[pypy-commit] pypy list-strategies: Merged default in, resolved merge conflicts (involved removing an optimization that had been done in a different way on default).

alex_gaynor noreply at buildbot.pypy.org
Wed Nov 2 14:52:11 CET 2011


Author: Alex Gaynor <alex.gaynor at gmail.com>
Branch: list-strategies
Changeset: r48658:56c40b33f07e
Date: 2011-11-02 09:51 -0400
http://bitbucket.org/pypy/pypy/changeset/56c40b33f07e/

Log:	Merged default in, resolved merge conflicts (involved removing an
	optimization that had been done in a different way on default).

diff too long, truncating to 10000 out of 19409 lines

diff --git a/lib-python/modified-2.7/heapq.py b/lib-python/modified-2.7/heapq.py
new file mode 100644
--- /dev/null
+++ b/lib-python/modified-2.7/heapq.py
@@ -0,0 +1,442 @@
+# -*- coding: latin-1 -*-
+
+"""Heap queue algorithm (a.k.a. priority queue).
+
+Heaps are arrays for which a[k] <= a[2*k+1] and a[k] <= a[2*k+2] for
+all k, counting elements from 0.  For the sake of comparison,
+non-existing elements are considered to be infinite.  The interesting
+property of a heap is that a[0] is always its smallest element.
+
+Usage:
+
+heap = []            # creates an empty heap
+heappush(heap, item) # pushes a new item on the heap
+item = heappop(heap) # pops the smallest item from the heap
+item = heap[0]       # smallest item on the heap without popping it
+heapify(x)           # transforms list into a heap, in-place, in linear time
+item = heapreplace(heap, item) # pops and returns smallest item, and adds
+                               # new item; the heap size is unchanged
+
+Our API differs from textbook heap algorithms as follows:
+
+- We use 0-based indexing.  This makes the relationship between the
+  index for a node and the indexes for its children slightly less
+  obvious, but is more suitable since Python uses 0-based indexing.
+
+- Our heappop() method returns the smallest item, not the largest.
+
+These two make it possible to view the heap as a regular Python list
+without surprises: heap[0] is the smallest item, and heap.sort()
+maintains the heap invariant!
+"""
+
+# Original code by Kevin O'Connor, augmented by Tim Peters and Raymond Hettinger
+
+__about__ = """Heap queues
+
+[explanation by Fran&#65533;ois Pinard]
+
+Heaps are arrays for which a[k] <= a[2*k+1] and a[k] <= a[2*k+2] for
+all k, counting elements from 0.  For the sake of comparison,
+non-existing elements are considered to be infinite.  The interesting
+property of a heap is that a[0] is always its smallest element.
+
+The strange invariant above is meant to be an efficient memory
+representation for a tournament.  The numbers below are `k', not a[k]:
+
+                                   0
+
+                  1                                 2
+
+          3               4                5               6
+
+      7       8       9       10      11      12      13      14
+
+    15 16   17 18   19 20   21 22   23 24   25 26   27 28   29 30
+
+
+In the tree above, each cell `k' is topping `2*k+1' and `2*k+2'.  In
+an usual binary tournament we see in sports, each cell is the winner
+over the two cells it tops, and we can trace the winner down the tree
+to see all opponents s/he had.  However, in many computer applications
+of such tournaments, we do not need to trace the history of a winner.
+To be more memory efficient, when a winner is promoted, we try to
+replace it by something else at a lower level, and the rule becomes
+that a cell and the two cells it tops contain three different items,
+but the top cell "wins" over the two topped cells.
+
+If this heap invariant is protected at all time, index 0 is clearly
+the overall winner.  The simplest algorithmic way to remove it and
+find the "next" winner is to move some loser (let's say cell 30 in the
+diagram above) into the 0 position, and then percolate this new 0 down
+the tree, exchanging values, until the invariant is re-established.
+This is clearly logarithmic on the total number of items in the tree.
+By iterating over all items, you get an O(n ln n) sort.
+
+A nice feature of this sort is that you can efficiently insert new
+items while the sort is going on, provided that the inserted items are
+not "better" than the last 0'th element you extracted.  This is
+especially useful in simulation contexts, where the tree holds all
+incoming events, and the "win" condition means the smallest scheduled
+time.  When an event schedule other events for execution, they are
+scheduled into the future, so they can easily go into the heap.  So, a
+heap is a good structure for implementing schedulers (this is what I
+used for my MIDI sequencer :-).
+
+Various structures for implementing schedulers have been extensively
+studied, and heaps are good for this, as they are reasonably speedy,
+the speed is almost constant, and the worst case is not much different
+than the average case.  However, there are other representations which
+are more efficient overall, yet the worst cases might be terrible.
+
+Heaps are also very useful in big disk sorts.  You most probably all
+know that a big sort implies producing "runs" (which are pre-sorted
+sequences, which size is usually related to the amount of CPU memory),
+followed by a merging passes for these runs, which merging is often
+very cleverly organised[1].  It is very important that the initial
+sort produces the longest runs possible.  Tournaments are a good way
+to that.  If, using all the memory available to hold a tournament, you
+replace and percolate items that happen to fit the current run, you'll
+produce runs which are twice the size of the memory for random input,
+and much better for input fuzzily ordered.
+
+Moreover, if you output the 0'th item on disk and get an input which
+may not fit in the current tournament (because the value "wins" over
+the last output value), it cannot fit in the heap, so the size of the
+heap decreases.  The freed memory could be cleverly reused immediately
+for progressively building a second heap, which grows at exactly the
+same rate the first heap is melting.  When the first heap completely
+vanishes, you switch heaps and start a new run.  Clever and quite
+effective!
+
+In a word, heaps are useful memory structures to know.  I use them in
+a few applications, and I think it is good to keep a `heap' module
+around. :-)
+
+--------------------
+[1] The disk balancing algorithms which are current, nowadays, are
+more annoying than clever, and this is a consequence of the seeking
+capabilities of the disks.  On devices which cannot seek, like big
+tape drives, the story was quite different, and one had to be very
+clever to ensure (far in advance) that each tape movement will be the
+most effective possible (that is, will best participate at
+"progressing" the merge).  Some tapes were even able to read
+backwards, and this was also used to avoid the rewinding time.
+Believe me, real good tape sorts were quite spectacular to watch!
+From all times, sorting has always been a Great Art! :-)
+"""
+
+__all__ = ['heappush', 'heappop', 'heapify', 'heapreplace', 'merge',
+           'nlargest', 'nsmallest', 'heappushpop']
+
+from itertools import islice, repeat, count, imap, izip, tee, chain
+from operator import itemgetter
+import bisect
+
+def heappush(heap, item):
+    """Push item onto heap, maintaining the heap invariant."""
+    heap.append(item)
+    _siftdown(heap, 0, len(heap)-1)
+
+def heappop(heap):
+    """Pop the smallest item off the heap, maintaining the heap invariant."""
+    lastelt = heap.pop()    # raises appropriate IndexError if heap is empty
+    if heap:
+        returnitem = heap[0]
+        heap[0] = lastelt
+        _siftup(heap, 0)
+    else:
+        returnitem = lastelt
+    return returnitem
+
+def heapreplace(heap, item):
+    """Pop and return the current smallest value, and add the new item.
+
+    This is more efficient than heappop() followed by heappush(), and can be
+    more appropriate when using a fixed-size heap.  Note that the value
+    returned may be larger than item!  That constrains reasonable uses of
+    this routine unless written as part of a conditional replacement:
+
+        if item > heap[0]:
+            item = heapreplace(heap, item)
+    """
+    returnitem = heap[0]    # raises appropriate IndexError if heap is empty
+    heap[0] = item
+    _siftup(heap, 0)
+    return returnitem
+
+def heappushpop(heap, item):
+    """Fast version of a heappush followed by a heappop."""
+    if heap and heap[0] < item:
+        item, heap[0] = heap[0], item
+        _siftup(heap, 0)
+    return item
+
+def heapify(x):
+    """Transform list into a heap, in-place, in O(len(heap)) time."""
+    n = len(x)
+    # Transform bottom-up.  The largest index there's any point to looking at
+    # is the largest with a child index in-range, so must have 2*i + 1 < n,
+    # or i < (n-1)/2.  If n is even = 2*j, this is (2*j-1)/2 = j-1/2 so
+    # j-1 is the largest, which is n//2 - 1.  If n is odd = 2*j+1, this is
+    # (2*j+1-1)/2 = j so j-1 is the largest, and that's again n//2-1.
+    for i in reversed(xrange(n//2)):
+        _siftup(x, i)
+
+def nlargest(n, iterable):
+    """Find the n largest elements in a dataset.
+
+    Equivalent to:  sorted(iterable, reverse=True)[:n]
+    """
+    if n < 0: # for consistency with the c impl
+        return []
+    it = iter(iterable)
+    result = list(islice(it, n))
+    if not result:
+        return result
+    heapify(result)
+    _heappushpop = heappushpop
+    for elem in it:
+        _heappushpop(result, elem)
+    result.sort(reverse=True)
+    return result
+
+def nsmallest(n, iterable):
+    """Find the n smallest elements in a dataset.
+
+    Equivalent to:  sorted(iterable)[:n]
+    """
+    if n < 0: # for consistency with the c impl
+        return []
+    if hasattr(iterable, '__len__') and n * 10 <= len(iterable):
+        # For smaller values of n, the bisect method is faster than a minheap.
+        # It is also memory efficient, consuming only n elements of space.
+        it = iter(iterable)
+        result = sorted(islice(it, 0, n))
+        if not result:
+            return result
+        insort = bisect.insort
+        pop = result.pop
+        los = result[-1]    # los --> Largest of the nsmallest
+        for elem in it:
+            if los <= elem:
+                continue
+            insort(result, elem)
+            pop()
+            los = result[-1]
+        return result
+    # An alternative approach manifests the whole iterable in memory but
+    # saves comparisons by heapifying all at once.  Also, saves time
+    # over bisect.insort() which has O(n) data movement time for every
+    # insertion.  Finding the n smallest of an m length iterable requires
+    #    O(m) + O(n log m) comparisons.
+    h = list(iterable)
+    heapify(h)
+    return map(heappop, repeat(h, min(n, len(h))))
+
+# 'heap' is a heap at all indices >= startpos, except possibly for pos.  pos
+# is the index of a leaf with a possibly out-of-order value.  Restore the
+# heap invariant.
+def _siftdown(heap, startpos, pos):
+    newitem = heap[pos]
+    # Follow the path to the root, moving parents down until finding a place
+    # newitem fits.
+    while pos > startpos:
+        parentpos = (pos - 1) >> 1
+        parent = heap[parentpos]
+        if newitem < parent:
+            heap[pos] = parent
+            pos = parentpos
+            continue
+        break
+    heap[pos] = newitem
+
+# The child indices of heap index pos are already heaps, and we want to make
+# a heap at index pos too.  We do this by bubbling the smaller child of
+# pos up (and so on with that child's children, etc) until hitting a leaf,
+# then using _siftdown to move the oddball originally at index pos into place.
+#
+# We *could* break out of the loop as soon as we find a pos where newitem <=
+# both its children, but turns out that's not a good idea, and despite that
+# many books write the algorithm that way.  During a heap pop, the last array
+# element is sifted in, and that tends to be large, so that comparing it
+# against values starting from the root usually doesn't pay (= usually doesn't
+# get us out of the loop early).  See Knuth, Volume 3, where this is
+# explained and quantified in an exercise.
+#
+# Cutting the # of comparisons is important, since these routines have no
+# way to extract "the priority" from an array element, so that intelligence
+# is likely to be hiding in custom __cmp__ methods, or in array elements
+# storing (priority, record) tuples.  Comparisons are thus potentially
+# expensive.
+#
+# On random arrays of length 1000, making this change cut the number of
+# comparisons made by heapify() a little, and those made by exhaustive
+# heappop() a lot, in accord with theory.  Here are typical results from 3
+# runs (3 just to demonstrate how small the variance is):
+#
+# Compares needed by heapify     Compares needed by 1000 heappops
+# --------------------------     --------------------------------
+# 1837 cut to 1663               14996 cut to 8680
+# 1855 cut to 1659               14966 cut to 8678
+# 1847 cut to 1660               15024 cut to 8703
+#
+# Building the heap by using heappush() 1000 times instead required
+# 2198, 2148, and 2219 compares:  heapify() is more efficient, when
+# you can use it.
+#
+# The total compares needed by list.sort() on the same lists were 8627,
+# 8627, and 8632 (this should be compared to the sum of heapify() and
+# heappop() compares):  list.sort() is (unsurprisingly!) more efficient
+# for sorting.
+
+def _siftup(heap, pos):
+    endpos = len(heap)
+    startpos = pos
+    newitem = heap[pos]
+    # Bubble up the smaller child until hitting a leaf.
+    childpos = 2*pos + 1    # leftmost child position
+    while childpos < endpos:
+        # Set childpos to index of smaller child.
+        rightpos = childpos + 1
+        if rightpos < endpos and not heap[childpos] < heap[rightpos]:
+            childpos = rightpos
+        # Move the smaller child up.
+        heap[pos] = heap[childpos]
+        pos = childpos
+        childpos = 2*pos + 1
+    # The leaf at pos is empty now.  Put newitem there, and bubble it up
+    # to its final resting place (by sifting its parents down).
+    heap[pos] = newitem
+    _siftdown(heap, startpos, pos)
+
+# If available, use C implementation
+try:
+    from _heapq import *
+except ImportError:
+    pass
+
+def merge(*iterables):
+    '''Merge multiple sorted inputs into a single sorted output.
+
+    Similar to sorted(itertools.chain(*iterables)) but returns a generator,
+    does not pull the data into memory all at once, and assumes that each of
+    the input streams is already sorted (smallest to largest).
+
+    >>> list(merge([1,3,5,7], [0,2,4,8], [5,10,15,20], [], [25]))
+    [0, 1, 2, 3, 4, 5, 5, 7, 8, 10, 15, 20, 25]
+
+    '''
+    _heappop, _heapreplace, _StopIteration = heappop, heapreplace, StopIteration
+
+    h = []
+    h_append = h.append
+    for itnum, it in enumerate(map(iter, iterables)):
+        try:
+            next = it.next
+            h_append([next(), itnum, next])
+        except _StopIteration:
+            pass
+    heapify(h)
+
+    while 1:
+        try:
+            while 1:
+                v, itnum, next = s = h[0]   # raises IndexError when h is empty
+                yield v
+                s[0] = next()               # raises StopIteration when exhausted
+                _heapreplace(h, s)          # restore heap condition
+        except _StopIteration:
+            _heappop(h)                     # remove empty iterator
+        except IndexError:
+            return
+
+# Extend the implementations of nsmallest and nlargest to use a key= argument
+_nsmallest = nsmallest
+def nsmallest(n, iterable, key=None):
+    """Find the n smallest elements in a dataset.
+
+    Equivalent to:  sorted(iterable, key=key)[:n]
+    """
+    # Short-cut for n==1 is to use min() when len(iterable)>0
+    if n == 1:
+        it = iter(iterable)
+        head = list(islice(it, 1))
+        if not head:
+            return []
+        if key is None:
+            return [min(chain(head, it))]
+        return [min(chain(head, it), key=key)]
+
+    # When n>=size, it's faster to use sort()
+    try:
+        size = len(iterable)
+    except (TypeError, AttributeError):
+        pass
+    else:
+        if n >= size:
+            return sorted(iterable, key=key)[:n]
+
+    # When key is none, use simpler decoration
+    if key is None:
+        it = izip(iterable, count())                        # decorate
+        result = _nsmallest(n, it)
+        return map(itemgetter(0), result)                   # undecorate
+
+    # General case, slowest method
+    in1, in2 = tee(iterable)
+    it = izip(imap(key, in1), count(), in2)                 # decorate
+    result = _nsmallest(n, it)
+    return map(itemgetter(2), result)                       # undecorate
+
+_nlargest = nlargest
+def nlargest(n, iterable, key=None):
+    """Find the n largest elements in a dataset.
+
+    Equivalent to:  sorted(iterable, key=key, reverse=True)[:n]
+    """
+
+    # Short-cut for n==1 is to use max() when len(iterable)>0
+    if n == 1:
+        it = iter(iterable)
+        head = list(islice(it, 1))
+        if not head:
+            return []
+        if key is None:
+            return [max(chain(head, it))]
+        return [max(chain(head, it), key=key)]
+
+    # When n>=size, it's faster to use sort()
+    try:
+        size = len(iterable)
+    except (TypeError, AttributeError):
+        pass
+    else:
+        if n >= size:
+            return sorted(iterable, key=key, reverse=True)[:n]
+
+    # When key is none, use simpler decoration
+    if key is None:
+        it = izip(iterable, count(0,-1))                    # decorate
+        result = _nlargest(n, it)
+        return map(itemgetter(0), result)                   # undecorate
+
+    # General case, slowest method
+    in1, in2 = tee(iterable)
+    it = izip(imap(key, in1), count(0,-1), in2)             # decorate
+    result = _nlargest(n, it)
+    return map(itemgetter(2), result)                       # undecorate
+
+if __name__ == "__main__":
+    # Simple sanity test
+    heap = []
+    data = [1, 3, 5, 7, 9, 2, 4, 6, 8, 0]
+    for item in data:
+        heappush(heap, item)
+    sort = []
+    while heap:
+        sort.append(heappop(heap))
+    print sort
+
+    import doctest
+    doctest.testmod()
diff --git a/lib-python/modified-2.7/httplib.py b/lib-python/modified-2.7/httplib.py
new file mode 100644
--- /dev/null
+++ b/lib-python/modified-2.7/httplib.py
@@ -0,0 +1,1377 @@
+"""HTTP/1.1 client library
+
+<intro stuff goes here>
+<other stuff, too>
+
+HTTPConnection goes through a number of "states", which define when a client
+may legally make another request or fetch the response for a particular
+request. This diagram details these state transitions:
+
+    (null)
+      |
+      | HTTPConnection()
+      v
+    Idle
+      |
+      | putrequest()
+      v
+    Request-started
+      |
+      | ( putheader() )*  endheaders()
+      v
+    Request-sent
+      |
+      | response = getresponse()
+      v
+    Unread-response   [Response-headers-read]
+      |\____________________
+      |                     |
+      | response.read()     | putrequest()
+      v                     v
+    Idle                  Req-started-unread-response
+                     ______/|
+                   /        |
+   response.read() |        | ( putheader() )*  endheaders()
+                   v        v
+       Request-started    Req-sent-unread-response
+                            |
+                            | response.read()
+                            v
+                          Request-sent
+
+This diagram presents the following rules:
+  -- a second request may not be started until {response-headers-read}
+  -- a response [object] cannot be retrieved until {request-sent}
+  -- there is no differentiation between an unread response body and a
+     partially read response body
+
+Note: this enforcement is applied by the HTTPConnection class. The
+      HTTPResponse class does not enforce this state machine, which
+      implies sophisticated clients may accelerate the request/response
+      pipeline. Caution should be taken, though: accelerating the states
+      beyond the above pattern may imply knowledge of the server's
+      connection-close behavior for certain requests. For example, it
+      is impossible to tell whether the server will close the connection
+      UNTIL the response headers have been read; this means that further
+      requests cannot be placed into the pipeline until it is known that
+      the server will NOT be closing the connection.
+
+Logical State                  __state            __response
+-------------                  -------            ----------
+Idle                           _CS_IDLE           None
+Request-started                _CS_REQ_STARTED    None
+Request-sent                   _CS_REQ_SENT       None
+Unread-response                _CS_IDLE           <response_class>
+Req-started-unread-response    _CS_REQ_STARTED    <response_class>
+Req-sent-unread-response       _CS_REQ_SENT       <response_class>
+"""
+
+from array import array
+import os
+import socket
+from sys import py3kwarning
+from urlparse import urlsplit
+import warnings
+with warnings.catch_warnings():
+    if py3kwarning:
+        warnings.filterwarnings("ignore", ".*mimetools has been removed",
+                                DeprecationWarning)
+    import mimetools
+
+try:
+    from cStringIO import StringIO
+except ImportError:
+    from StringIO import StringIO
+
+__all__ = ["HTTP", "HTTPResponse", "HTTPConnection",
+           "HTTPException", "NotConnected", "UnknownProtocol",
+           "UnknownTransferEncoding", "UnimplementedFileMode",
+           "IncompleteRead", "InvalidURL", "ImproperConnectionState",
+           "CannotSendRequest", "CannotSendHeader", "ResponseNotReady",
+           "BadStatusLine", "error", "responses"]
+
+HTTP_PORT = 80
+HTTPS_PORT = 443
+
+_UNKNOWN = 'UNKNOWN'
+
+# connection states
+_CS_IDLE = 'Idle'
+_CS_REQ_STARTED = 'Request-started'
+_CS_REQ_SENT = 'Request-sent'
+
+# status codes
+# informational
+CONTINUE = 100
+SWITCHING_PROTOCOLS = 101
+PROCESSING = 102
+
+# successful
+OK = 200
+CREATED = 201
+ACCEPTED = 202
+NON_AUTHORITATIVE_INFORMATION = 203
+NO_CONTENT = 204
+RESET_CONTENT = 205
+PARTIAL_CONTENT = 206
+MULTI_STATUS = 207
+IM_USED = 226
+
+# redirection
+MULTIPLE_CHOICES = 300
+MOVED_PERMANENTLY = 301
+FOUND = 302
+SEE_OTHER = 303
+NOT_MODIFIED = 304
+USE_PROXY = 305
+TEMPORARY_REDIRECT = 307
+
+# client error
+BAD_REQUEST = 400
+UNAUTHORIZED = 401
+PAYMENT_REQUIRED = 402
+FORBIDDEN = 403
+NOT_FOUND = 404
+METHOD_NOT_ALLOWED = 405
+NOT_ACCEPTABLE = 406
+PROXY_AUTHENTICATION_REQUIRED = 407
+REQUEST_TIMEOUT = 408
+CONFLICT = 409
+GONE = 410
+LENGTH_REQUIRED = 411
+PRECONDITION_FAILED = 412
+REQUEST_ENTITY_TOO_LARGE = 413
+REQUEST_URI_TOO_LONG = 414
+UNSUPPORTED_MEDIA_TYPE = 415
+REQUESTED_RANGE_NOT_SATISFIABLE = 416
+EXPECTATION_FAILED = 417
+UNPROCESSABLE_ENTITY = 422
+LOCKED = 423
+FAILED_DEPENDENCY = 424
+UPGRADE_REQUIRED = 426
+
+# server error
+INTERNAL_SERVER_ERROR = 500
+NOT_IMPLEMENTED = 501
+BAD_GATEWAY = 502
+SERVICE_UNAVAILABLE = 503
+GATEWAY_TIMEOUT = 504
+HTTP_VERSION_NOT_SUPPORTED = 505
+INSUFFICIENT_STORAGE = 507
+NOT_EXTENDED = 510
+
+# Mapping status codes to official W3C names
+responses = {
+    100: 'Continue',
+    101: 'Switching Protocols',
+
+    200: 'OK',
+    201: 'Created',
+    202: 'Accepted',
+    203: 'Non-Authoritative Information',
+    204: 'No Content',
+    205: 'Reset Content',
+    206: 'Partial Content',
+
+    300: 'Multiple Choices',
+    301: 'Moved Permanently',
+    302: 'Found',
+    303: 'See Other',
+    304: 'Not Modified',
+    305: 'Use Proxy',
+    306: '(Unused)',
+    307: 'Temporary Redirect',
+
+    400: 'Bad Request',
+    401: 'Unauthorized',
+    402: 'Payment Required',
+    403: 'Forbidden',
+    404: 'Not Found',
+    405: 'Method Not Allowed',
+    406: 'Not Acceptable',
+    407: 'Proxy Authentication Required',
+    408: 'Request Timeout',
+    409: 'Conflict',
+    410: 'Gone',
+    411: 'Length Required',
+    412: 'Precondition Failed',
+    413: 'Request Entity Too Large',
+    414: 'Request-URI Too Long',
+    415: 'Unsupported Media Type',
+    416: 'Requested Range Not Satisfiable',
+    417: 'Expectation Failed',
+
+    500: 'Internal Server Error',
+    501: 'Not Implemented',
+    502: 'Bad Gateway',
+    503: 'Service Unavailable',
+    504: 'Gateway Timeout',
+    505: 'HTTP Version Not Supported',
+}
+
+# maximal amount of data to read at one time in _safe_read
+MAXAMOUNT = 1048576
+
+class HTTPMessage(mimetools.Message):
+
+    def addheader(self, key, value):
+        """Add header for field key handling repeats."""
+        prev = self.dict.get(key)
+        if prev is None:
+            self.dict[key] = value
+        else:
+            combined = ", ".join((prev, value))
+            self.dict[key] = combined
+
+    def addcontinue(self, key, more):
+        """Add more field data from a continuation line."""
+        prev = self.dict[key]
+        self.dict[key] = prev + "\n " + more
+
+    def readheaders(self):
+        """Read header lines.
+
+        Read header lines up to the entirely blank line that terminates them.
+        The (normally blank) line that ends the headers is skipped, but not
+        included in the returned list.  If a non-header line ends the headers,
+        (which is an error), an attempt is made to backspace over it; it is
+        never included in the returned list.
+
+        The variable self.status is set to the empty string if all went well,
+        otherwise it is an error message.  The variable self.headers is a
+        completely uninterpreted list of lines contained in the header (so
+        printing them will reproduce the header exactly as it appears in the
+        file).
+
+        If multiple header fields with the same name occur, they are combined
+        according to the rules in RFC 2616 sec 4.2:
+
+        Appending each subsequent field-value to the first, each separated
+        by a comma. The order in which header fields with the same field-name
+        are received is significant to the interpretation of the combined
+        field value.
+        """
+        # XXX The implementation overrides the readheaders() method of
+        # rfc822.Message.  The base class design isn't amenable to
+        # customized behavior here so the method here is a copy of the
+        # base class code with a few small changes.
+
+        self.dict = {}
+        self.unixfrom = ''
+        self.headers = hlist = []
+        self.status = ''
+        headerseen = ""
+        firstline = 1
+        startofline = unread = tell = None
+        if hasattr(self.fp, 'unread'):
+            unread = self.fp.unread
+        elif self.seekable:
+            tell = self.fp.tell
+        while True:
+            if tell:
+                try:
+                    startofline = tell()
+                except IOError:
+                    startofline = tell = None
+                    self.seekable = 0
+            line = self.fp.readline()
+            if not line:
+                self.status = 'EOF in headers'
+                break
+            # Skip unix From name time lines
+            if firstline and line.startswith('From '):
+                self.unixfrom = self.unixfrom + line
+                continue
+            firstline = 0
+            if headerseen and line[0] in ' \t':
+                # XXX Not sure if continuation lines are handled properly
+                # for http and/or for repeating headers
+                # It's a continuation line.
+                hlist.append(line)
+                self.addcontinue(headerseen, line.strip())
+                continue
+            elif self.iscomment(line):
+                # It's a comment.  Ignore it.
+                continue
+            elif self.islast(line):
+                # Note! No pushback here!  The delimiter line gets eaten.
+                break
+            headerseen = self.isheader(line)
+            if headerseen:
+                # It's a legal header line, save it.
+                hlist.append(line)
+                self.addheader(headerseen, line[len(headerseen)+1:].strip())
+                continue
+            else:
+                # It's not a header line; throw it back and stop here.
+                if not self.dict:
+                    self.status = 'No headers'
+                else:
+                    self.status = 'Non-header line where header expected'
+                # Try to undo the read.
+                if unread:
+                    unread(line)
+                elif tell:
+                    self.fp.seek(startofline)
+                else:
+                    self.status = self.status + '; bad seek'
+                break
+
+class HTTPResponse:
+
+    # strict: If true, raise BadStatusLine if the status line can't be
+    # parsed as a valid HTTP/1.0 or 1.1 status line.  By default it is
+    # false because it prevents clients from talking to HTTP/0.9
+    # servers.  Note that a response with a sufficiently corrupted
+    # status line will look like an HTTP/0.9 response.
+
+    # See RFC 2616 sec 19.6 and RFC 1945 sec 6 for details.
+
+    def __init__(self, sock, debuglevel=0, strict=0, method=None, buffering=False):
+        if buffering:
+            # The caller won't be using any sock.recv() calls, so buffering
+            # is fine and recommended for performance.
+            self.fp = sock.makefile('rb')
+        else:
+            # The buffer size is specified as zero, because the headers of
+            # the response are read with readline().  If the reads were
+            # buffered the readline() calls could consume some of the
+            # response, which make be read via a recv() on the underlying
+            # socket.
+            self.fp = sock.makefile('rb', 0)
+        self.debuglevel = debuglevel
+        self.strict = strict
+        self._method = method
+
+        self.msg = None
+
+        # from the Status-Line of the response
+        self.version = _UNKNOWN # HTTP-Version
+        self.status = _UNKNOWN  # Status-Code
+        self.reason = _UNKNOWN  # Reason-Phrase
+
+        self.chunked = _UNKNOWN         # is "chunked" being used?
+        self.chunk_left = _UNKNOWN      # bytes left to read in current chunk
+        self.length = _UNKNOWN          # number of bytes left in response
+        self.will_close = _UNKNOWN      # conn will close at end of response
+
+    def _read_status(self):
+        # Initialize with Simple-Response defaults
+        line = self.fp.readline()
+        if self.debuglevel > 0:
+            print "reply:", repr(line)
+        if not line:
+            # Presumably, the server closed the connection before
+            # sending a valid response.
+            raise BadStatusLine(line)
+        try:
+            [version, status, reason] = line.split(None, 2)
+        except ValueError:
+            try:
+                [version, status] = line.split(None, 1)
+                reason = ""
+            except ValueError:
+                # empty version will cause next test to fail and status
+                # will be treated as 0.9 response.
+                version = ""
+        if not version.startswith('HTTP/'):
+            if self.strict:
+                self.close()
+                raise BadStatusLine(line)
+            else:
+                # assume it's a Simple-Response from an 0.9 server
+                self.fp = LineAndFileWrapper(line, self.fp)
+                return "HTTP/0.9", 200, ""
+
+        # The status code is a three-digit number
+        try:
+            status = int(status)
+            if status < 100 or status > 999:
+                raise BadStatusLine(line)
+        except ValueError:
+            raise BadStatusLine(line)
+        return version, status, reason
+
+    def begin(self):
+        if self.msg is not None:
+            # we've already started reading the response
+            return
+
+        # read until we get a non-100 response
+        while True:
+            version, status, reason = self._read_status()
+            if status != CONTINUE:
+                break
+            # skip the header from the 100 response
+            while True:
+                skip = self.fp.readline().strip()
+                if not skip:
+                    break
+                if self.debuglevel > 0:
+                    print "header:", skip
+
+        self.status = status
+        self.reason = reason.strip()
+        if version == 'HTTP/1.0':
+            self.version = 10
+        elif version.startswith('HTTP/1.'):
+            self.version = 11   # use HTTP/1.1 code for HTTP/1.x where x>=1
+        elif version == 'HTTP/0.9':
+            self.version = 9
+        else:
+            raise UnknownProtocol(version)
+
+        if self.version == 9:
+            self.length = None
+            self.chunked = 0
+            self.will_close = 1
+            self.msg = HTTPMessage(StringIO())
+            return
+
+        self.msg = HTTPMessage(self.fp, 0)
+        if self.debuglevel > 0:
+            for hdr in self.msg.headers:
+                print "header:", hdr,
+
+        # don't let the msg keep an fp
+        self.msg.fp = None
+
+        # are we using the chunked-style of transfer encoding?
+        tr_enc = self.msg.getheader('transfer-encoding')
+        if tr_enc and tr_enc.lower() == "chunked":
+            self.chunked = 1
+            self.chunk_left = None
+        else:
+            self.chunked = 0
+
+        # will the connection close at the end of the response?
+        self.will_close = self._check_close()
+
+        # do we have a Content-Length?
+        # NOTE: RFC 2616, S4.4, #3 says we ignore this if tr_enc is "chunked"
+        length = self.msg.getheader('content-length')
+        if length and not self.chunked:
+            try:
+                self.length = int(length)
+            except ValueError:
+                self.length = None
+            else:
+                if self.length < 0:  # ignore nonsensical negative lengths
+                    self.length = None
+        else:
+            self.length = None
+
+        # does the body have a fixed length? (of zero)
+        if (status == NO_CONTENT or status == NOT_MODIFIED or
+            100 <= status < 200 or      # 1xx codes
+            self._method == 'HEAD'):
+            self.length = 0
+
+        # if the connection remains open, and we aren't using chunked, and
+        # a content-length was not provided, then assume that the connection
+        # WILL close.
+        if not self.will_close and \
+           not self.chunked and \
+           self.length is None:
+            self.will_close = 1
+
+    def _check_close(self):
+        conn = self.msg.getheader('connection')
+        if self.version == 11:
+            # An HTTP/1.1 proxy is assumed to stay open unless
+            # explicitly closed.
+            conn = self.msg.getheader('connection')
+            if conn and "close" in conn.lower():
+                return True
+            return False
+
+        # Some HTTP/1.0 implementations have support for persistent
+        # connections, using rules different than HTTP/1.1.
+
+        # For older HTTP, Keep-Alive indicates persistent connection.
+        if self.msg.getheader('keep-alive'):
+            return False
+
+        # At least Akamai returns a "Connection: Keep-Alive" header,
+        # which was supposed to be sent by the client.
+        if conn and "keep-alive" in conn.lower():
+            return False
+
+        # Proxy-Connection is a netscape hack.
+        pconn = self.msg.getheader('proxy-connection')
+        if pconn and "keep-alive" in pconn.lower():
+            return False
+
+        # otherwise, assume it will close
+        return True
+
+    def close(self):
+        if self.fp:
+            self.fp.close()
+            self.fp = None
+
+    def isclosed(self):
+        # NOTE: it is possible that we will not ever call self.close(). This
+        #       case occurs when will_close is TRUE, length is None, and we
+        #       read up to the last byte, but NOT past it.
+        #
+        # IMPLIES: if will_close is FALSE, then self.close() will ALWAYS be
+        #          called, meaning self.isclosed() is meaningful.
+        return self.fp is None
+
+    # XXX It would be nice to have readline and __iter__ for this, too.
+
+    def read(self, amt=None):
+        if self.fp is None:
+            return ''
+
+        if self._method == 'HEAD':
+            self.close()
+            return ''
+
+        if self.chunked:
+            return self._read_chunked(amt)
+
+        if amt is None:
+            # unbounded read
+            if self.length is None:
+                s = self.fp.read()
+            else:
+                s = self._safe_read(self.length)
+                self.length = 0
+            self.close()        # we read everything
+            return s
+
+        if self.length is not None:
+            if amt > self.length:
+                # clip the read to the "end of response"
+                amt = self.length
+
+        # we do not use _safe_read() here because this may be a .will_close
+        # connection, and the user is reading more bytes than will be provided
+        # (for example, reading in 1k chunks)
+        s = self.fp.read(amt)
+        if self.length is not None:
+            self.length -= len(s)
+            if not self.length:
+                self.close()
+        return s
+
+    def _read_chunked(self, amt):
+        assert self.chunked != _UNKNOWN
+        chunk_left = self.chunk_left
+        value = []
+        while True:
+            if chunk_left is None:
+                line = self.fp.readline()
+                i = line.find(';')
+                if i >= 0:
+                    line = line[:i] # strip chunk-extensions
+                try:
+                    chunk_left = int(line, 16)
+                except ValueError:
+                    # close the connection as protocol synchronisation is
+                    # probably lost
+                    self.close()
+                    raise IncompleteRead(''.join(value))
+                if chunk_left == 0:
+                    break
+            if amt is None:
+                value.append(self._safe_read(chunk_left))
+            elif amt < chunk_left:
+                value.append(self._safe_read(amt))
+                self.chunk_left = chunk_left - amt
+                return ''.join(value)
+            elif amt == chunk_left:
+                value.append(self._safe_read(amt))
+                self._safe_read(2)  # toss the CRLF at the end of the chunk
+                self.chunk_left = None
+                return ''.join(value)
+            else:
+                value.append(self._safe_read(chunk_left))
+                amt -= chunk_left
+
+            # we read the whole chunk, get another
+            self._safe_read(2)      # toss the CRLF at the end of the chunk
+            chunk_left = None
+
+        # read and discard trailer up to the CRLF terminator
+        ### note: we shouldn't have any trailers!
+        while True:
+            line = self.fp.readline()
+            if not line:
+                # a vanishingly small number of sites EOF without
+                # sending the trailer
+                break
+            if line == '\r\n':
+                break
+
+        # we read everything; close the "file"
+        self.close()
+
+        return ''.join(value)
+
+    def _safe_read(self, amt):
+        """Read the number of bytes requested, compensating for partial reads.
+
+        Normally, we have a blocking socket, but a read() can be interrupted
+        by a signal (resulting in a partial read).
+
+        Note that we cannot distinguish between EOF and an interrupt when zero
+        bytes have been read. IncompleteRead() will be raised in this
+        situation.
+
+        This function should be used when <amt> bytes "should" be present for
+        reading. If the bytes are truly not available (due to EOF), then the
+        IncompleteRead exception can be used to detect the problem.
+        """
+        # NOTE(gps): As of svn r74426 socket._fileobject.read(x) will never
+        # return less than x bytes unless EOF is encountered.  It now handles
+        # signal interruptions (socket.error EINTR) internally.  This code
+        # never caught that exception anyways.  It seems largely pointless.
+        # self.fp.read(amt) will work fine.
+        s = []
+        while amt > 0:
+            chunk = self.fp.read(min(amt, MAXAMOUNT))
+            if not chunk:
+                raise IncompleteRead(''.join(s), amt)
+            s.append(chunk)
+            amt -= len(chunk)
+        return ''.join(s)
+
+    def fileno(self):
+        return self.fp.fileno()
+
+    def getheader(self, name, default=None):
+        if self.msg is None:
+            raise ResponseNotReady()
+        return self.msg.getheader(name, default)
+
+    def getheaders(self):
+        """Return list of (header, value) tuples."""
+        if self.msg is None:
+            raise ResponseNotReady()
+        return self.msg.items()
+
+
+class HTTPConnection:
+
+    _http_vsn = 11
+    _http_vsn_str = 'HTTP/1.1'
+
+    response_class = HTTPResponse
+    default_port = HTTP_PORT
+    auto_open = 1
+    debuglevel = 0
+    strict = 0
+
+    def __init__(self, host, port=None, strict=None,
+                 timeout=socket._GLOBAL_DEFAULT_TIMEOUT, source_address=None):
+        self.timeout = timeout
+        self.source_address = source_address
+        self.sock = None
+        self._buffer = []
+        self.__response = None
+        self.__state = _CS_IDLE
+        self._method = None
+        self._tunnel_host = None
+        self._tunnel_port = None
+        self._tunnel_headers = {}
+
+        self._set_hostport(host, port)
+        if strict is not None:
+            self.strict = strict
+
+    def set_tunnel(self, host, port=None, headers=None):
+        """ Sets up the host and the port for the HTTP CONNECT Tunnelling.
+
+        The headers argument should be a mapping of extra HTTP headers
+        to send with the CONNECT request.
+        """
+        self._tunnel_host = host
+        self._tunnel_port = port
+        if headers:
+            self._tunnel_headers = headers
+        else:
+            self._tunnel_headers.clear()
+
+    def _set_hostport(self, host, port):
+        if port is None:
+            i = host.rfind(':')
+            j = host.rfind(']')         # ipv6 addresses have [...]
+            if i > j:
+                try:
+                    port = int(host[i+1:])
+                except ValueError:
+                    raise InvalidURL("nonnumeric port: '%s'" % host[i+1:])
+                host = host[:i]
+            else:
+                port = self.default_port
+            if host and host[0] == '[' and host[-1] == ']':
+                host = host[1:-1]
+        self.host = host
+        self.port = port
+
+    def set_debuglevel(self, level):
+        self.debuglevel = level
+
+    def _tunnel(self):
+        self._set_hostport(self._tunnel_host, self._tunnel_port)
+        self.send("CONNECT %s:%d HTTP/1.0\r\n" % (self.host, self.port))
+        for header, value in self._tunnel_headers.iteritems():
+            self.send("%s: %s\r\n" % (header, value))
+        self.send("\r\n")
+        response = self.response_class(self.sock, strict = self.strict,
+                                       method = self._method)
+        (version, code, message) = response._read_status()
+
+        if code != 200:
+            self.close()
+            raise socket.error("Tunnel connection failed: %d %s" % (code,
+                                                                    message.strip()))
+        while True:
+            line = response.fp.readline()
+            if line == '\r\n': break
+
+
+    def connect(self):
+        """Connect to the host and port specified in __init__."""
+        self.sock = socket.create_connection((self.host,self.port),
+                                             self.timeout, self.source_address)
+
+        if self._tunnel_host:
+            self._tunnel()
+
+    def close(self):
+        """Close the connection to the HTTP server."""
+        if self.sock:
+            self.sock.close()   # close it manually... there may be other refs
+            self.sock = None
+        if self.__response:
+            self.__response.close()
+            self.__response = None
+        self.__state = _CS_IDLE
+
+    def send(self, data):
+        """Send `data' to the server."""
+        if self.sock is None:
+            if self.auto_open:
+                self.connect()
+            else:
+                raise NotConnected()
+
+        if self.debuglevel > 0:
+            print "send:", repr(data)
+        blocksize = 8192
+        if hasattr(data,'read') and not isinstance(data, array):
+            if self.debuglevel > 0: print "sendIng a read()able"
+            datablock = data.read(blocksize)
+            while datablock:
+                self.sock.sendall(datablock)
+                datablock = data.read(blocksize)
+        else:
+            self.sock.sendall(data)
+
+    def _output(self, s):
+        """Add a line of output to the current request buffer.
+
+        Assumes that the line does *not* end with \\r\\n.
+        """
+        self._buffer.append(s)
+
+    def _send_output(self, message_body=None):
+        """Send the currently buffered request and clear the buffer.
+
+        Appends an extra \\r\\n to the buffer.
+        A message_body may be specified, to be appended to the request.
+        """
+        self._buffer.extend(("", ""))
+        msg = "\r\n".join(self._buffer)
+        del self._buffer[:]
+        # If msg and message_body are sent in a single send() call,
+        # it will avoid performance problems caused by the interaction
+        # between delayed ack and the Nagle algorithim.
+        if isinstance(message_body, str):
+            msg += message_body
+            message_body = None
+        self.send(msg)
+        if message_body is not None:
+            #message_body was not a string (i.e. it is a file) and
+            #we must run the risk of Nagle
+            self.send(message_body)
+
+    def putrequest(self, method, url, skip_host=0, skip_accept_encoding=0):
+        """Send a request to the server.
+
+        `method' specifies an HTTP request method, e.g. 'GET'.
+        `url' specifies the object being requested, e.g. '/index.html'.
+        `skip_host' if True does not add automatically a 'Host:' header
+        `skip_accept_encoding' if True does not add automatically an
+           'Accept-Encoding:' header
+        """
+
+        # if a prior response has been completed, then forget about it.
+        if self.__response and self.__response.isclosed():
+            self.__response = None
+
+
+        # in certain cases, we cannot issue another request on this connection.
+        # this occurs when:
+        #   1) we are in the process of sending a request.   (_CS_REQ_STARTED)
+        #   2) a response to a previous request has signalled that it is going
+        #      to close the connection upon completion.
+        #   3) the headers for the previous response have not been read, thus
+        #      we cannot determine whether point (2) is true.   (_CS_REQ_SENT)
+        #
+        # if there is no prior response, then we can request at will.
+        #
+        # if point (2) is true, then we will have passed the socket to the
+        # response (effectively meaning, "there is no prior response"), and
+        # will open a new one when a new request is made.
+        #
+        # Note: if a prior response exists, then we *can* start a new request.
+        #       We are not allowed to begin fetching the response to this new
+        #       request, however, until that prior response is complete.
+        #
+        if self.__state == _CS_IDLE:
+            self.__state = _CS_REQ_STARTED
+        else:
+            raise CannotSendRequest()
+
+        # Save the method we use, we need it later in the response phase
+        self._method = method
+        if not url:
+            url = '/'
+        hdr = '%s %s %s' % (method, url, self._http_vsn_str)
+
+        self._output(hdr)
+
+        if self._http_vsn == 11:
+            # Issue some standard headers for better HTTP/1.1 compliance
+
+            if not skip_host:
+                # this header is issued *only* for HTTP/1.1
+                # connections. more specifically, this means it is
+                # only issued when the client uses the new
+                # HTTPConnection() class. backwards-compat clients
+                # will be using HTTP/1.0 and those clients may be
+                # issuing this header themselves. we should NOT issue
+                # it twice; some web servers (such as Apache) barf
+                # when they see two Host: headers
+
+                # If we need a non-standard port,include it in the
+                # header.  If the request is going through a proxy,
+                # but the host of the actual URL, not the host of the
+                # proxy.
+
+                netloc = ''
+                if url.startswith('http'):
+                    nil, netloc, nil, nil, nil = urlsplit(url)
+
+                if netloc:
+                    try:
+                        netloc_enc = netloc.encode("ascii")
+                    except UnicodeEncodeError:
+                        netloc_enc = netloc.encode("idna")
+                    self.putheader('Host', netloc_enc)
+                else:
+                    try:
+                        host_enc = self.host.encode("ascii")
+                    except UnicodeEncodeError:
+                        host_enc = self.host.encode("idna")
+                    # Wrap the IPv6 Host Header with [] (RFC 2732)
+                    if host_enc.find(':') >= 0:
+                        host_enc = "[" + host_enc + "]"
+                    if self.port == self.default_port:
+                        self.putheader('Host', host_enc)
+                    else:
+                        self.putheader('Host', "%s:%s" % (host_enc, self.port))
+
+            # note: we are assuming that clients will not attempt to set these
+            #       headers since *this* library must deal with the
+            #       consequences. this also means that when the supporting
+            #       libraries are updated to recognize other forms, then this
+            #       code should be changed (removed or updated).
+
+            # we only want a Content-Encoding of "identity" since we don't
+            # support encodings such as x-gzip or x-deflate.
+            if not skip_accept_encoding:
+                self.putheader('Accept-Encoding', 'identity')
+
+            # we can accept "chunked" Transfer-Encodings, but no others
+            # NOTE: no TE header implies *only* "chunked"
+            #self.putheader('TE', 'chunked')
+
+            # if TE is supplied in the header, then it must appear in a
+            # Connection header.
+            #self.putheader('Connection', 'TE')
+
+        else:
+            # For HTTP/1.0, the server will assume "not chunked"
+            pass
+
+    def putheader(self, header, *values):
+        """Send a request header line to the server.
+
+        For example: h.putheader('Accept', 'text/html')
+        """
+        if self.__state != _CS_REQ_STARTED:
+            raise CannotSendHeader()
+
+        hdr = '%s: %s' % (header, '\r\n\t'.join([str(v) for v in values]))
+        self._output(hdr)
+
+    def endheaders(self, message_body=None):
+        """Indicate that the last header line has been sent to the server.
+
+        This method sends the request to the server.  The optional
+        message_body argument can be used to pass message body
+        associated with the request.  The message body will be sent in
+        the same packet as the message headers if possible.  The
+        message_body should be a string.
+        """
+        if self.__state == _CS_REQ_STARTED:
+            self.__state = _CS_REQ_SENT
+        else:
+            raise CannotSendHeader()
+        self._send_output(message_body)
+
+    def request(self, method, url, body=None, headers={}):
+        """Send a complete request to the server."""
+        self._send_request(method, url, body, headers)
+
+    def _set_content_length(self, body):
+        # Set the content-length based on the body.
+        thelen = None
+        try:
+            thelen = str(len(body))
+        except TypeError, te:
+            # If this is a file-like object, try to
+            # fstat its file descriptor
+            try:
+                thelen = str(os.fstat(body.fileno()).st_size)
+            except (AttributeError, OSError):
+                # Don't send a length if this failed
+                if self.debuglevel > 0: print "Cannot stat!!"
+
+        if thelen is not None:
+            self.putheader('Content-Length', thelen)
+
+    def _send_request(self, method, url, body, headers):
+        # Honor explicitly requested Host: and Accept-Encoding: headers.
+        header_names = dict.fromkeys([k.lower() for k in headers])
+        skips = {}
+        if 'host' in header_names:
+            skips['skip_host'] = 1
+        if 'accept-encoding' in header_names:
+            skips['skip_accept_encoding'] = 1
+
+        self.putrequest(method, url, **skips)
+
+        if body and ('content-length' not in header_names):
+            self._set_content_length(body)
+        for hdr, value in headers.iteritems():
+            self.putheader(hdr, value)
+        self.endheaders(body)
+
+    def getresponse(self, buffering=False):
+        "Get the response from the server."
+
+        # if a prior response has been completed, then forget about it.
+        if self.__response and self.__response.isclosed():
+            self.__response = None
+
+        #
+        # if a prior response exists, then it must be completed (otherwise, we
+        # cannot read this response's header to determine the connection-close
+        # behavior)
+        #
+        # note: if a prior response existed, but was connection-close, then the
+        # socket and response were made independent of this HTTPConnection
+        # object since a new request requires that we open a whole new
+        # connection
+        #
+        # this means the prior response had one of two states:
+        #   1) will_close: this connection was reset and the prior socket and
+        #                  response operate independently
+        #   2) persistent: the response was retained and we await its
+        #                  isclosed() status to become true.
+        #
+        if self.__state != _CS_REQ_SENT or self.__response:
+            raise ResponseNotReady()
+
+        args = (self.sock,)
+        kwds = {"strict":self.strict, "method":self._method}
+        if self.debuglevel > 0:
+            args += (self.debuglevel,)
+        if buffering:
+            #only add this keyword if non-default, for compatibility with
+            #other response_classes.
+            kwds["buffering"] = True;
+        response = self.response_class(*args, **kwds)
+
+        try:
+            response.begin()
+        except:
+            response.close()
+            raise
+        assert response.will_close != _UNKNOWN
+        self.__state = _CS_IDLE
+
+        if response.will_close:
+            # this effectively passes the connection to the response
+            self.close()
+        else:
+            # remember this, so we can tell when it is complete
+            self.__response = response
+
+        return response
+
+
+class HTTP:
+    "Compatibility class with httplib.py from 1.5."
+
+    _http_vsn = 10
+    _http_vsn_str = 'HTTP/1.0'
+
+    debuglevel = 0
+
+    _connection_class = HTTPConnection
+
+    def __init__(self, host='', port=None, strict=None):
+        "Provide a default host, since the superclass requires one."
+
+        # some joker passed 0 explicitly, meaning default port
+        if port == 0:
+            port = None
+
+        # Note that we may pass an empty string as the host; this will throw
+        # an error when we attempt to connect. Presumably, the client code
+        # will call connect before then, with a proper host.
+        self._setup(self._connection_class(host, port, strict))
+
+    def _setup(self, conn):
+        self._conn = conn
+
+        # set up delegation to flesh out interface
+        self.send = conn.send
+        self.putrequest = conn.putrequest
+        self.putheader = conn.putheader
+        self.endheaders = conn.endheaders
+        self.set_debuglevel = conn.set_debuglevel
+
+        conn._http_vsn = self._http_vsn
+        conn._http_vsn_str = self._http_vsn_str
+
+        self.file = None
+
+    def connect(self, host=None, port=None):
+        "Accept arguments to set the host/port, since the superclass doesn't."
+
+        if host is not None:
+            self._conn._set_hostport(host, port)
+        self._conn.connect()
+
+    def getfile(self):
+        "Provide a getfile, since the superclass' does not use this concept."
+        return self.file
+
+    def getreply(self, buffering=False):
+        """Compat definition since superclass does not define it.
+
+        Returns a tuple consisting of:
+        - server status code (e.g. '200' if all goes well)
+        - server "reason" corresponding to status code
+        - any RFC822 headers in the response from the server
+        """
+        try:
+            if not buffering:
+                response = self._conn.getresponse()
+            else:
+                #only add this keyword if non-default for compatibility
+                #with other connection classes
+                response = self._conn.getresponse(buffering)
+        except BadStatusLine, e:
+            ### hmm. if getresponse() ever closes the socket on a bad request,
+            ### then we are going to have problems with self.sock
+
+            ### should we keep this behavior? do people use it?
+            # keep the socket open (as a file), and return it
+            self.file = self._conn.sock.makefile('rb', 0)
+
+            # close our socket -- we want to restart after any protocol error
+            self.close()
+
+            self.headers = None
+            return -1, e.line, None
+
+        self.headers = response.msg
+        self.file = response.fp
+        return response.status, response.reason, response.msg
+
+    def close(self):
+        self._conn.close()
+
+        # note that self.file == response.fp, which gets closed by the
+        # superclass. just clear the object ref here.
+        ### hmm. messy. if status==-1, then self.file is owned by us.
+        ### well... we aren't explicitly closing, but losing this ref will
+        ### do it
+        self.file = None
+
+try:
+    import ssl
+except ImportError:
+    pass
+else:
+    class HTTPSConnection(HTTPConnection):
+        "This class allows communication via SSL."
+
+        default_port = HTTPS_PORT
+
+        def __init__(self, host, port=None, key_file=None, cert_file=None,
+                     strict=None, timeout=socket._GLOBAL_DEFAULT_TIMEOUT,
+                     source_address=None):
+            HTTPConnection.__init__(self, host, port, strict, timeout,
+                                    source_address)
+            self.key_file = key_file
+            self.cert_file = cert_file
+
+        def connect(self):
+            "Connect to a host on a given (SSL) port."
+
+            sock = socket.create_connection((self.host, self.port),
+                                            self.timeout, self.source_address)
+            if self._tunnel_host:
+                self.sock = sock
+                self._tunnel()
+            self.sock = ssl.wrap_socket(sock, self.key_file, self.cert_file)
+
+    __all__.append("HTTPSConnection")
+
+    class HTTPS(HTTP):
+        """Compatibility with 1.5 httplib interface
+
+        Python 1.5.2 did not have an HTTPS class, but it defined an
+        interface for sending http requests that is also useful for
+        https.
+        """
+
+        _connection_class = HTTPSConnection
+
+        def __init__(self, host='', port=None, key_file=None, cert_file=None,
+                     strict=None):
+            # provide a default host, pass the X509 cert info
+
+            # urf. compensate for bad input.
+            if port == 0:
+                port = None
+            self._setup(self._connection_class(host, port, key_file,
+                                               cert_file, strict))
+
+            # we never actually use these for anything, but we keep them
+            # here for compatibility with post-1.5.2 CVS.
+            self.key_file = key_file
+            self.cert_file = cert_file
+
+
+    def FakeSocket (sock, sslobj):
+        warnings.warn("FakeSocket is deprecated, and won't be in 3.x.  " +
+                      "Use the result of ssl.wrap_socket() directly instead.",
+                      DeprecationWarning, stacklevel=2)
+        return sslobj
+
+
+class HTTPException(Exception):
+    # Subclasses that define an __init__ must call Exception.__init__
+    # or define self.args.  Otherwise, str() will fail.
+    pass
+
+class NotConnected(HTTPException):
+    pass
+
+class InvalidURL(HTTPException):
+    pass
+
+class UnknownProtocol(HTTPException):
+    def __init__(self, version):
+        self.args = version,
+        self.version = version
+
+class UnknownTransferEncoding(HTTPException):
+    pass
+
+class UnimplementedFileMode(HTTPException):
+    pass
+
+class IncompleteRead(HTTPException):
+    def __init__(self, partial, expected=None):
+        self.args = partial,
+        self.partial = partial
+        self.expected = expected
+    def __repr__(self):
+        if self.expected is not None:
+            e = ', %i more expected' % self.expected
+        else:
+            e = ''
+        return 'IncompleteRead(%i bytes read%s)' % (len(self.partial), e)
+    def __str__(self):
+        return repr(self)
+
+class ImproperConnectionState(HTTPException):
+    pass
+
+class CannotSendRequest(ImproperConnectionState):
+    pass
+
+class CannotSendHeader(ImproperConnectionState):
+    pass
+
+class ResponseNotReady(ImproperConnectionState):
+    pass
+
+class BadStatusLine(HTTPException):
+    def __init__(self, line):
+        if not line:
+            line = repr(line)
+        self.args = line,
+        self.line = line
+
+# for backwards compatibility
+error = HTTPException
+
+class LineAndFileWrapper:
+    """A limited file-like object for HTTP/0.9 responses."""
+
+    # The status-line parsing code calls readline(), which normally
+    # get the HTTP status line.  For a 0.9 response, however, this is
+    # actually the first line of the body!  Clients need to get a
+    # readable file object that contains that line.
+
+    def __init__(self, line, file):
+        self._line = line
+        self._file = file
+        self._line_consumed = 0
+        self._line_offset = 0
+        self._line_left = len(line)
+
+    def __getattr__(self, attr):
+        return getattr(self._file, attr)
+
+    def _done(self):
+        # called when the last byte is read from the line.  After the
+        # call, all read methods are delegated to the underlying file
+        # object.
+        self._line_consumed = 1
+        self.read = self._file.read
+        self.readline = self._file.readline
+        self.readlines = self._file.readlines
+
+    def read(self, amt=None):
+        if self._line_consumed:
+            return self._file.read(amt)
+        assert self._line_left
+        if amt is None or amt > self._line_left:
+            s = self._line[self._line_offset:]
+            self._done()
+            if amt is None:
+                return s + self._file.read()
+            else:
+                return s + self._file.read(amt - len(s))
+        else:
+            assert amt <= self._line_left
+            i = self._line_offset
+            j = i + amt
+            s = self._line[i:j]
+            self._line_offset = j
+            self._line_left -= amt
+            if self._line_left == 0:
+                self._done()
+            return s
+
+    def readline(self):
+        if self._line_consumed:
+            return self._file.readline()
+        assert self._line_left
+        s = self._line[self._line_offset:]
+        self._done()
+        return s
+
+    def readlines(self, size=None):
+        if self._line_consumed:
+            return self._file.readlines(size)
+        assert self._line_left
+        L = [self._line[self._line_offset:]]
+        self._done()
+        if size is None:
+            return L + self._file.readlines()
+        else:
+            return L + self._file.readlines(size)
+
+def test():
+    """Test this module.
+
+    A hodge podge of tests collected here, because they have too many
+    external dependencies for the regular test suite.
+    """
+
+    import sys
+    import getopt
+    opts, args = getopt.getopt(sys.argv[1:], 'd')
+    dl = 0
+    for o, a in opts:
+        if o == '-d': dl = dl + 1
+    host = 'www.python.org'
+    selector = '/'
+    if args[0:]: host = args[0]
+    if args[1:]: selector = args[1]
+    h = HTTP()
+    h.set_debuglevel(dl)
+    h.connect(host)
+    h.putrequest('GET', selector)
+    h.endheaders()
+    status, reason, headers = h.getreply()
+    print 'status =', status
+    print 'reason =', reason
+    print "read", len(h.getfile().read())
+    print
+    if headers:
+        for header in headers.headers: print header.strip()
+    print
+
+    # minimal test that code to extract host from url works
+    class HTTP11(HTTP):
+        _http_vsn = 11
+        _http_vsn_str = 'HTTP/1.1'
+
+    h = HTTP11('www.python.org')
+    h.putrequest('GET', 'http://www.python.org/~jeremy/')
+    h.endheaders()
+    h.getreply()
+    h.close()
+
+    try:
+        import ssl
+    except ImportError:
+        pass
+    else:
+
+        for host, selector in (('sourceforge.net', '/projects/python'),
+                               ):
+            print "https://%s%s" % (host, selector)
+            hs = HTTPS()
+            hs.set_debuglevel(dl)
+            hs.connect(host)
+            hs.putrequest('GET', selector)
+            hs.endheaders()
+            status, reason, headers = hs.getreply()
+            print 'status =', status
+            print 'reason =', reason
+            print "read", len(hs.getfile().read())
+            print
+            if headers:
+                for header in headers.headers: print header.strip()
+            print
+
+if __name__ == '__main__':
+    test()
diff --git a/lib-python/modified-2.7/json/encoder.py b/lib-python/modified-2.7/json/encoder.py
--- a/lib-python/modified-2.7/json/encoder.py
+++ b/lib-python/modified-2.7/json/encoder.py
@@ -2,14 +2,7 @@
 """
 import re
 
-try:
-    from _json import encode_basestring_ascii as c_encode_basestring_ascii
-except ImportError:
-    c_encode_basestring_ascii = None
-try:
-    from _json import make_encoder as c_make_encoder
-except ImportError:
-    c_make_encoder = None
+from __pypy__.builders import StringBuilder, UnicodeBuilder
 
 ESCAPE = re.compile(r'[\x00-\x1f\\"\b\f\n\r\t]')
 ESCAPE_ASCII = re.compile(r'([\\"]|[^\ -~])')
@@ -24,8 +17,7 @@
     '\t': '\\t',
 }
 for i in range(0x20):
-    ESCAPE_DCT.setdefault(chr(i), '\\u{0:04x}'.format(i))
-    #ESCAPE_DCT.setdefault(chr(i), '\\u%04x' % (i,))
+    ESCAPE_DCT.setdefault(chr(i), '\\u%04x' % (i,))
 
 # Assume this produces an infinity on all machines (probably not guaranteed)
 INFINITY = float('1e66666')
@@ -37,10 +29,9 @@
     """
     def replace(match):
         return ESCAPE_DCT[match.group(0)]
-    return '"' + ESCAPE.sub(replace, s) + '"'
+    return ESCAPE.sub(replace, s)
 
-
-def py_encode_basestring_ascii(s):
+def encode_basestring_ascii(s):
     """Return an ASCII-only JSON representation of a Python string
 
     """
@@ -53,20 +44,18 @@
         except KeyError:
             n = ord(s)
             if n < 0x10000:
-                return '\\u{0:04x}'.format(n)
-                #return '\\u%04x' % (n,)
+                return '\\u%04x' % (n,)
             else:
                 # surrogate pair
                 n -= 0x10000
                 s1 = 0xd800 | ((n >> 10) & 0x3ff)
                 s2 = 0xdc00 | (n & 0x3ff)
-                return '\\u{0:04x}\\u{1:04x}'.format(s1, s2)
-                #return '\\u%04x\\u%04x' % (s1, s2)
-    return '"' + str(ESCAPE_ASCII.sub(replace, s)) + '"'
-
-
-encode_basestring_ascii = (
-    c_encode_basestring_ascii or py_encode_basestring_ascii)
+                return '\\u%04x\\u%04x' % (s1, s2)
+    if ESCAPE_ASCII.search(s):
+        return str(ESCAPE_ASCII.sub(replace, s))
+    return s
+py_encode_basestring_ascii = lambda s: '"' + encode_basestring_ascii(s) + '"'
+c_encode_basestring_ascii = None
 
 class JSONEncoder(object):
     """Extensible JSON <http://json.org> encoder for Python data structures.
@@ -147,6 +136,17 @@
 
         self.skipkeys = skipkeys
         self.ensure_ascii = ensure_ascii
+        if ensure_ascii:
+            self.encoder = encode_basestring_ascii
+        else:
+            self.encoder = encode_basestring
+        if encoding != 'utf-8':
+            orig_encoder = self.encoder
+            def encoder(o):
+                if isinstance(o, str):
+                    o = o.decode(encoding)
+                return orig_encoder(o)
+            self.encoder = encoder
         self.check_circular = check_circular
         self.allow_nan = allow_nan
         self.sort_keys = sort_keys
@@ -184,24 +184,126 @@
         '{"foo": ["bar", "baz"]}'
 
         """
-        # This is for extremely simple cases and benchmarks.
+        if self.check_circular:
+            markers = {}
+        else:
+            markers = None
+        if self.ensure_ascii:
+            builder = StringBuilder()
+        else:
+            builder = UnicodeBuilder()
+        self._encode(o, markers, builder, 0)
+        return builder.build()
+
+    def _emit_indent(self, builder, _current_indent_level):
+        if self.indent is not None:
+            _current_indent_level += 1
+            newline_indent = '\n' + (' ' * (self.indent *
+                                            _current_indent_level))
+            separator = self.item_separator + newline_indent
+            builder.append(newline_indent)
+        else:
+            separator = self.item_separator
+        return separator, _current_indent_level
+
+    def _emit_unindent(self, builder, _current_indent_level):
+        if self.indent is not None:
+            builder.append('\n')
+            builder.append(' ' * (self.indent * (_current_indent_level - 1)))
+
+    def _encode(self, o, markers, builder, _current_indent_level):
         if isinstance(o, basestring):
-            if isinstance(o, str):
-                _encoding = self.encoding
-                if (_encoding is not None
-                        and not (_encoding == 'utf-8')):
-                    o = o.decode(_encoding)
-            if self.ensure_ascii:
-                return encode_basestring_ascii(o)
+            builder.append('"')
+            builder.append(self.encoder(o))
+            builder.append('"')
+        elif o is None:
+            builder.append('null')
+        elif o is True:
+            builder.append('true')
+        elif o is False:
+            builder.append('false')
+        elif isinstance(o, (int, long)):
+            builder.append(str(o))
+        elif isinstance(o, float):
+            builder.append(self._floatstr(o))
+        elif isinstance(o, (list, tuple)):
+            if not o:
+                builder.append('[]')
+                return
+            self._encode_list(o, markers, builder, _current_indent_level)
+        elif isinstance(o, dict):
+            if not o:
+                builder.append('{}')
+                return
+            self._encode_dict(o, markers, builder, _current_indent_level)
+        else:
+            self._mark_markers(markers, o)
+            res = self.default(o)
+            self._encode(res, markers, builder, _current_indent_level)
+            self._remove_markers(markers, o)
+            return res
+
+    def _encode_list(self, l, markers, builder, _current_indent_level):
+        self._mark_markers(markers, l)
+        builder.append('[')
+        first = True
+        separator, _current_indent_level = self._emit_indent(builder,
+                                                      _current_indent_level)
+        for elem in l:
+            if first:
+                first = False
             else:
-                return encode_basestring(o)
-        # This doesn't pass the iterator directly to ''.join() because the
-        # exceptions aren't as detailed.  The list call should be roughly
-        # equivalent to the PySequence_Fast that ''.join() would do.
-        chunks = self.iterencode(o, _one_shot=True)
-        if not isinstance(chunks, (list, tuple)):
-            chunks = list(chunks)
-        return ''.join(chunks)
+                builder.append(separator)
+            self._encode(elem, markers, builder, _current_indent_level)
+            del elem # XXX grumble
+        self._emit_unindent(builder, _current_indent_level)
+        builder.append(']')
+        self._remove_markers(markers, l)
+
+    def _encode_dict(self, d, markers, builder, _current_indent_level):
+        self._mark_markers(markers, d)
+        first = True
+        builder.append('{')
+        separator, _current_indent_level = self._emit_indent(builder,
+                                                         _current_indent_level)
+        if self.sort_keys:
+            items = sorted(d.items(), key=lambda kv: kv[0])
+        else:
+            items = d.iteritems()
+
+        for key, v in items:
+            if first:
+                first = False
+            else:
+                builder.append(separator)
+            if isinstance(key, basestring):
+                pass
+            # JavaScript is weakly typed for these, so it makes sense to
+            # also allow them.  Many encoders seem to do something like this.
+            elif isinstance(key, float):
+                key = self._floatstr(key)
+            elif key is True:
+                key = 'true'
+            elif key is False:
+                key = 'false'
+            elif key is None:
+                key = 'null'
+            elif isinstance(key, (int, long)):
+                key = str(key)
+            elif self.skipkeys:
+                continue
+            else:
+                raise TypeError("key " + repr(key) + " is not a string")
+            builder.append('"')
+            builder.append(self.encoder(key))
+            builder.append('"')
+            builder.append(self.key_separator)
+            self._encode(v, markers, builder, _current_indent_level)
+            del key
+            del v # XXX grumble
+        self._emit_unindent(builder, _current_indent_level)
+        builder.append('}')
+        self._remove_markers(markers, d)
 
     def iterencode(self, o, _one_shot=False):
         """Encode the given object and yield each string
@@ -217,86 +319,54 @@
             markers = {}
         else:
             markers = None
-        if self.ensure_ascii:
-            _encoder = encode_basestring_ascii
+        return self._iterencode(o, markers, 0)
+
+    def _floatstr(self, o):
+        # Check for specials.  Note that this type of test is processor
+        # and/or platform-specific, so do tests which don't depend on the
+        # internals.
+
+        if o != o:
+            text = 'NaN'
+        elif o == INFINITY:
+            text = 'Infinity'
+        elif o == -INFINITY:
+            text = '-Infinity'
         else:
-            _encoder = encode_basestring
-        if self.encoding != 'utf-8':
-            def _encoder(o, _orig_encoder=_encoder, _encoding=self.encoding):
-                if isinstance(o, str):
-                    o = o.decode(_encoding)
-                return _orig_encoder(o)
+            return FLOAT_REPR(o)
 
-        def floatstr(o, allow_nan=self.allow_nan,
-                _repr=FLOAT_REPR, _inf=INFINITY, _neginf=-INFINITY):
-            # Check for specials.  Note that this type of test is processor
-            # and/or platform-specific, so do tests which don't depend on the
-            # internals.
+        if not self.allow_nan:
+            raise ValueError(
+                "Out of range float values are not JSON compliant: " +
+                repr(o))
 
-            if o != o:
-                text = 'NaN'
-            elif o == _inf:
-                text = 'Infinity'
-            elif o == _neginf:
-                text = '-Infinity'
-            else:
-                return _repr(o)
+        return text
 
-            if not allow_nan:
-                raise ValueError(
-                    "Out of range float values are not JSON compliant: " +
-                    repr(o))
+    def _mark_markers(self, markers, o):
+        if markers is not None:
+            if id(o) in markers:
+                raise ValueError("Circular reference detected")
+            markers[id(o)] = None
 
-            return text
+    def _remove_markers(self, markers, o):
+        if markers is not None:
+            del markers[id(o)]
 
-
-        if (_one_shot and c_make_encoder is not None
-                and not self.indent and not self.sort_keys):
-            _iterencode = c_make_encoder(
-                markers, self.default, _encoder, self.indent,
-                self.key_separator, self.item_separator, self.sort_keys,
-                self.skipkeys, self.allow_nan)
-        else:
-            _iterencode = _make_iterencode(
-                markers, self.default, _encoder, self.indent, floatstr,
-                self.key_separator, self.item_separator, self.sort_keys,
-                self.skipkeys, _one_shot)
-        return _iterencode(o, 0)
-
-def _make_iterencode(markers, _default, _encoder, _indent, _floatstr,
-        _key_separator, _item_separator, _sort_keys, _skipkeys, _one_shot,
-        ## HACK: hand-optimized bytecode; turn globals into locals
-        ValueError=ValueError,
-        basestring=basestring,
-        dict=dict,
-        float=float,
-        id=id,
-        int=int,
-        isinstance=isinstance,
-        list=list,
-        long=long,
-        str=str,
-        tuple=tuple,
-    ):
-
-    def _iterencode_list(lst, _current_indent_level):
+    def _iterencode_list(self, lst, markers, _current_indent_level):
         if not lst:
             yield '[]'
             return
-        if markers is not None:
-            markerid = id(lst)
-            if markerid in markers:
-                raise ValueError("Circular reference detected")
-            markers[markerid] = lst
+        self._mark_markers(markers, lst)
         buf = '['
-        if _indent is not None:
+        if self.indent is not None:
             _current_indent_level += 1
-            newline_indent = '\n' + (' ' * (_indent * _current_indent_level))
-            separator = _item_separator + newline_indent
+            newline_indent = '\n' + (' ' * (self.indent *
+                                            _current_indent_level))
+            separator = self.item_separator + newline_indent
             buf += newline_indent
         else:
             newline_indent = None
-            separator = _item_separator
+            separator = self.item_separator
         first = True
         for value in lst:
             if first:
@@ -304,7 +374,7 @@
             else:
                 buf = separator
             if isinstance(value, basestring):
-                yield buf + _encoder(value)
+                yield buf + '"' + self.encoder(value) + '"'
             elif value is None:
                 yield buf + 'null'
             elif value is True:
@@ -314,44 +384,43 @@
             elif isinstance(value, (int, long)):
                 yield buf + str(value)
             elif isinstance(value, float):
-                yield buf + _floatstr(value)
+                yield buf + self._floatstr(value)
             else:
                 yield buf
                 if isinstance(value, (list, tuple)):
-                    chunks = _iterencode_list(value, _current_indent_level)
+                    chunks = self._iterencode_list(value, markers,
+                                                   _current_indent_level)
                 elif isinstance(value, dict):
-                    chunks = _iterencode_dict(value, _current_indent_level)
+                    chunks = self._iterencode_dict(value, markers,
+                                                   _current_indent_level)
                 else:
-                    chunks = _iterencode(value, _current_indent_level)
+                    chunks = self._iterencode(value, markers,
+                                              _current_indent_level)
                 for chunk in chunks:
                     yield chunk
         if newline_indent is not None:
             _current_indent_level -= 1
-            yield '\n' + (' ' * (_indent * _current_indent_level))
+            yield '\n' + (' ' * (self.indent * _current_indent_level))
         yield ']'
-        if markers is not None:
-            del markers[markerid]
+        self._remove_markers(markers, lst)
 
-    def _iterencode_dict(dct, _current_indent_level):
+    def _iterencode_dict(self, dct, markers, _current_indent_level):
         if not dct:
             yield '{}'
             return
-        if markers is not None:
-            markerid = id(dct)
-            if markerid in markers:
-                raise ValueError("Circular reference detected")
-            markers[markerid] = dct
+        self._mark_markers(markers, dct)
         yield '{'
-        if _indent is not None:
+        if self.indent is not None:
             _current_indent_level += 1
-            newline_indent = '\n' + (' ' * (_indent * _current_indent_level))
-            item_separator = _item_separator + newline_indent
+            newline_indent = '\n' + (' ' * (self.indent *
+                                            _current_indent_level))
+            item_separator = self.item_separator + newline_indent
             yield newline_indent
         else:
             newline_indent = None
-            item_separator = _item_separator
+            item_separator = self.item_separator
         first = True
-        if _sort_keys:
+        if self.sort_keys:
             items = sorted(dct.items(), key=lambda kv: kv[0])
         else:
             items = dct.iteritems()
@@ -361,7 +430,7 @@
             # JavaScript is weakly typed for these, so it makes sense to
             # also allow them.  Many encoders seem to do something like this.
             elif isinstance(key, float):
-                key = _floatstr(key)
+                key = self._floatstr(key)
             elif key is True:
                 key = 'true'
             elif key is False:
@@ -370,7 +439,7 @@
                 key = 'null'
             elif isinstance(key, (int, long)):
                 key = str(key)
-            elif _skipkeys:
+            elif self.skipkeys:
                 continue
             else:
                 raise TypeError("key " + repr(key) + " is not a string")
@@ -378,10 +447,10 @@
                 first = False
             else:
                 yield item_separator
-            yield _encoder(key)
-            yield _key_separator
+            yield '"' + self.encoder(key) + '"'
+            yield self.key_separator
             if isinstance(value, basestring):
-                yield _encoder(value)
+                yield '"' + self.encoder(value) + '"'
             elif value is None:
                 yield 'null'
             elif value is True:
@@ -391,26 +460,28 @@
             elif isinstance(value, (int, long)):
                 yield str(value)
             elif isinstance(value, float):
-                yield _floatstr(value)
+                yield self._floatstr(value)
             else:
                 if isinstance(value, (list, tuple)):
-                    chunks = _iterencode_list(value, _current_indent_level)
+                    chunks = self._iterencode_list(value, markers,
+                                                   _current_indent_level)
                 elif isinstance(value, dict):
-                    chunks = _iterencode_dict(value, _current_indent_level)
+                    chunks = self._iterencode_dict(value, markers,
+                                                   _current_indent_level)
                 else:
-                    chunks = _iterencode(value, _current_indent_level)
+                    chunks = self._iterencode(value, markers,
+                                              _current_indent_level)
                 for chunk in chunks:
                     yield chunk
         if newline_indent is not None:
             _current_indent_level -= 1
-            yield '\n' + (' ' * (_indent * _current_indent_level))
+            yield '\n' + (' ' * (self.indent * _current_indent_level))
         yield '}'
-        if markers is not None:
-            del markers[markerid]
+        self._remove_markers(markers, dct)
 
-    def _iterencode(o, _current_indent_level):
+    def _iterencode(self, o, markers, _current_indent_level):
         if isinstance(o, basestring):
-            yield _encoder(o)
+            yield '"' + self.encoder(o) + '"'
         elif o is None:
             yield 'null'
         elif o is True:
@@ -420,23 +491,19 @@
         elif isinstance(o, (int, long)):
             yield str(o)
         elif isinstance(o, float):
-            yield _floatstr(o)
+            yield self._floatstr(o)
         elif isinstance(o, (list, tuple)):
-            for chunk in _iterencode_list(o, _current_indent_level):
+            for chunk in self._iterencode_list(o, markers,
+                                               _current_indent_level):
                 yield chunk
         elif isinstance(o, dict):
-            for chunk in _iterencode_dict(o, _current_indent_level):
+            for chunk in self._iterencode_dict(o, markers,
+                                               _current_indent_level):
                 yield chunk
         else:
-            if markers is not None:
-                markerid = id(o)
-                if markerid in markers:
-                    raise ValueError("Circular reference detected")
-                markers[markerid] = o
-            o = _default(o)
-            for chunk in _iterencode(o, _current_indent_level):
+            self._mark_markers(markers, o)
+            obj = self.default(o)
+            for chunk in self._iterencode(obj, markers,
+                                          _current_indent_level):
                 yield chunk
-            if markers is not None:
-                del markers[markerid]
-
-    return _iterencode
+            self._remove_markers(markers, o)
diff --git a/lib-python/modified-2.7/json/tests/test_unicode.py b/lib-python/modified-2.7/json/tests/test_unicode.py
--- a/lib-python/modified-2.7/json/tests/test_unicode.py
+++ b/lib-python/modified-2.7/json/tests/test_unicode.py
@@ -80,3 +80,9 @@
         self.assertEqual(type(json.loads(u'["a"]')[0]), unicode)
         # Issue 10038.
         self.assertEqual(type(json.loads('"foo"')), unicode)
+
+    def test_encode_not_utf_8(self):
+        self.assertEqual(json.dumps('\xb1\xe6', encoding='iso8859-2'),
+                         '"\\u0105\\u0107"')
+        self.assertEqual(json.dumps(['\xb1\xe6'], encoding='iso8859-2'),
+                         '["\\u0105\\u0107"]')
diff --git a/lib-python/modified-2.7/test/test_array.py b/lib-python/modified-2.7/test/test_array.py
--- a/lib-python/modified-2.7/test/test_array.py
+++ b/lib-python/modified-2.7/test/test_array.py
@@ -295,9 +295,10 @@
         )
 
         b = array.array(self.badtypecode())
-        self.assertRaises(TypeError, "a + b")
-
-        self.assertRaises(TypeError, "a + 'bad'")
+        with self.assertRaises(TypeError):
+            a + b
+        with self.assertRaises(TypeError):
+            a + 'bad'
 
     def test_iadd(self):
         a = array.array(self.typecode, self.example[::-1])
@@ -316,9 +317,10 @@
         )
 
         b = array.array(self.badtypecode())
-        self.assertRaises(TypeError, "a += b")
-
-        self.assertRaises(TypeError, "a += 'bad'")
+        with self.assertRaises(TypeError):
+            a += b
+        with self.assertRaises(TypeError):
+            a += 'bad'
 
     def test_mul(self):
         a = 5*array.array(self.typecode, self.example)
@@ -345,7 +347,8 @@
             array.array(self.typecode)
         )
 
-        self.assertRaises(TypeError, "a * 'bad'")
+        with self.assertRaises(TypeError):
+            a * 'bad'
 
     def test_imul(self):
         a = array.array(self.typecode, self.example)
@@ -374,7 +377,8 @@
         a *= -1
         self.assertEqual(a, array.array(self.typecode))
 
-        self.assertRaises(TypeError, "a *= 'bad'")
+        with self.assertRaises(TypeError):
+            a *= 'bad'
 
     def test_getitem(self):
         a = array.array(self.typecode, self.example)
diff --git a/lib-python/modified-2.7/test/test_heapq.py b/lib-python/modified-2.7/test/test_heapq.py
--- a/lib-python/modified-2.7/test/test_heapq.py
+++ b/lib-python/modified-2.7/test/test_heapq.py
@@ -186,6 +186,11 @@
         self.assertFalse(sys.modules['heapq'] is self.module)
         self.assertTrue(hasattr(self.module.heapify, 'func_code'))
 
+    def test_islice_protection(self):
+        m = self.module
+        self.assertFalse(m.nsmallest(-1, [1]))
+        self.assertFalse(m.nlargest(-1, [1]))
+
 
 class TestHeapC(TestHeap):
     module = c_heapq
diff --git a/lib-python/modified-2.7/test/test_sys_settrace.py b/lib-python/modified-2.7/test/test_sys_settrace.py
--- a/lib-python/modified-2.7/test/test_sys_settrace.py
+++ b/lib-python/modified-2.7/test/test_sys_settrace.py
@@ -286,11 +286,11 @@
         self.compare_events(func.func_code.co_firstlineno,
                             tracer.events, func.events)
 
-    def set_and_retrieve_none(self):
+    def test_set_and_retrieve_none(self):
         sys.settrace(None)
         assert sys.gettrace() is None
 
-    def set_and_retrieve_func(self):
+    def test_set_and_retrieve_func(self):
         def fn(*args):
             pass
 
diff --git a/lib-python/modified-2.7/test/test_urllib2.py b/lib-python/modified-2.7/test/test_urllib2.py
--- a/lib-python/modified-2.7/test/test_urllib2.py
+++ b/lib-python/modified-2.7/test/test_urllib2.py
@@ -307,6 +307,9 @@
     def getresponse(self):
         return MockHTTPResponse(MockFile(), {}, 200, "OK")
 
+    def close(self):
+        pass
+
 class MockHandler:
     # useful for testing handler machinery
     # see add_ordered_mock_handlers() docstring
diff --git a/lib-python/modified-2.7/urllib2.py b/lib-python/modified-2.7/urllib2.py
new file mode 100644
--- /dev/null
+++ b/lib-python/modified-2.7/urllib2.py
@@ -0,0 +1,1436 @@
+"""An extensible library for opening URLs using a variety of protocols
+
+The simplest way to use this module is to call the urlopen function,
+which accepts a string containing a URL or a Request object (described
+below).  It opens the URL and returns the results as file-like
+object; the returned object has some extra methods described below.
+
+The OpenerDirector manages a collection of Handler objects that do
+all the actual work.  Each Handler implements a particular protocol or
+option.  The OpenerDirector is a composite object that invokes the
+Handlers needed to open the requested URL.  For example, the
+HTTPHandler performs HTTP GET and POST requests and deals with
+non-error returns.  The HTTPRedirectHandler automatically deals with
+HTTP 301, 302, 303 and 307 redirect errors, and the HTTPDigestAuthHandler
+deals with digest authentication.
+
+urlopen(url, data=None) -- Basic usage is the same as original
+urllib.  pass the url and optionally data to post to an HTTP URL, and
+get a file-like object back.  One difference is that you can also pass
+a Request instance instead of URL.  Raises a URLError (subclass of
+IOError); for HTTP errors, raises an HTTPError, which can also be
+treated as a valid response.
+
+build_opener -- Function that creates a new OpenerDirector instance.
+Will install the default handlers.  Accepts one or more Handlers as
+arguments, either instances or Handler classes that it will
+instantiate.  If one of the argument is a subclass of the default
+handler, the argument will be installed instead of the default.
+
+install_opener -- Installs a new opener as the default opener.
+
+objects of interest:
+
+OpenerDirector -- Sets up the User Agent as the Python-urllib client and manages
+the Handler classes, while dealing with requests and responses.
+
+Request -- An object that encapsulates the state of a request.  The
+state can be as simple as the URL.  It can also include extra HTTP
+headers, e.g. a User-Agent.
+
+BaseHandler --
+
+exceptions:
+URLError -- A subclass of IOError, individual protocols have their own
+specific subclass.
+
+HTTPError -- Also a valid HTTP response, so you can treat an HTTP error
+as an exceptional event or valid response.
+
+internals:
+BaseHandler and parent
+_call_chain conventions
+
+Example usage:
+
+import urllib2
+
+# set up authentication info
+authinfo = urllib2.HTTPBasicAuthHandler()
+authinfo.add_password(realm='PDQ Application',
+                      uri='https://mahler:8092/site-updates.py',
+                      user='klem',
+                      passwd='geheim$parole')
+
+proxy_support = urllib2.ProxyHandler({"http" : "http://ahad-haam:3128"})
+
+# build a new opener that adds authentication and caching FTP handlers
+opener = urllib2.build_opener(proxy_support, authinfo, urllib2.CacheFTPHandler)
+
+# install it
+urllib2.install_opener(opener)
+
+f = urllib2.urlopen('http://www.python.org/')
+
+
+"""
+
+# XXX issues:
+# If an authentication error handler that tries to perform
+# authentication for some reason but fails, how should the error be
+# signalled?  The client needs to know the HTTP error code.  But if
+# the handler knows that the problem was, e.g., that it didn't know
+# that hash algo that requested in the challenge, it would be good to
+# pass that information along to the client, too.
+# ftp errors aren't handled cleanly
+# check digest against correct (i.e. non-apache) implementation
+
+# Possible extensions:
+# complex proxies  XXX not sure what exactly was meant by this
+# abstract factory for opener
+
+import base64
+import hashlib
+import httplib
+import mimetools
+import os
+import posixpath
+import random
+import re
+import socket
+import sys
+import time
+import urlparse
+import bisect
+
+try:
+    from cStringIO import StringIO
+except ImportError:
+    from StringIO import StringIO
+
+from urllib import (unwrap, unquote, splittype, splithost, quote,
+     addinfourl, splitport, splittag,
+     splitattr, ftpwrapper, splituser, splitpasswd, splitvalue)
+
+# support for FileHandler, proxies via environment variables
+from urllib import localhost, url2pathname, getproxies, proxy_bypass
+
+# used in User-Agent header sent
+__version__ = sys.version[:3]
+
+_opener = None
+def urlopen(url, data=None, timeout=socket._GLOBAL_DEFAULT_TIMEOUT):
+    global _opener
+    if _opener is None:
+        _opener = build_opener()
+    return _opener.open(url, data, timeout)
+
+def install_opener(opener):
+    global _opener
+    _opener = opener
+
+# do these error classes make sense?
+# make sure all of the IOError stuff is overridden.  we just want to be
+# subtypes.
+
+class URLError(IOError):
+    # URLError is a sub-type of IOError, but it doesn't share any of
+    # the implementation.  need to override __init__ and __str__.
+    # It sets self.args for compatibility with other EnvironmentError
+    # subclasses, but args doesn't have the typical format with errno in
+    # slot 0 and strerror in slot 1.  This may be better than nothing.
+    def __init__(self, reason):
+        self.args = reason,
+        self.reason = reason
+
+    def __str__(self):
+        return '<urlopen error %s>' % self.reason
+
+class HTTPError(URLError, addinfourl):
+    """Raised when HTTP error occurs, but also acts like non-error return"""
+    __super_init = addinfourl.__init__
+
+    def __init__(self, url, code, msg, hdrs, fp):
+        self.code = code
+        self.msg = msg
+        self.hdrs = hdrs
+        self.fp = fp
+        self.filename = url
+        # The addinfourl classes depend on fp being a valid file
+        # object.  In some cases, the HTTPError may not have a valid
+        # file object.  If this happens, the simplest workaround is to
+        # not initialize the base classes.
+        if fp is not None:
+            self.__super_init(fp, hdrs, url, code)
+
+    def __str__(self):
+        return 'HTTP Error %s: %s' % (self.code, self.msg)
+
+# copied from cookielib.py
+_cut_port_re = re.compile(r":\d+$")
+def request_host(request):
+    """Return request-host, as defined by RFC 2965.
+
+    Variation from RFC: returned value is lowercased, for convenient
+    comparison.
+
+    """
+    url = request.get_full_url()
+    host = urlparse.urlparse(url)[1]
+    if host == "":
+        host = request.get_header("Host", "")
+
+    # remove port, if present
+    host = _cut_port_re.sub("", host, 1)
+    return host.lower()
+
+class Request:
+
+    def __init__(self, url, data=None, headers={},
+                 origin_req_host=None, unverifiable=False):
+        # unwrap('<URL:type://host/path>') --> 'type://host/path'
+        self.__original = unwrap(url)
+        self.__original, fragment = splittag(self.__original)
+        self.type = None
+        # self.__r_type is what's left after doing the splittype
+        self.host = None
+        self.port = None
+        self._tunnel_host = None
+        self.data = data
+        self.headers = {}
+        for key, value in headers.items():
+            self.add_header(key, value)
+        self.unredirected_hdrs = {}
+        if origin_req_host is None:
+            origin_req_host = request_host(self)
+        self.origin_req_host = origin_req_host
+        self.unverifiable = unverifiable
+
+    def __getattr__(self, attr):
+        # XXX this is a fallback mechanism to guard against these
+        # methods getting called in a non-standard order.  this may be
+        # too complicated and/or unnecessary.
+        # XXX should the __r_XXX attributes be public?
+        if attr[:12] == '_Request__r_':
+            name = attr[12:]
+            if hasattr(Request, 'get_' + name):
+                getattr(self, 'get_' + name)()
+                return getattr(self, attr)
+        raise AttributeError, attr
+
+    def get_method(self):
+        if self.has_data():
+            return "POST"
+        else:
+            return "GET"
+
+    # XXX these helper methods are lame
+
+    def add_data(self, data):
+        self.data = data
+
+    def has_data(self):
+        return self.data is not None
+
+    def get_data(self):
+        return self.data
+
+    def get_full_url(self):
+        return self.__original
+
+    def get_type(self):
+        if self.type is None:
+            self.type, self.__r_type = splittype(self.__original)
+            if self.type is None:
+                raise ValueError, "unknown url type: %s" % self.__original
+        return self.type
+
+    def get_host(self):
+        if self.host is None:
+            self.host, self.__r_host = splithost(self.__r_type)
+            if self.host:
+                self.host = unquote(self.host)
+        return self.host
+
+    def get_selector(self):
+        return self.__r_host
+
+    def set_proxy(self, host, type):
+        if self.type == 'https' and not self._tunnel_host:
+            self._tunnel_host = self.host
+        else:
+            self.type = type
+            self.__r_host = self.__original
+
+        self.host = host
+
+    def has_proxy(self):
+        return self.__r_host == self.__original
+
+    def get_origin_req_host(self):
+        return self.origin_req_host
+
+    def is_unverifiable(self):
+        return self.unverifiable
+
+    def add_header(self, key, val):
+        # useful for something like authentication
+        self.headers[key.capitalize()] = val
+
+    def add_unredirected_header(self, key, val):
+        # will not be added to a redirected request
+        self.unredirected_hdrs[key.capitalize()] = val
+
+    def has_header(self, header_name):
+        return (header_name in self.headers or
+                header_name in self.unredirected_hdrs)
+
+    def get_header(self, header_name, default=None):
+        return self.headers.get(
+            header_name,
+            self.unredirected_hdrs.get(header_name, default))
+
+    def header_items(self):
+        hdrs = self.unredirected_hdrs.copy()
+        hdrs.update(self.headers)
+        return hdrs.items()
+
+class OpenerDirector:
+    def __init__(self):
+        client_version = "Python-urllib/%s" % __version__
+        self.addheaders = [('User-agent', client_version)]
+        # manage the individual handlers
+        self.handlers = []
+        self.handle_open = {}
+        self.handle_error = {}
+        self.process_response = {}
+        self.process_request = {}
+
+    def add_handler(self, handler):
+        if not hasattr(handler, "add_parent"):
+            raise TypeError("expected BaseHandler instance, got %r" %
+                            type(handler))
+
+        added = False
+        for meth in dir(handler):
+            if meth in ["redirect_request", "do_open", "proxy_open"]:
+                # oops, coincidental match
+                continue
+
+            i = meth.find("_")
+            protocol = meth[:i]
+            condition = meth[i+1:]
+
+            if condition.startswith("error"):
+                j = condition.find("_") + i + 1
+                kind = meth[j+1:]
+                try:
+                    kind = int(kind)
+                except ValueError:
+                    pass
+                lookup = self.handle_error.get(protocol, {})
+                self.handle_error[protocol] = lookup
+            elif condition == "open":
+                kind = protocol
+                lookup = self.handle_open
+            elif condition == "response":
+                kind = protocol
+                lookup = self.process_response
+            elif condition == "request":
+                kind = protocol
+                lookup = self.process_request
+            else:
+                continue
+
+            handlers = lookup.setdefault(kind, [])
+            if handlers:
+                bisect.insort(handlers, handler)
+            else:
+                handlers.append(handler)
+            added = True
+
+        if added:
+            # the handlers must work in an specific order, the order
+            # is specified in a Handler attribute
+            bisect.insort(self.handlers, handler)
+            handler.add_parent(self)
+
+    def close(self):
+        # Only exists for backwards compatibility.
+        pass
+
+    def _call_chain(self, chain, kind, meth_name, *args):
+        # Handlers raise an exception if no one else should try to handle
+        # the request, or return None if they can't but another handler
+        # could.  Otherwise, they return the response.
+        handlers = chain.get(kind, ())
+        for handler in handlers:
+            func = getattr(handler, meth_name)
+
+            result = func(*args)
+            if result is not None:
+                return result
+
+    def open(self, fullurl, data=None, timeout=socket._GLOBAL_DEFAULT_TIMEOUT):
+        # accept a URL or a Request object
+        if isinstance(fullurl, basestring):
+            req = Request(fullurl, data)
+        else:
+            req = fullurl
+            if data is not None:
+                req.add_data(data)
+
+        req.timeout = timeout
+        protocol = req.get_type()
+
+        # pre-process request
+        meth_name = protocol+"_request"
+        for processor in self.process_request.get(protocol, []):
+            meth = getattr(processor, meth_name)
+            req = meth(req)
+
+        response = self._open(req, data)
+
+        # post-process response
+        meth_name = protocol+"_response"
+        for processor in self.process_response.get(protocol, []):
+            meth = getattr(processor, meth_name)
+            response = meth(req, response)
+
+        return response
+
+    def _open(self, req, data=None):
+        result = self._call_chain(self.handle_open, 'default',
+                                  'default_open', req)
+        if result:
+            return result
+
+        protocol = req.get_type()
+        result = self._call_chain(self.handle_open, protocol, protocol +
+                                  '_open', req)
+        if result:
+            return result
+
+        return self._call_chain(self.handle_open, 'unknown',
+                                'unknown_open', req)
+
+    def error(self, proto, *args):
+        if proto in ('http', 'https'):
+            # XXX http[s] protocols are special-cased
+            dict = self.handle_error['http'] # https is not different than http
+            proto = args[2]  # YUCK!
+            meth_name = 'http_error_%s' % proto
+            http_err = 1
+            orig_args = args
+        else:
+            dict = self.handle_error
+            meth_name = proto + '_error'
+            http_err = 0
+        args = (dict, proto, meth_name) + args
+        result = self._call_chain(*args)
+        if result:
+            return result
+
+        if http_err:
+            args = (dict, 'default', 'http_error_default') + orig_args
+            return self._call_chain(*args)
+
+# XXX probably also want an abstract factory that knows when it makes
+# sense to skip a superclass in favor of a subclass and when it might
+# make sense to include both
+
+def build_opener(*handlers):
+    """Create an opener object from a list of handlers.
+
+    The opener will use several default handlers, including support
+    for HTTP, FTP and when applicable, HTTPS.
+
+    If any of the handlers passed as arguments are subclasses of the
+    default handlers, the default handlers will not be used.
+    """
+    import types
+    def isclass(obj):
+        return isinstance(obj, (types.ClassType, type))
+
+    opener = OpenerDirector()
+    default_classes = [ProxyHandler, UnknownHandler, HTTPHandler,
+                       HTTPDefaultErrorHandler, HTTPRedirectHandler,
+                       FTPHandler, FileHandler, HTTPErrorProcessor]
+    if hasattr(httplib, 'HTTPS'):
+        default_classes.append(HTTPSHandler)
+    skip = set()
+    for klass in default_classes:
+        for check in handlers:
+            if isclass(check):
+                if issubclass(check, klass):
+                    skip.add(klass)
+            elif isinstance(check, klass):
+                skip.add(klass)
+    for klass in skip:
+        default_classes.remove(klass)
+
+    for klass in default_classes:
+        opener.add_handler(klass())
+
+    for h in handlers:
+        if isclass(h):
+            h = h()
+        opener.add_handler(h)
+    return opener
+
+class BaseHandler:
+    handler_order = 500
+
+    def add_parent(self, parent):
+        self.parent = parent
+
+    def close(self):
+        # Only exists for backwards compatibility
+        pass
+
+    def __lt__(self, other):
+        if not hasattr(other, "handler_order"):
+            # Try to preserve the old behavior of having custom classes
+            # inserted after default ones (works only for custom user
+            # classes which are not aware of handler_order).
+            return True
+        return self.handler_order < other.handler_order
+
+
+class HTTPErrorProcessor(BaseHandler):
+    """Process HTTP error responses."""
+    handler_order = 1000  # after all other processing
+
+    def http_response(self, request, response):
+        code, msg, hdrs = response.code, response.msg, response.info()
+
+        # According to RFC 2616, "2xx" code indicates that the client's
+        # request was successfully received, understood, and accepted.
+        if not (200 <= code < 300):
+            response = self.parent.error(
+                'http', request, response, code, msg, hdrs)
+
+        return response
+
+    https_response = http_response
+
+class HTTPDefaultErrorHandler(BaseHandler):
+    def http_error_default(self, req, fp, code, msg, hdrs):
+        raise HTTPError(req.get_full_url(), code, msg, hdrs, fp)
+
+class HTTPRedirectHandler(BaseHandler):
+    # maximum number of redirections to any single URL
+    # this is needed because of the state that cookies introduce
+    max_repeats = 4
+    # maximum total number of redirections (regardless of URL) before
+    # assuming we're in a loop
+    max_redirections = 10
+
+    def redirect_request(self, req, fp, code, msg, headers, newurl):
+        """Return a Request or None in response to a redirect.
+
+        This is called by the http_error_30x methods when a
+        redirection response is received.  If a redirection should
+        take place, return a new Request to allow http_error_30x to
+        perform the redirect.  Otherwise, raise HTTPError if no-one
+        else should try to handle this url.  Return None if you can't
+        but another Handler might.
+        """
+        m = req.get_method()
+        if (code in (301, 302, 303, 307) and m in ("GET", "HEAD")
+            or code in (301, 302, 303) and m == "POST"):
+            # Strictly (according to RFC 2616), 301 or 302 in response
+            # to a POST MUST NOT cause a redirection without confirmation
+            # from the user (of urllib2, in this case).  In practice,
+            # essentially all clients do redirect in this case, so we
+            # do the same.
+            # be conciliant with URIs containing a space
+            newurl = newurl.replace(' ', '%20')
+            newheaders = dict((k,v) for k,v in req.headers.items()
+                              if k.lower() not in ("content-length", "content-type")
+                             )
+            return Request(newurl,
+                           headers=newheaders,
+                           origin_req_host=req.get_origin_req_host(),
+                           unverifiable=True)
+        else:
+            raise HTTPError(req.get_full_url(), code, msg, headers, fp)
+
+    # Implementation note: To avoid the server sending us into an
+    # infinite loop, the request object needs to track what URLs we
+    # have already seen.  Do this by adding a handler-specific
+    # attribute to the Request object.
+    def http_error_302(self, req, fp, code, msg, headers):
+        # Some servers (incorrectly) return multiple Location headers
+        # (so probably same goes for URI).  Use first header.
+        if 'location' in headers:
+            newurl = headers.getheaders('location')[0]
+        elif 'uri' in headers:
+            newurl = headers.getheaders('uri')[0]
+        else:
+            return
+
+        # fix a possible malformed URL
+        urlparts = urlparse.urlparse(newurl)
+        if not urlparts.path:
+            urlparts = list(urlparts)
+            urlparts[2] = "/"
+        newurl = urlparse.urlunparse(urlparts)
+
+        newurl = urlparse.urljoin(req.get_full_url(), newurl)
+
+        # XXX Probably want to forget about the state of the current
+        # request, although that might interact poorly with other
+        # handlers that also use handler-specific request attributes
+        new = self.redirect_request(req, fp, code, msg, headers, newurl)
+        if new is None:
+            return
+
+        # loop detection
+        # .redirect_dict has a key url if url was previously visited.
+        if hasattr(req, 'redirect_dict'):
+            visited = new.redirect_dict = req.redirect_dict
+            if (visited.get(newurl, 0) >= self.max_repeats or
+                len(visited) >= self.max_redirections):
+                raise HTTPError(req.get_full_url(), code,
+                                self.inf_msg + msg, headers, fp)
+        else:
+            visited = new.redirect_dict = req.redirect_dict = {}
+        visited[newurl] = visited.get(newurl, 0) + 1
+
+        # Don't close the fp until we are sure that we won't use it
+        # with HTTPError.
+        fp.read()
+        fp.close()
+
+        return self.parent.open(new, timeout=req.timeout)
+
+    http_error_301 = http_error_303 = http_error_307 = http_error_302
+
+    inf_msg = "The HTTP server returned a redirect error that would " \
+              "lead to an infinite loop.\n" \
+              "The last 30x error message was:\n"
+
+
+def _parse_proxy(proxy):
+    """Return (scheme, user, password, host/port) given a URL or an authority.
+
+    If a URL is supplied, it must have an authority (host:port) component.
+    According to RFC 3986, having an authority component means the URL must
+    have two slashes after the scheme:
+
+    >>> _parse_proxy('file:/ftp.example.com/')
+    Traceback (most recent call last):
+    ValueError: proxy URL with no authority: 'file:/ftp.example.com/'
+
+    The first three items of the returned tuple may be None.
+
+    Examples of authority parsing:
+
+    >>> _parse_proxy('proxy.example.com')
+    (None, None, None, 'proxy.example.com')
+    >>> _parse_proxy('proxy.example.com:3128')
+    (None, None, None, 'proxy.example.com:3128')
+
+    The authority component may optionally include userinfo (assumed to be
+    username:password):
+
+    >>> _parse_proxy('joe:password at proxy.example.com')
+    (None, 'joe', 'password', 'proxy.example.com')
+    >>> _parse_proxy('joe:password at proxy.example.com:3128')
+    (None, 'joe', 'password', 'proxy.example.com:3128')
+
+    Same examples, but with URLs instead:
+
+    >>> _parse_proxy('http://proxy.example.com/')
+    ('http', None, None, 'proxy.example.com')
+    >>> _parse_proxy('http://proxy.example.com:3128/')
+    ('http', None, None, 'proxy.example.com:3128')
+    >>> _parse_proxy('http://joe:password@proxy.example.com/')
+    ('http', 'joe', 'password', 'proxy.example.com')
+    >>> _parse_proxy('http://joe:password@proxy.example.com:3128')
+    ('http', 'joe', 'password', 'proxy.example.com:3128')
+
+    Everything after the authority is ignored:
+
+    >>> _parse_proxy('ftp://joe:password@proxy.example.com/rubbish:3128')
+    ('ftp', 'joe', 'password', 'proxy.example.com')
+
+    Test for no trailing '/' case:
+
+    >>> _parse_proxy('http://joe:password@proxy.example.com')
+    ('http', 'joe', 'password', 'proxy.example.com')
+
+    """
+    scheme, r_scheme = splittype(proxy)
+    if not r_scheme.startswith("/"):
+        # authority
+        scheme = None
+        authority = proxy
+    else:
+        # URL
+        if not r_scheme.startswith("//"):
+            raise ValueError("proxy URL with no authority: %r" % proxy)
+        # We have an authority, so for RFC 3986-compliant URLs (by ss 3.
+        # and 3.3.), path is empty or starts with '/'
+        end = r_scheme.find("/", 2)
+        if end == -1:
+            end = None
+        authority = r_scheme[2:end]
+    userinfo, hostport = splituser(authority)
+    if userinfo is not None:
+        user, password = splitpasswd(userinfo)
+    else:
+        user = password = None
+    return scheme, user, password, hostport
+
+class ProxyHandler(BaseHandler):
+    # Proxies must be in front
+    handler_order = 100
+
+    def __init__(self, proxies=None):
+        if proxies is None:
+            proxies = getproxies()
+        assert hasattr(proxies, 'has_key'), "proxies must be a mapping"
+        self.proxies = proxies
+        for type, url in proxies.items():
+            setattr(self, '%s_open' % type,
+                    lambda r, proxy=url, type=type, meth=self.proxy_open: \
+                    meth(r, proxy, type))
+
+    def proxy_open(self, req, proxy, type):
+        orig_type = req.get_type()
+        proxy_type, user, password, hostport = _parse_proxy(proxy)
+
+        if proxy_type is None:
+            proxy_type = orig_type
+
+        if req.host and proxy_bypass(req.host):
+            return None
+
+        if user and password:
+            user_pass = '%s:%s' % (unquote(user), unquote(password))
+            creds = base64.b64encode(user_pass).strip()
+            req.add_header('Proxy-authorization', 'Basic ' + creds)
+        hostport = unquote(hostport)
+        req.set_proxy(hostport, proxy_type)
+
+        if orig_type == proxy_type or orig_type == 'https':
+            # let other handlers take care of it
+            return None
+        else:
+            # need to start over, because the other handlers don't
+            # grok the proxy's URL type
+            # e.g. if we have a constructor arg proxies like so:
+            # {'http': 'ftp://proxy.example.com'}, we may end up turning
+            # a request for http://acme.example.com/a into one for
+            # ftp://proxy.example.com/a
+            return self.parent.open(req, timeout=req.timeout)
+
+class HTTPPasswordMgr:
+
+    def __init__(self):
+        self.passwd = {}
+
+    def add_password(self, realm, uri, user, passwd):
+        # uri could be a single URI or a sequence
+        if isinstance(uri, basestring):
+            uri = [uri]
+        if not realm in self.passwd:
+            self.passwd[realm] = {}
+        for default_port in True, False:
+            reduced_uri = tuple(
+                [self.reduce_uri(u, default_port) for u in uri])
+            self.passwd[realm][reduced_uri] = (user, passwd)
+
+    def find_user_password(self, realm, authuri):
+        domains = self.passwd.get(realm, {})
+        for default_port in True, False:
+            reduced_authuri = self.reduce_uri(authuri, default_port)
+            for uris, authinfo in domains.iteritems():
+                for uri in uris:
+                    if self.is_suburi(uri, reduced_authuri):
+                        return authinfo
+        return None, None
+
+    def reduce_uri(self, uri, default_port=True):
+        """Accept authority or URI and extract only the authority and path."""
+        # note HTTP URLs do not have a userinfo component
+        parts = urlparse.urlsplit(uri)
+        if parts[1]:
+            # URI
+            scheme = parts[0]
+            authority = parts[1]
+            path = parts[2] or '/'
+        else:
+            # host or host:port
+            scheme = None
+            authority = uri
+            path = '/'
+        host, port = splitport(authority)
+        if default_port and port is None and scheme is not None:
+            dport = {"http": 80,
+                     "https": 443,
+                     }.get(scheme)
+            if dport is not None:
+                authority = "%s:%d" % (host, dport)
+        return authority, path
+
+    def is_suburi(self, base, test):
+        """Check if test is below base in a URI tree
+
+        Both args must be URIs in reduced form.
+        """
+        if base == test:
+            return True
+        if base[0] != test[0]:
+            return False
+        common = posixpath.commonprefix((base[1], test[1]))
+        if len(common) == len(base[1]):
+            return True
+        return False
+
+
+class HTTPPasswordMgrWithDefaultRealm(HTTPPasswordMgr):
+
+    def find_user_password(self, realm, authuri):
+        user, password = HTTPPasswordMgr.find_user_password(self, realm,
+                                                            authuri)
+        if user is not None:
+            return user, password
+        return HTTPPasswordMgr.find_user_password(self, None, authuri)
+
+
+class AbstractBasicAuthHandler:
+
+    # XXX this allows for multiple auth-schemes, but will stupidly pick
+    # the last one with a realm specified.
+
+    # allow for double- and single-quoted realm values
+    # (single quotes are a violation of the RFC, but appear in the wild)
+    rx = re.compile('(?:.*,)*[ \t]*([^ \t]+)[ \t]+'
+                    'realm=(["\'])(.*?)\\2', re.I)
+
+    # XXX could pre-emptively send auth info already accepted (RFC 2617,
+    # end of section 2, and section 1.2 immediately after "credentials"
+    # production).
+
+    def __init__(self, password_mgr=None):
+        if password_mgr is None:
+            password_mgr = HTTPPasswordMgr()
+        self.passwd = password_mgr
+        self.add_password = self.passwd.add_password
+        self.retried = 0
+
+    def reset_retry_count(self):
+        self.retried = 0
+
+    def http_error_auth_reqed(self, authreq, host, req, headers):
+        # host may be an authority (without userinfo) or a URL with an
+        # authority
+        # XXX could be multiple headers
+        authreq = headers.get(authreq, None)
+
+        if self.retried > 5:
+            # retry sending the username:password 5 times before failing.
+            raise HTTPError(req.get_full_url(), 401, "basic auth failed",
+                            headers, None)
+        else:
+            self.retried += 1
+
+        if authreq:
+            mo = AbstractBasicAuthHandler.rx.search(authreq)
+            if mo:
+                scheme, quote, realm = mo.groups()
+                if scheme.lower() == 'basic':
+                    response = self.retry_http_basic_auth(host, req, realm)
+                    if response and response.code != 401:
+                        self.retried = 0
+                    return response
+
+    def retry_http_basic_auth(self, host, req, realm):
+        user, pw = self.passwd.find_user_password(realm, host)
+        if pw is not None:
+            raw = "%s:%s" % (user, pw)
+            auth = 'Basic %s' % base64.b64encode(raw).strip()
+            if req.headers.get(self.auth_header, None) == auth:
+                return None
+            req.add_unredirected_header(self.auth_header, auth)
+            return self.parent.open(req, timeout=req.timeout)
+        else:
+            return None
+
+
+class HTTPBasicAuthHandler(AbstractBasicAuthHandler, BaseHandler):
+
+    auth_header = 'Authorization'
+
+    def http_error_401(self, req, fp, code, msg, headers):
+        url = req.get_full_url()
+        response = self.http_error_auth_reqed('www-authenticate',
+                                              url, req, headers)
+        self.reset_retry_count()
+        return response
+
+
+class ProxyBasicAuthHandler(AbstractBasicAuthHandler, BaseHandler):
+
+    auth_header = 'Proxy-authorization'
+
+    def http_error_407(self, req, fp, code, msg, headers):
+        # http_error_auth_reqed requires that there is no userinfo component in
+        # authority.  Assume there isn't one, since urllib2 does not (and
+        # should not, RFC 3986 s. 3.2.1) support requests for URLs containing
+        # userinfo.
+        authority = req.get_host()
+        response = self.http_error_auth_reqed('proxy-authenticate',
+                                          authority, req, headers)
+        self.reset_retry_count()
+        return response
+
+
+def randombytes(n):
+    """Return n random bytes."""
+    # Use /dev/urandom if it is available.  Fall back to random module
+    # if not.  It might be worthwhile to extend this function to use
+    # other platform-specific mechanisms for getting random bytes.
+    if os.path.exists("/dev/urandom"):
+        f = open("/dev/urandom")
+        s = f.read(n)
+        f.close()
+        return s
+    else:
+        L = [chr(random.randrange(0, 256)) for i in range(n)]
+        return "".join(L)
+
+class AbstractDigestAuthHandler:
+    # Digest authentication is specified in RFC 2617.
+
+    # XXX The client does not inspect the Authentication-Info header
+    # in a successful response.
+
+    # XXX It should be possible to test this implementation against
+    # a mock server that just generates a static set of challenges.
+
+    # XXX qop="auth-int" supports is shaky
+
+    def __init__(self, passwd=None):
+        if passwd is None:
+            passwd = HTTPPasswordMgr()
+        self.passwd = passwd
+        self.add_password = self.passwd.add_password
+        self.retried = 0
+        self.nonce_count = 0
+        self.last_nonce = None
+
+    def reset_retry_count(self):
+        self.retried = 0
+
+    def http_error_auth_reqed(self, auth_header, host, req, headers):
+        authreq = headers.get(auth_header, None)
+        if self.retried > 5:
+            # Don't fail endlessly - if we failed once, we'll probably
+            # fail a second time. Hm. Unless the Password Manager is
+            # prompting for the information. Crap. This isn't great
+            # but it's better than the current 'repeat until recursion
+            # depth exceeded' approach <wink>
+            raise HTTPError(req.get_full_url(), 401, "digest auth failed",
+                            headers, None)
+        else:
+            self.retried += 1
+        if authreq:
+            scheme = authreq.split()[0]
+            if scheme.lower() == 'digest':
+                return self.retry_http_digest_auth(req, authreq)
+
+    def retry_http_digest_auth(self, req, auth):
+        token, challenge = auth.split(' ', 1)
+        chal = parse_keqv_list(parse_http_list(challenge))
+        auth = self.get_authorization(req, chal)
+        if auth:
+            auth_val = 'Digest %s' % auth
+            if req.headers.get(self.auth_header, None) == auth_val:
+                return None
+            req.add_unredirected_header(self.auth_header, auth_val)
+            resp = self.parent.open(req, timeout=req.timeout)
+            return resp
+
+    def get_cnonce(self, nonce):
+        # The cnonce-value is an opaque
+        # quoted string value provided by the client and used by both client
+        # and server to avoid chosen plaintext attacks, to provide mutual
+        # authentication, and to provide some message integrity protection.
+        # This isn't a fabulous effort, but it's probably Good Enough.
+        dig = hashlib.sha1("%s:%s:%s:%s" % (self.nonce_count, nonce, time.ctime(),
+                                            randombytes(8))).hexdigest()
+        return dig[:16]
+
+    def get_authorization(self, req, chal):
+        try:
+            realm = chal['realm']
+            nonce = chal['nonce']
+            qop = chal.get('qop')
+            algorithm = chal.get('algorithm', 'MD5')
+            # mod_digest doesn't send an opaque, even though it isn't
+            # supposed to be optional
+            opaque = chal.get('opaque', None)
+        except KeyError:
+            return None
+
+        H, KD = self.get_algorithm_impls(algorithm)
+        if H is None:
+            return None
+
+        user, pw = self.passwd.find_user_password(realm, req.get_full_url())
+        if user is None:
+            return None
+
+        # XXX not implemented yet
+        if req.has_data():
+            entdig = self.get_entity_digest(req.get_data(), chal)
+        else:
+            entdig = None
+
+        A1 = "%s:%s:%s" % (user, realm, pw)
+        A2 = "%s:%s" % (req.get_method(),
+                        # XXX selector: what about proxies and full urls
+                        req.get_selector())
+        if qop == 'auth':
+            if nonce == self.last_nonce:
+                self.nonce_count += 1
+            else:
+                self.nonce_count = 1
+                self.last_nonce = nonce
+
+            ncvalue = '%08x' % self.nonce_count
+            cnonce = self.get_cnonce(nonce)
+            noncebit = "%s:%s:%s:%s:%s" % (nonce, ncvalue, cnonce, qop, H(A2))
+            respdig = KD(H(A1), noncebit)
+        elif qop is None:
+            respdig = KD(H(A1), "%s:%s" % (nonce, H(A2)))
+        else:
+            # XXX handle auth-int.
+            raise URLError("qop '%s' is not supported." % qop)
+
+        # XXX should the partial digests be encoded too?
+
+        base = 'username="%s", realm="%s", nonce="%s", uri="%s", ' \
+               'response="%s"' % (user, realm, nonce, req.get_selector(),
+                                  respdig)
+        if opaque:
+            base += ', opaque="%s"' % opaque
+        if entdig:
+            base += ', digest="%s"' % entdig
+        base += ', algorithm="%s"' % algorithm
+        if qop:
+            base += ', qop=auth, nc=%s, cnonce="%s"' % (ncvalue, cnonce)
+        return base
+
+    def get_algorithm_impls(self, algorithm):
+        # algorithm should be case-insensitive according to RFC2617
+        algorithm = algorithm.upper()
+        # lambdas assume digest modules are imported at the top level
+        if algorithm == 'MD5':
+            H = lambda x: hashlib.md5(x).hexdigest()
+        elif algorithm == 'SHA':
+            H = lambda x: hashlib.sha1(x).hexdigest()
+        # XXX MD5-sess
+        KD = lambda s, d: H("%s:%s" % (s, d))
+        return H, KD
+
+    def get_entity_digest(self, data, chal):
+        # XXX not implemented yet
+        return None
+
+
+class HTTPDigestAuthHandler(BaseHandler, AbstractDigestAuthHandler):
+    """An authentication protocol defined by RFC 2069
+
+    Digest authentication improves on basic authentication because it
+    does not transmit passwords in the clear.
+    """
+
+    auth_header = 'Authorization'
+    handler_order = 490  # before Basic auth
+
+    def http_error_401(self, req, fp, code, msg, headers):
+        host = urlparse.urlparse(req.get_full_url())[1]
+        retry = self.http_error_auth_reqed('www-authenticate',
+                                           host, req, headers)
+        self.reset_retry_count()
+        return retry
+
+
+class ProxyDigestAuthHandler(BaseHandler, AbstractDigestAuthHandler):
+
+    auth_header = 'Proxy-Authorization'
+    handler_order = 490  # before Basic auth
+
+    def http_error_407(self, req, fp, code, msg, headers):
+        host = req.get_host()
+        retry = self.http_error_auth_reqed('proxy-authenticate',
+                                           host, req, headers)
+        self.reset_retry_count()
+        return retry
+
+class AbstractHTTPHandler(BaseHandler):
+
+    def __init__(self, debuglevel=0):
+        self._debuglevel = debuglevel
+
+    def set_http_debuglevel(self, level):
+        self._debuglevel = level
+
+    def do_request_(self, request):
+        host = request.get_host()
+        if not host:
+            raise URLError('no host given')
+
+        if request.has_data():  # POST
+            data = request.get_data()
+            if not request.has_header('Content-type'):
+                request.add_unredirected_header(
+                    'Content-type',
+                    'application/x-www-form-urlencoded')
+            if not request.has_header('Content-length'):
+                request.add_unredirected_header(
+                    'Content-length', '%d' % len(data))
+
+        sel_host = host
+        if request.has_proxy():
+            scheme, sel = splittype(request.get_selector())
+            sel_host, sel_path = splithost(sel)
+
+        if not request.has_header('Host'):
+            request.add_unredirected_header('Host', sel_host)
+        for name, value in self.parent.addheaders:
+            name = name.capitalize()
+            if not request.has_header(name):
+                request.add_unredirected_header(name, value)
+
+        return request
+
+    def do_open(self, http_class, req):
+        """Return an addinfourl object for the request, using http_class.
+
+        http_class must implement the HTTPConnection API from httplib.
+        The addinfourl return value is a file-like object.  It also
+        has methods and attributes including:
+            - info(): return a mimetools.Message object for the headers
+            - geturl(): return the original request URL
+            - code: HTTP status code
+        """
+        host = req.get_host()
+        if not host:
+            raise URLError('no host given')
+
+        h = http_class(host, timeout=req.timeout) # will parse host:port
+        h.set_debuglevel(self._debuglevel)
+
+        headers = dict(req.unredirected_hdrs)
+        headers.update(dict((k, v) for k, v in req.headers.items()
+                            if k not in headers))
+
+        # We want to make an HTTP/1.1 request, but the addinfourl
+        # class isn't prepared to deal with a persistent connection.
+        # It will try to read all remaining data from the socket,
+        # which will block while the server waits for the next request.
+        # So make sure the connection gets closed after the (only)
+        # request.
+        headers["Connection"] = "close"
+        headers = dict(
+            (name.title(), val) for name, val in headers.items())
+
+        if req._tunnel_host:
+            tunnel_headers = {}
+            proxy_auth_hdr = "Proxy-Authorization"
+            if proxy_auth_hdr in headers:
+                tunnel_headers[proxy_auth_hdr] = headers[proxy_auth_hdr]
+                # Proxy-Authorization should not be sent to origin
+                # server.
+                del headers[proxy_auth_hdr]
+            h.set_tunnel(req._tunnel_host, headers=tunnel_headers)
+
+        try:
+            h.request(req.get_method(), req.get_selector(), req.data, headers)
+            try:
+                r = h.getresponse(buffering=True)
+            except TypeError: #buffering kw not supported
+                r = h.getresponse()
+        except socket.error, err: # XXX what error?
+            h.close()
+            raise URLError(err)
+
+        # Pick apart the HTTPResponse object to get the addinfourl
+        # object initialized properly.
+
+        # Wrap the HTTPResponse object in socket's file object adapter
+        # for Windows.  That adapter calls recv(), so delegate recv()
+        # to read().  This weird wrapping allows the returned object to
+        # have readline() and readlines() methods.
+
+        # XXX It might be better to extract the read buffering code
+        # out of socket._fileobject() and into a base class.
+
+        r.recv = r.read
+        fp = socket._fileobject(r, close=True)
+
+        resp = addinfourl(fp, r.msg, req.get_full_url())
+        resp.code = r.status
+        resp.msg = r.reason
+        return resp
+
+
+class HTTPHandler(AbstractHTTPHandler):
+
+    def http_open(self, req):
+        return self.do_open(httplib.HTTPConnection, req)
+
+    http_request = AbstractHTTPHandler.do_request_
+
+if hasattr(httplib, 'HTTPS'):
+    class HTTPSHandler(AbstractHTTPHandler):
+
+        def https_open(self, req):
+            return self.do_open(httplib.HTTPSConnection, req)
+
+        https_request = AbstractHTTPHandler.do_request_
+
+class HTTPCookieProcessor(BaseHandler):
+    def __init__(self, cookiejar=None):
+        import cookielib
+        if cookiejar is None:
+            cookiejar = cookielib.CookieJar()
+        self.cookiejar = cookiejar
+
+    def http_request(self, request):
+        self.cookiejar.add_cookie_header(request)
+        return request
+
+    def http_response(self, request, response):
+        self.cookiejar.extract_cookies(response, request)
+        return response
+
+    https_request = http_request
+    https_response = http_response
+
+class UnknownHandler(BaseHandler):
+    def unknown_open(self, req):
+        type = req.get_type()
+        raise URLError('unknown url type: %s' % type)
+
+def parse_keqv_list(l):
+    """Parse list of key=value strings where keys are not duplicated."""
+    parsed = {}
+    for elt in l:
+        k, v = elt.split('=', 1)
+        if v[0] == '"' and v[-1] == '"':
+            v = v[1:-1]
+        parsed[k] = v
+    return parsed
+
+def parse_http_list(s):
+    """Parse lists as described by RFC 2068 Section 2.
+
+    In particular, parse comma-separated lists where the elements of
+    the list may include quoted-strings.  A quoted-string could
+    contain a comma.  A non-quoted string could have quotes in the
+    middle.  Neither commas nor quotes count if they are escaped.
+    Only double-quotes count, not single-quotes.
+    """
+    res = []
+    part = ''
+
+    escape = quote = False
+    for cur in s:
+        if escape:
+            part += cur
+            escape = False
+            continue
+        if quote:
+            if cur == '\\':
+                escape = True
+                continue
+            elif cur == '"':
+                quote = False
+            part += cur
+            continue
+
+        if cur == ',':
+            res.append(part)
+            part = ''
+            continue
+
+        if cur == '"':
+            quote = True
+
+        part += cur
+
+    # append last part
+    if part:
+        res.append(part)
+
+    return [part.strip() for part in res]
+
+def _safe_gethostbyname(host):
+    try:
+        return socket.gethostbyname(host)
+    except socket.gaierror:
+        return None
+
+class FileHandler(BaseHandler):
+    # Use local file or FTP depending on form of URL
+    def file_open(self, req):
+        url = req.get_selector()
+        if url[:2] == '//' and url[2:3] != '/' and (req.host and
+                req.host != 'localhost'):
+            req.type = 'ftp'
+            return self.parent.open(req)
+        else:
+            return self.open_local_file(req)
+
+    # names for the localhost
+    names = None
+    def get_names(self):
+        if FileHandler.names is None:
+            try:
+                FileHandler.names = tuple(
+                    socket.gethostbyname_ex('localhost')[2] +
+                    socket.gethostbyname_ex(socket.gethostname())[2])
+            except socket.gaierror:
+                FileHandler.names = (socket.gethostbyname('localhost'),)
+        return FileHandler.names
+
+    # not entirely sure what the rules are here
+    def open_local_file(self, req):
+        import email.utils
+        import mimetypes
+        host = req.get_host()
+        filename = req.get_selector()
+        localfile = url2pathname(filename)
+        try:
+            stats = os.stat(localfile)
+            size = stats.st_size
+            modified = email.utils.formatdate(stats.st_mtime, usegmt=True)
+            mtype = mimetypes.guess_type(filename)[0]
+            headers = mimetools.Message(StringIO(
+                'Content-type: %s\nContent-length: %d\nLast-modified: %s\n' %
+                (mtype or 'text/plain', size, modified)))
+            if host:
+                host, port = splitport(host)
+            if not host or \
+                (not port and _safe_gethostbyname(host) in self.get_names()):
+                if host:
+                    origurl = 'file://' + host + filename
+                else:
+                    origurl = 'file://' + filename
+                return addinfourl(open(localfile, 'rb'), headers, origurl)
+        except OSError, msg:
+            # urllib2 users shouldn't expect OSErrors coming from urlopen()
+            raise URLError(msg)
+        raise URLError('file not on local host')
+
+class FTPHandler(BaseHandler):
+    def ftp_open(self, req):
+        import ftplib
+        import mimetypes
+        host = req.get_host()
+        if not host:
+            raise URLError('ftp error: no host given')
+        host, port = splitport(host)
+        if port is None:
+            port = ftplib.FTP_PORT
+        else:
+            port = int(port)
+
+        # username/password handling
+        user, host = splituser(host)
+        if user:
+            user, passwd = splitpasswd(user)
+        else:
+            passwd = None
+        host = unquote(host)
+        user = user or ''
+        passwd = passwd or ''
+
+        try:
+            host = socket.gethostbyname(host)
+        except socket.error, msg:
+            raise URLError(msg)
+        path, attrs = splitattr(req.get_selector())
+        dirs = path.split('/')
+        dirs = map(unquote, dirs)
+        dirs, file = dirs[:-1], dirs[-1]
+        if dirs and not dirs[0]:
+            dirs = dirs[1:]
+        try:
+            fw = self.connect_ftp(user, passwd, host, port, dirs, req.timeout)
+            type = file and 'I' or 'D'
+            for attr in attrs:
+                attr, value = splitvalue(attr)
+                if attr.lower() == 'type' and \
+                   value in ('a', 'A', 'i', 'I', 'd', 'D'):
+                    type = value.upper()
+            fp, retrlen = fw.retrfile(file, type)
+            headers = ""
+            mtype = mimetypes.guess_type(req.get_full_url())[0]
+            if mtype:
+                headers += "Content-type: %s\n" % mtype
+            if retrlen is not None and retrlen >= 0:
+                headers += "Content-length: %d\n" % retrlen
+            sf = StringIO(headers)
+            headers = mimetools.Message(sf)
+            return addinfourl(fp, headers, req.get_full_url())
+        except ftplib.all_errors, msg:
+            raise URLError, ('ftp error: %s' % msg), sys.exc_info()[2]
+
+    def connect_ftp(self, user, passwd, host, port, dirs, timeout):
+        fw = ftpwrapper(user, passwd, host, port, dirs, timeout)
+##        fw.ftp.set_debuglevel(1)
+        return fw
+
+class CacheFTPHandler(FTPHandler):
+    # XXX would be nice to have pluggable cache strategies
+    # XXX this stuff is definitely not thread safe
+    def __init__(self):
+        self.cache = {}
+        self.timeout = {}
+        self.soonest = 0
+        self.delay = 60
+        self.max_conns = 16
+
+    def setTimeout(self, t):
+        self.delay = t
+
+    def setMaxConns(self, m):
+        self.max_conns = m
+
+    def connect_ftp(self, user, passwd, host, port, dirs, timeout):
+        key = user, host, port, '/'.join(dirs), timeout
+        if key in self.cache:
+            self.timeout[key] = time.time() + self.delay
+        else:
+            self.cache[key] = ftpwrapper(user, passwd, host, port, dirs, timeout)
+            self.timeout[key] = time.time() + self.delay
+        self.check_cache()
+        return self.cache[key]
+
+    def check_cache(self):
+        # first check for old ones
+        t = time.time()
+        if self.soonest <= t:
+            for k, v in self.timeout.items():
+                if v < t:
+                    self.cache[k].close()
+                    del self.cache[k]
+                    del self.timeout[k]
+        self.soonest = min(self.timeout.values())
+
+        # then check the size
+        if len(self.cache) == self.max_conns:
+            for k, v in self.timeout.items():
+                if v == self.soonest:
+                    del self.cache[k]
+                    del self.timeout[k]
+                    break
+            self.soonest = min(self.timeout.values())
diff --git a/lib_pypy/_pypy_irc_topic.py b/lib_pypy/_pypy_irc_topic.py
--- a/lib_pypy/_pypy_irc_topic.py
+++ b/lib_pypy/_pypy_irc_topic.py
@@ -1,117 +1,6 @@
-"""qvfgbcvna naq hgbcvna punvef
-qlfgbcvna naq hgbcvna punvef
-V'z fbeel, pbhyq lbh cyrnfr abg nterr jvgu gur png nf jryy?
-V'z fbeel, pbhyq lbh cyrnfr abg nterr jvgu gur punve nf jryy?
-jr cnffrq gur RH erivrj
-cbfg RhebClguba fcevag fgnegf 12.IVV.2007, 10nz
-RhebClguba raqrq
-n Pyrna Ragrecevfrf cebqhpgvba
-npnqrzl vf n pbzcyvpngrq ebyr tnzr
-npnqrzvn vf n pbzcyvpngrq ebyr tnzr
-jbexvat pbqr vf crn fbhc
-abg lbhe snhyg, zber yvxr vg'f n zbivat gnetrg
-guvf fragrapr vf snyfr
-abguvat vf gehr
-Yncfnat Fbhpubat
-Oenpunzhgnaqn
-fbeel, V'yy grnpu gur pnpghf ubj gb fjvz yngre
-Jul fb znal znal znal znal znal ivbyvaf?
-Jul fb znal znal znal znal znal bowrpgf?
-"eha njnl naq yvir ba n snez" nccebnpu gb fbsgjner qrirybczrag
-"va snpg, lbh zvtug xabj zber nobhg gur genafyngvba gbbypunva nsgre znfgrevat eclguba guna fbzr angvir fcrnxre xabjf nobhg uvf zbgure gbathr" - kbeNkNk
-"jurer qvq nyy gur ivbyvaf tb?"
-- ClCl fgnghf oybt: uggc://zberclcl.oybtfcbg.pbz/
-uggc://kxpq.pbz/353/
-pnfhnyvgl ivbyngvbaf naq sylvat
-wrgmg abpu fpubxbynqvtre
-R09 2X @PNN:85?
-vs lbh'er gelvat gb oybj hc fghss, jub pnerf?
-vs fghss oybjf hc, lbh pner
-2008 jvyy or gur lrne bs clcl ba gur qrfxgbc
-2008 jvyy or gur lrne bs gur qrfxgbc ba #clcl
-2008 jvyy or gur lrne bs gur qrfxgbc ba #clcl, Wnahnel jvyy or gur zbagu bs gur nyc gbcf
-lrf, ohg jung'g gur frafr bs 0 < "qhena qhena"
-eclguba: flagnk naq frznagvpf bs clguba, fcrrq bs p, erfgevpgvbaf bs wnin naq pbzcvyre reebe zrffntrf nf crargenoyr nf ZHZCF
+"""eclguba: flagnk naq frznagvpf bs clguba, fcrrq bs p, erfgevpgvbaf bs wnin naq pbzcvyre reebe zrffntrf nf crargenoyr nf ZHZCF
 pglcrf unf n fcva bs 1/3
 ' ' vf n fcnpr gbb
-2009 jvyy or gur lrne bs WVG ba gur qrfxgbc
-N ynathntr vf n qvnyrpg jvgu na nezl naq anil
-gbcvpf ner sbe gur srroyr zvaqrq
-2009 vf gur lrne bs ersyrpgvba ba gur qrfxgbc
-gur tybor vf bhe cbal, gur pbfzbf bhe erny ubefr
-jub nz V naq vs lrf, ubj znal?
-cebtenzzvat va orq vf n cresrpgyl svar npgvivgl
-zbber'f ynj vf n qeht jvgu gur jbefg pbzr qbja
-EClguba: jr hfr vg fb lbh qba'g unir gb
-Zbber'f ynj vf n qeht jvgu gur jbefg pbzr qbja. EClguba: haqrpvqrq.
-guvatf jvyy or avpr naq fghss
-qba'g cbfg yvaxf gb cngragf urer
-Abg lbhe hfhny nanylfrf.
-Gur Neg bs gur Punaary
-Clguba 300
-V fhccbfr ZO bs UGZY cre frpbaq vf abg gur hfhny fcrrq zrnfher crbcyr jbhyq rkcrpg sbe n wvg
-gur fha arire frgf ba gur ClCl rzcver
-ghegyrf ner snfgre guna lbh guvax
-cebtenzzvat vf na nrfgrguvp raqrnibhe
-P vf tbbq sbe fbzrguvat, whfg abg sbe jevgvat fbsgjner
-trezna vf tbbq sbe fbzrguvat, whfg abg sbe jevgvat fbsgjner
-trezna vf tbbq sbe artngvbaf, whfg abg sbe jevgvat fbsgjner
-# nffreg qvq abg penfu
-lbh fubhyq fgneg n cresrpg fbsgjner zbirzrag
-lbh fubhyq fgneg n cresrpg punaary gbcvp zbirzrag
-guvf vf n cresrpg punaary gbcvp
-guvf vf n frys-ersreragvny punaary gbcvp
-crrcubcr bcgvzvmngvbaf ner jung n Fhssvpvragyl Fzneg Pbzcvyre hfrf
-"crrcubcr" bcgvzvmngvbaf ner jung na bcgvzvfgvp Pbzcvyre hfrf
-pubbfr lbhe unpx
-gur 'fhcre' xrljbeq vf abg gung uhttnoyr
-wlguba cngpurf ner abg rabhtu sbe clcl
-- qb lbh xabj oreyva? - nyy bs vg? - jryy, whfg oreyva
-- ubj jvyy gur snpg gung gurl ner hfrq va bhe ercy punatr bhe gbcvpf?
-- ubj pna vg rire unir jbexrq?
-- jurer fubhyq gur unpx or fgberq?
-- Vg'f uneq gb fnl rknpgyl jung pbafgvghgrf erfrnepu va gur pbzchgre jbeyq, ohg nf n svefg nccebkvzngvba, vg'f fbsgjner gung qbrfa'g unir hfref.
-- Cebtenzzvat vf nyy nobhg xabjvat jura gb obvy gur benatr fcbatr qbaxrl npebff gur cuvyyvcvarf
-- Jul fb znal, znal, znal, znal, znal, znal qhpxyvatf?
-- ab qrgnvy vf bofpher rabhtu gb abg unir fbzr pbqr qrcraqvat ba vg.
-- jung V trarenyyl jnag vf serr fcrrqhcf
-- nyy bs ClCl vf kv-dhnyvgl
-"lbh pna nyjnlf xvyy -9 be bf._rkvg() vs lbh'er va n uheel"
-Ohernhpengf ohvyq npnqrzvp rzcverf juvpu puhea bhg zrnavatyrff fbyhgvbaf gb veeryrinag ceboyrzf.
-vg'f abg n unpx, vg'f n jbexnebhaq
-ClCl qbrfa'g unir pbcbylinevnqvp qrcraqragyl-zbabzbecurq ulcresyhknqf
-ClCl qbrfa'g punatr gur shaqnzragny culfvpf pbafgnagf
-Qnapr bs gur Fhtnecyhz Snvel
-Wnin vf whfg tbbq rabhtu gb or cenpgvpny, ohg abg tbbq rabhtu gb or hfnoyr.
-RhebClguba vf unccravat, qba'g rkcrpg nal dhvpx erfcbafr gvzrf.
-"V jbhyq yvxr gb fgnl njnl sebz ernyvgl gura"
-"gung'f jul gur 'be' vf ernyyl na 'naq' "
-jvgu nyy nccebcevngr pbagrkghnyvfngvbavat
-qba'g gevc ba gur cbjre pbeq
-vzcyrzragvat YBTB va YBTB: "ghegyrf nyy gur jnl qbja"
-gur ohooyrfbeg jbhyq or gur jebat jnl gb tb
-gur cevapvcyr bs pbafreingvba bs zrff
-gb fnir n gerr, rng n ornire
-Qre Ovore znpugf evpugvt: Antg nyyrf xnchgg.
-"Nal jbeyqivrj gung vfag jenpxrq ol frys-qbhog naq pbashfvba bire vgf bja vqragvgl vf abg n jbeyqivrj sbe zr." - Fpbgg Nnebafba
-jr oryvrir va cnapnxrf, znlor
-jr oryvrir va ghegyrf, znlor
-jr qrsvavgryl oryvrir va zrgn
-gur zngevk unf lbh
-"Yvsr vf uneq, gura lbh anc" - n png
-Vf Nezva ubzr jura gur havirefr prnfrf gb rkvfg?
-Qhrffryqbes fcevag fgnegrq
-frys.nobeeg("pnaabg ybnq negvpyrf")
-QRAGVFGEL FLZOBY YVTUG IREGVPNY NAQ JNIR
-"Gur UUH pnzchf vf n tbbq Dhnxr yriry" - Nezva
-"Gur UUH pnzchf jbhyq or n greevoyr dhnxr yriry - lbh'q arire unir n pyhr jurer lbh ner" - zvpunry
-N enqvbnpgvir png unf 18 unys-yvirf.
-<rfp> : j  [fvtu] <onpxfcnpr> <onpxfcnpr> <pgey>-f
-pbybe-pbqrq oyhrf
-"Neebtnapr va pbzchgre fpvrapr vf zrnfherq va anab-Qvwxfgenf."
-ClCl arrqf n Whfg-va-Gvzr WVG
-"Lbh pna'g gvzr geniry whfg ol frggvat lbhe pybpxf jebat"
-Gjb guernqf jnyx vagb n one. Gur onexrrcre ybbxf hc naq lryyf, "url, V jnag qba'g nal pbaqvgvbaf enpr yvxr gvzr ynfg!"
 Clguba 2.k rfg cerfdhr zbeg, ivir Clguba!
 Clguba 2.k vf abg qrnq
 Riregvzr fbzrbar nethrf jvgu "Fznyygnyx unf nyjnlf qbar K", vg vf  nyjnlf n tbbq uvag gung fbzrguvat arrqf gb or punatrq snfg. - Znephf Qraxre
@@ -119,7 +8,6 @@
 __kkk__ naq __ekkk__ if bcrengvba fybgf: cnegvpyr dhnaghz fhcrecbfvgvba xvaq bs sha
 ClCl vf na rkpvgvat grpuabybtl gung yrgf lbh gb jevgr snfg, cbegnoyr, zhygv-cyngsbez vagrecergref jvgu yrff rssbeg
 Nezva: "Cebybt vf n zrff.", PS: "Ab, vg'f irel pbby!", Nezva: "Vfa'g guvf jung V fnvq?"
-<nevtngb> tbbq, grfgf ner hfrshy fbzrgvzrf :-)
 ClCl vf yvxr nofheq gurngre
 jr unir ab nagv-vzcbffvoyr fgvpx gung znxrf fher gung nyy lbhe cebtenzf unyg
 clcl vf n enpr orgjrra crbcyr funivat lnxf naq gur havirefr cebqhpvat zber orneqrq lnxf. Fb sne, gur havirefr vf jvaavat
@@ -136,14 +24,14 @@
 ClCl 1.1.0orgn eryrnfrq: uggc://pbqrfcrnx.arg/clcl/qvfg/clcl/qbp/eryrnfr-1.1.0.ugzy
 "gurer fubhyq or bar naq bayl bar boivbhf jnl gb qb vg". ClCl inevnag: "gurer pna or A unys-ohttl jnlf gb qb vg"
 1.1 svany eryrnfrq: uggc://pbqrfcrnx.arg/clcl/qvfg/clcl/qbp/eryrnfr-1.1.0.ugzy
-1.1 svany eryrnfrq | <svwny> nzq64 naq ccp ner bayl ninvynoyr va ragrecevfr irefvba
+<svwny> nzq64 naq ccp ner bayl ninvynoyr va ragrecevfr irefvba
 Vf gurer n clcl gvzr? - vs lbh pna srry vg (?) gura gurer vf
 <nevtngb> ab, abezny jbex vf fhpu zhpu yrff gvevat guna inpngvbaf
 <nevtngb> ab, abezny jbex vf fb zhpu yrff gvevat guna inpngvbaf
-SVEFG gurl vtaber lbh, gura gurl ynhtu ng lbh, gura gurl svtug lbh, gura lbh jva.
+-SVEFG gurl vtaber lbh, gura gurl ynhtu ng lbh, gura gurl svtug lbh, gura lbh jva.-
 vg'f Fhaqnl, znlor
 vg'f Fhaqnl, ntnva
-"3 + 3 = 8"  Nagb va gur WVG gnyx
+"3 + 3 = 8" - Nagb va gur WVG gnyx
 RPBBC vf unccravat
 RPBBC vf svavfurq
 cflpb rngf bar oenva cre vapu bs cebterff
@@ -175,10 +63,108 @@
 "nu, whfg va gvzr qbphzragngvba" (__nc__)
 ClCl vf abg n erny IZ: ab frtsnhyg unaqyref gb qb gur ener pnfrf
 lbh pna'g unir obgu pbairavrapr naq fcrrq
-gur WVG qbrfa'g jbex ba BF/K (abi'09)
-ab fhccbeg sbe BF/K evtug abj! (abi'09)
 fyvccref urvtug pna or zrnfherq va k86 ertvfgref
 clcl vf n enpr orgjrra gur vaqhfgel gelvat gb ohvyq znpuvarf jvgu zber naq zber erfbheprf, naq gur clcl qrirybcref gelvat gb rng nyy bs gurz. Fb sne, gur jvaare vf fgvyy hapyrne
+"znl pbagnva ahgf naq/be lbhat cbvagref"
+vg'f nyy irel fvzcyr, yvxr gur ubyvqnlf
+unccl ClCl'f lrne 2010!
+fnzhryr fnlf gung jr ybfg n enmbe. fb jr pna'g funir lnxf
+"yrg'f abg or bofpher, hayrff jr ernyyl arrq gb"
+<nevtngb> (abg guernq-fnsr, ohg jryy, abguvat vf)
+clcl unf znal ceboyrzf, ohg rnpu bar unf znal fbyhgvbaf
+whfg nabgure vgrz (1.333...) ba bhe erny-ahzorerq gbqb yvfg
+ClCl vf Fuveg Bevtnzv erfrnepu
+<svwny> nafjrevat n dhrfgvba: "ab -- sbe ng yrnfg bar cbffvoyr vagrecergngvba bs lbhe fragrapr"
+eryrnfr 1.2 hcpbzvat
+ClCl 1.2 eryrnfrq - uggc://clcl.bet/
+AB IPF QVFPHFFVBAF
+EClguba vf n svar pnzry unve oehfu
+ClCl vf n npghnyyl n ivfhnyvmngvba cebwrpg, jr whfg ohvyq vagrecergref gb unir vagrerfgvat qngn gb ivfhnyvmr
+clcl vf yvxr fnhfntrf
+naq abj sbe fbzrguvat pbzcyrgryl qvssrerag
+n 10gu bs sberire vf 1u45
+pbeerpg pbqr qbrfag arrq nal grfgf <qhpx>
+cbfgfgehpghenyvfz rgp.
+clcl UVG trarengbe
+gur arj clcl fcbeg vf gb cnff clcl ohtf nf pclguba ohtf
+jr unir zhpu zber vagrecergref guna hfref
+ClCl 1.3 njnvgvat eryrnfr
+ClCl 1.3 eryrnfrq
+vg frrzf gb zr gung bapr lbh frggyr ba na rkrphgvba / bowrpg zbqry naq / be olgrpbqr sbezng, lbh'ir nyernql qrpvqrq jung ynathntrf (jurer gur 'f' frrzf fhcresyhbhf) fhccbeg vf tbvat gb or svefg pynff sbe
+"Nyy ceboyrzf va ClCl pna or fbyirq ol nabgure yriry bs vagrecergngvba"
+ClCl 1.3 eryrnfrq (jvaqbjf ovanevrf vapyhqrq)
+jul qvq lbh thlf unir gb znxr gur ohvygva sbeghar zber vagrerfgvat guna npghny jbex? v whfg pngpurq zlfrys erfgnegvat clcl 20 gvzrf
+"jr hfrq gb unir n zrff jvgu na bofpher vagresnpr, abj jr unir zrff urer naq bofpher vagresnpr gurer. cebterff" crqebavf ba n clcl fcevag
+"phcf bs pbssrr ner yvxr nanybtvrf va gung V'z znxvat bar evtug abj"
+"vg'f nyjnlf hc gb hf, va n jnl be gur bgure"
+ClCl vf infg, naq pbagnvaf zhygvghqrf
+qravny vf eneryl n tbbq qrohttvat grpuavdhr
+"Yrg'f tb." - "Jr pna'g" - "Jul abg?" - "Jr'er jnvgvat sbe n Genafyngvba." - (qrfcnvevatyl) "Nu!"
+'gung'f qrsvavgryl n pnfr bs "hu????"'
+va gurbel gurer vf gur Ybbc, va cenpgvpr gurer ner oevqtrf
+gur uneqqevir - pbafgnag qngn cvytevzntr
+ClCl vf n gbby gb xrrc bgurejvfr qnatrebhf zvaqf fnsryl bpphcvrq.
+jr ner n trareny senzrjbex ohvyg ba pbafvfgrag nccyvpngvba bs nqubp-arff
+gur jnl gb nibvq n jbexnebhaq vf gb vagebqhpr n fgebatre jbexnebhaq fbzrjurer ryfr
+pnyyvat gur genafyngvba gbby punva n 'fpevcg' vf xvaq bs bssrafvir
+ehaavat clcl-p genafyngr.cl vf n ovg yvxr jngpuvat n guevyyre zbivr, vg pbhyq pbafhzr nyy gur zrzbel ng nal gvzr
+ehaavat clcl-p genafyngr.cl vf n ovg yvxr jngpuvat n guevyyre zbivr, vg pbhyq qvr ng nal gvzr orpnhfr bs gur 32-ovg 4TO yvzvg bs ENZ
+Qh jvefg rora tranh qnf reervpura, jbena xrvare tynhog
+vs fjvgmreynaq jrer jurer terrpr vf (ba vfynaqf) jbhyq gurl nyy or pbaarpgrq ol oevqtrf?
+genafyngvat clcl jvgu pclguba vf fbbbbbb fybj
+ClCl 1.4 eryrnfrq!
+Jr ner abg urebrf, whfg irel cngvrag.
+QBAR zrnaf vg'f qbar
+jul gurer vf ab "ClCl 1.4 eryrnfrq" va gbcvp nal zber?
+fabj! fabj!
+svanyyl, zrephevny zvtengvba vf unccravat!
+Gur zvtengvba gb zrephevny vf pbzcyrgrq! uggc://ovgohpxrg.bet/clcl/clcl
+fabj! fabj! (gre)
+unccl arj lrne
+naq anaanaw gb lbh nf jryy
+Frrvat nf gur ynjf bs culfvpf ner ntnvafg lbh, lbh unir gb pnershyyl pbafvqre lbhe fpbcr fb gung lbhe tbnyf ner ernfbanoyr.
+nf hfhny va clcl, gur fbyhgvba nccrnef pbzcyrgryl qvfcebcbegvbangr gb gur ceboyrz naq vafgrnq jr'yy tb sbe n pbzcyrgryl qvssrerag fvzcyre nccebnpu gb gur bevtvany ceboyrz
+fabj, fabj!
+va clcl lbh ner nyjnlf ng gur jebat yriry, va bar jnl be gur bgure
+jryy, vg'f jebat ohg abg fb "irel jebat" nf vg ybbxrq
+<svwny> V ybir clcl
+ynmvarff vzcngvrapr naq uhoevf
+fabj, fabj
+EClguba: guvatf lbh jbhyqa'g qb va Clguba, naq pna'g qb va P.
+vg vf gur rkcrpgrq orunivbe, rkprcg jura lbh qba'g rkcrpg vg
+erqrsvavat lryybj frrzf yvxr n orggre vqrn
+"gung'f ubjrire whfg ratvarrevat" (svwny)
+"[vg] whfg fubjf ntnva gung eclguba vf bofpher" (psobym)
+"naljnl, clguba vf n infg ynathntr" (svwny)
+bhg-bs-yvr-thneqf
+"gurer ner qnlf ba juvpu lbh ybbx nebhaq naq abguvat fubhyq unir rire jbexrq" (svwny)
+clcl vf n orggre xvaq bs sbbyvfuarff - ynp
+ehaavat grfgf vf rffragvny sbe qrirybcvat clcl -- hu? qvq V oernx gur grfg? (svwny)
+V'ir tbg guvf sybbe jnk gung'f nyfb n TERNG qrffreg gbccvat!!
+rknexha: "gur cneg gung V gubhtug jnf tbvat gb or uneq jnf gevivny, fb abj V whfg unir guvf cneg gung V qvqa'g rira guvax bs gung vf uneq"
+V fhccbfr jr pna yvir jvgu gur bofphevgl, nf ybat nf gurer vf n pbzzrag znxvat vg yvtugre
+V nz n ovt oryvrire va ernfbaf. ohg gur nccnerag xvaq ner zl snibevgr.
+clcl: trg n WVG sbe serr (jryy gur svefg qnl lbh jba'g znantr naq vg jvyy or irel sehfgengvat)
+<Nyrk_Tnlabe> thgjbegu: bu, jr fubhyq znxr gur WVG zntvpnyyl orggre, jvgu qrpbengbef naq fghss
+vg'f n pbzcyrgr unpx, ohg n irel zvavzny bar (nevtngb)
+svefg gurl ynhtu ng lbh, gura gurl vtaber lbh, gura gurl svtug lbh, gura lbh jva
+ClCl vf snzvyl sevraqyl
+jr yvxr pbzcynvagf
+gbqnl jr'er snfgre guna lrfgreqnl (hfhnyyl)
+ClCl naq PClguba: gurl ner zbegny rarzvrf vagrag ba xvyyvat rnpu bgure
+nethnoyl, rirelguvat vf n avpur
+clcl unf ynlref yvxr bavbaf: crryvat gurz onpx jvyy znxr lbh pel
+EClguba zntvpnyyl znxrf lbh evpu naq snzbhf (fnlf fb ba gur gva)
+Vf evtbobg nebhaq jura gur havirefr prnfrf gb rkvfg?
+ClCl vf gbb pbby sbe dhrelfgevatf.
+< nevtngb> gura jung bpphef? < svwny> tbbq fghss V oryvrir
+ClCl 1.6 eryrnfrq!
+<psobym> jurer ner gur grfgf?
+uggc://gjvgcvp.pbz/52nr8s
+N enaqbz dhbgr
+Nyy rkprcgoybpxf frrz fnar.
+N cvax tyvggrel ebgngvat ynzoqn
+"vg'f yvxryl grzcbenel hagvy sberire" nevtb
 """
 
 def some_topic():
diff --git a/lib_pypy/pyrepl/readline.py b/lib_pypy/pyrepl/readline.py
--- a/lib_pypy/pyrepl/readline.py
+++ b/lib_pypy/pyrepl/readline.py
@@ -231,7 +231,11 @@
         return ''.join(chars)
 
     def _histline(self, line):
-        return unicode(line.rstrip('\n'), ENCODING)
+        line = line.rstrip('\n')
+        try:
+            return unicode(line, ENCODING)
+        except UnicodeDecodeError:   # bah, silently fall back...
+            return unicode(line, 'utf-8')
 
     def get_history_length(self):
         return self.saved_history_length
@@ -268,7 +272,10 @@
         f = open(os.path.expanduser(filename), 'w')
         for entry in history:
             if isinstance(entry, unicode):
-                entry = entry.encode(ENCODING)
+                try:
+                    entry = entry.encode(ENCODING)
+                except UnicodeEncodeError:   # bah, silently fall back...
+                    entry = entry.encode('utf-8')
             entry = entry.replace('\n', '\r\n')   # multiline history support
             f.write(entry + '\n')
         f.close()
@@ -395,9 +402,21 @@
     _wrapper.f_in = f_in
     _wrapper.f_out = f_out
 
-    if hasattr(sys, '__raw_input__'):    # PyPy
-        _old_raw_input = sys.__raw_input__
+    if '__pypy__' in sys.builtin_module_names:    # PyPy
+
+        def _old_raw_input(prompt=''):
+            # sys.__raw_input__() is only called when stdin and stdout are
+            # as expected and are ttys.  If it is the case, then get_reader()
+            # should not really fail in _wrapper.raw_input().  If it still
+            # does, then we will just cancel the redirection and call again
+            # the built-in raw_input().
+            try:
+                del sys.__raw_input__
+            except AttributeError:
+                pass
+            return raw_input(prompt)
         sys.__raw_input__ = _wrapper.raw_input
+
     else:
         # this is not really what readline.c does.  Better than nothing I guess
         import __builtin__
diff --git a/lib_pypy/resource.py b/lib_pypy/resource.py
--- a/lib_pypy/resource.py
+++ b/lib_pypy/resource.py
@@ -7,7 +7,7 @@
 
 from ctypes_support import standard_c_lib as libc
 from ctypes_support import get_errno
-from ctypes import Structure, c_int, c_long, byref, sizeof, POINTER
+from ctypes import Structure, c_int, c_long, byref, POINTER
 from errno import EINVAL, EPERM
 import _structseq
 
@@ -165,7 +165,6 @@
 
 @builtinify
 def getpagesize():
-    pagesize = 0
     if _getpagesize:
         return _getpagesize()
     else:
diff --git a/pypy/annotation/classdef.py b/pypy/annotation/classdef.py
--- a/pypy/annotation/classdef.py
+++ b/pypy/annotation/classdef.py
@@ -276,8 +276,8 @@
         # create the Attribute and do the generalization asked for
         newattr = Attribute(attr, self.bookkeeper)
         if s_value:
-            if newattr.name == 'intval' and getattr(s_value, 'unsigned', False):
-                import pdb; pdb.set_trace()
+            #if newattr.name == 'intval' and getattr(s_value, 'unsigned', False):
+            #    import pdb; pdb.set_trace()
             newattr.s_value = s_value
 
         # keep all subattributes' values
diff --git a/pypy/config/pypyoption.py b/pypy/config/pypyoption.py
--- a/pypy/config/pypyoption.py
+++ b/pypy/config/pypyoption.py
@@ -72,6 +72,7 @@
     del working_modules['fcntl']  # LOCK_NB not defined
     del working_modules["_minimal_curses"]
     del working_modules["termios"]
+    del working_modules["_multiprocessing"]   # depends on rctime
 
 
 
@@ -91,7 +92,7 @@
 
 module_import_dependencies = {
     # no _rawffi if importing pypy.rlib.clibffi raises ImportError
-    # or CompilationError
+    # or CompilationError or py.test.skip.Exception
     "_rawffi"   : ["pypy.rlib.clibffi"],
     "_ffi"      : ["pypy.rlib.clibffi"],
 
@@ -112,7 +113,7 @@
             try:
                 for name in modlist:
                     __import__(name)
-            except (ImportError, CompilationError), e:
+            except (ImportError, CompilationError, py.test.skip.Exception), e:
                 errcls = e.__class__.__name__
                 config.add_warning(
                     "The module %r is disabled\n" % (modname,) +
@@ -127,7 +128,7 @@
 
 pypy_optiondescription = OptionDescription("objspace", "Object Space Options", [
     ChoiceOption("name", "Object Space name",
-                 ["std", "flow", "thunk", "dump", "taint"],
+                 ["std", "flow", "thunk", "dump"],
                  "std",
                  cmdline='--objspace -o'),
 
diff --git a/pypy/doc/__pypy__-module.rst b/pypy/doc/__pypy__-module.rst
--- a/pypy/doc/__pypy__-module.rst
+++ b/pypy/doc/__pypy__-module.rst
@@ -37,29 +37,6 @@
 .. _`thunk object space docs`: objspace-proxies.html#thunk
 .. _`interface section of the thunk object space docs`: objspace-proxies.html#thunk-interface
 
-.. broken:
-
-    Taint Object Space Functionality
-    ================================
-
-    When the taint object space is used (choose with :config:`objspace.name`),
-    the following names are put into ``__pypy__``:
-
-     - ``taint``
-     - ``is_tainted``
-     - ``untaint``
-     - ``taint_atomic``
-     - ``_taint_debug``
-     - ``_taint_look``
-     - ``TaintError``
-
-    Those are all described in the `interface section of the taint object space
-    docs`_.
-
-    For more detailed explanations and examples see the `taint object space docs`_.
-
-    .. _`taint object space docs`: objspace-proxies.html#taint
-    .. _`interface section of the taint object space docs`: objspace-proxies.html#taint-interface
 
 Transparent Proxy Functionality
 ===============================
diff --git a/pypy/doc/config/objspace.name.txt b/pypy/doc/config/objspace.name.txt
--- a/pypy/doc/config/objspace.name.txt
+++ b/pypy/doc/config/objspace.name.txt
@@ -4,7 +4,6 @@
 for normal usage):
 
   * thunk_: The thunk object space adds lazy evaluation to PyPy.
-  * taint_: The taint object space adds soft security features.
   * dump_:  Using this object spaces results in the dumpimp of all operations
     to a log.
 
@@ -12,5 +11,4 @@
 .. _`Object Space Proxies`: ../objspace-proxies.html
 .. _`Standard Object Space`: ../objspace.html#standard-object-space
 .. _thunk: ../objspace-proxies.html#thunk
-.. _taint: ../objspace-proxies.html#taint
 .. _dump: ../objspace-proxies.html#dump
diff --git a/pypy/doc/index.rst b/pypy/doc/index.rst
--- a/pypy/doc/index.rst
+++ b/pypy/doc/index.rst
@@ -309,7 +309,6 @@
 .. _`object space`: objspace.html
 .. _FlowObjSpace: objspace.html#the-flow-object-space 
 .. _`trace object space`: objspace.html#the-trace-object-space 
-.. _`taint object space`: objspace-proxies.html#taint
 .. _`thunk object space`: objspace-proxies.html#thunk
 .. _`transparent proxies`: objspace-proxies.html#tproxy
 .. _`Differences between PyPy and CPython`: cpython_differences.html
diff --git a/pypy/doc/objspace-proxies.rst b/pypy/doc/objspace-proxies.rst
--- a/pypy/doc/objspace-proxies.rst
+++ b/pypy/doc/objspace-proxies.rst
@@ -129,297 +129,6 @@
    function behaves lazily: all calls to it return a thunk object.
 
 
-.. broken right now:
-
-    .. _taint:
-
-    The Taint Object Space
-    ======================
-
-    Motivation
-    ----------
-
-    The Taint Object Space provides a form of security: "tainted objects",
-    inspired by various sources, see [D12.1]_ for a more detailed discussion. 
-
-    The basic idea of this kind of security is not to protect against
-    malicious code but to help with handling and boxing sensitive data. 
-    It covers two kinds of sensitive data: secret data which should not leak, 
-    and untrusted data coming from an external source and that must be 
-    validated before it is used.
-
-    The idea is that, considering a large application that handles these
-    kinds of sensitive data, there are typically only a small number of
-    places that need to explicitly manipulate that sensitive data; all the
-    other places merely pass it around, or do entirely unrelated things.
-
-    Nevertheless, if a large application needs to be reviewed for security,
-    it must be entirely carefully checked, because it is possible that a
-    bug at some apparently unrelated place could lead to a leak of sensitive
-    information in a way that an external attacker could exploit.  For
-    example, if any part of the application provides web services, an
-    attacker might be able to issue unexpected requests with a regular web
-    browser and deduce secret information from the details of the answers he
-    gets.  Another example is the common CGI attack where an attacker sends
-    malformed inputs and causes the CGI script to do unintended things.
-
-    An approach like that of the Taint Object Space allows the small parts
-    of the program that manipulate sensitive data to be explicitly marked.
-    The effect of this is that although these small parts still need a
-    careful security review, the rest of the application no longer does,
-    because even a bug would be unable to leak the information.
-
-    We have implemented a simple two-level model: objects are either
-    regular (untainted), or sensitive (tainted).  Objects are marked as
-    sensitive if they are secret or untrusted, and only declassified at
-    carefully-checked positions (e.g. where the secret data is needed, or
-    after the untrusted data has been fully validated).
-
-    It would be simple to extend the code for more fine-grained scales of
-    secrecy.  For example it is typical in the literature to consider
-    user-specified lattices of secrecy levels, corresponding to multiple
-    "owners" that cannot access data belonging to another "owner" unless
-    explicitly authorized to do so.
-
-    Tainting and untainting
-    -----------------------
-
-    Start a py.py with the Taint Object Space and try the following example::
-
-        $ py.py -o taint
-        >>>> from __pypy__ import taint
-        >>>> x = taint(6)
-
-        # x is hidden from now on.  We can pass it around and
-        # even operate on it, but not inspect it.  Taintness
-        # is propagated to operation results.
-
-        >>>> x
-        TaintError
-
-        >>>> if x > 5: y = 2   # see below
-        TaintError
-
-        >>>> y = x + 5         # ok
-        >>>> lst = [x, y]
-        >>>> z = lst.pop()
-        >>>> t = type(z)       # type() works too, tainted answer
-        >>>> t
-        TaintError
-        >>>> u = t is int      # even 'is' works
-        >>>> u
-        TaintError
-
-    Notice that using a tainted boolean like ``x > 5`` in an ``if``
-    statement is forbidden.  This is because knowing which path is followed
-    would give away a hint about ``x``; in the example above, if the
-    statement ``if x > 5: y = 2`` was allowed to run, we would know
-    something about the value of ``x`` by looking at the (untainted) value
-    in the variable ``y``.
-
-    Of course, there is a way to inspect tainted objects.  The basic way is
-    to explicitly "declassify" it with the ``untaint()`` function.  In an
-    application, the places that use ``untaint()`` are the places that need
-    careful security review.  To avoid unexpected objects showing up, the
-    ``untaint()`` function must be called with the exact type of the object
-    to declassify.  It will raise ``TaintError`` if the type doesn't match::
-
-        >>>> from __pypy__ import taint
-        >>>> untaint(int, x)
-        6
-        >>>> untaint(int, z)
-        11
-        >>>> untaint(bool, x > 5)
-        True
-        >>>> untaint(int, x > 5)
-        TaintError
-
-
-    Taint Bombs
-    -----------
-
-    In this area, a common problem is what to do about failing operations.
-    If an operation raises an exception when manipulating a tainted object,
-    then the very presence of the exception can leak information about the
-    tainted object itself.  Consider::
-
-        >>>> 5 / (x-6)
-
-    By checking if this raises ``ZeroDivisionError`` or not, we would know
-    if ``x`` was equal to 6 or not.  The solution to this problem in the
-    Taint Object Space is to introduce *Taint Bombs*.  They are a kind of
-    tainted object that doesn't contain a real object, but a pending
-    exception.  Taint Bombs are indistinguishable from normal tainted
-    objects to unprivileged code. See::
-
-        >>>> x = taint(6)
-        >>>> i = 5 / (x-6)     # no exception here
-        >>>> j = i + 1         # nor here
-        >>>> k = j + 5         # nor here
-        >>>> untaint(int, k)
-        TaintError
-
-    In the above example, all of ``i``, ``j`` and ``k`` contain a Taint
-    Bomb.  Trying to untaint it raises an exception - a generic
-    ``TaintError``.  What we win is that the exception gives little away,
-    and most importantly it occurs at the point where ``untaint()`` is
-    called, not where the operation failed.  This means that all calls to
-    ``untaint()`` - but not the rest of the code - must be carefully
-    reviewed for what occurs if they receive a Taint Bomb; they might catch
-    the ``TaintError`` and give the user a generic message that something
-    went wrong, if we are reasonably careful that the message or even its
-    presence doesn't give information away.  This might be a
-    problem by itself, but there is no satisfying general solution here:
-    it must be considered on a case-by-case basis.  Again, what the
-    Taint Object Space approach achieves is not solving these problems, but
-    localizing them to well-defined small parts of the application - namely,
-    around calls to ``untaint()``.
-
-    The ``TaintError`` exception deliberately does not include any
-    useful error messages, because they might give information away.
-    Of course, this makes debugging quite a bit harder; a difficult
-    problem to solve properly.  So far we have implemented a way to peek in a Taint
-    Box or Bomb, ``__pypy__._taint_look(x)``, and a "debug mode" that
-    prints the exception as soon as a Bomb is created - both write
-    information to the low-level stderr of the application, where we hope
-    that it is unlikely to be seen by anyone but the application
-    developer.
-
-
-    Taint Atomic functions
-    ----------------------
-
-    Occasionally, a more complicated computation must be performed on a
-    tainted object.  This requires first untainting the object, performing the
-    computations, and then carefully tainting the result again (including
-    hiding all exceptions into Bombs).
-
-    There is a built-in decorator that does this for you::
-
-        >>>> @__pypy__.taint_atomic
-        >>>> def myop(x, y):
-        ....     while x > 0:
-        ....         x -= y
-        ....     return x
-        ....
-        >>>> myop(42, 10)
-        -8
-        >>>> z = myop(taint(42), 10)
-        >>>> z
-        TaintError
-        >>>> untaint(int, z)
-        -8
-
-    The decorator makes a whole function behave like a built-in operation.
-    If no tainted argument is passed in, the function behaves normally.  But
-    if any of the arguments is tainted, it is automatically untainted - so
-    the function body always sees untainted arguments - and the eventual
-    result is tainted again (possibly in a Taint Bomb).
-
-    It is important for the function marked as ``taint_atomic`` to have no
-    visible side effects, as these could cause information leakage.
-    This is currently not enforced, which means that all ``taint_atomic``
-    functions have to be carefully reviewed for security (but not the
-    callers of ``taint_atomic`` functions).
-
-    A possible future extension would be to forbid side-effects on
-    non-tainted objects from all ``taint_atomic`` functions.
-
-    An example of usage: given a tainted object ``passwords_db`` that
-    references a database of passwords, we can write a function
-    that checks if a password is valid as follows::
-
-        @taint_atomic
-        def validate(passwords_db, username, password):
-            assert type(passwords_db) is PasswordDatabase
-            assert type(username) is str
-            assert type(password) is str
-            ...load username entry from passwords_db...
-            return expected_password == password
-
-    It returns a tainted boolean answer, or a Taint Bomb if something
-    went wrong.  A caller can do::
-
-        ok = validate(passwords_db, 'john', '1234')
-        ok = untaint(bool, ok)
-
-    This can give three outcomes: ``True``, ``False``, or a ``TaintError``
-    exception (with no information on it) if anything went wrong.  If even
-    this is considered giving too much information away, the ``False`` case
-    can be made indistinguishable from the ``TaintError`` case (simply by
-    raising an exception in ``validate()`` if the password is wrong).
-
-    In the above example, the security results achieved are the following:
-    as long as ``validate()`` does not leak information, no other part of
-    the code can obtain more information about a passwords database than a
-    Yes/No answer to a precise query.
-
-    A possible extension of the ``taint_atomic`` decorator would be to check
-    the argument types, as ``untaint()`` does, for the same reason: to
-    prevent bugs where a function like ``validate()`` above is accidentally
-    called with the wrong kind of tainted object, which would make it
-    misbehave.  For now, all ``taint_atomic`` functions should be
-    conservative and carefully check all assumptions on their input
-    arguments.
-
-
-    .. _`taint-interface`:
-
-    Interface
-    ---------
-
-    .. _`like a built-in operation`:
-
-    The basic rule of the Tainted Object Space is that it introduces two new
-    kinds of objects, Tainted Boxes and Tainted Bombs (which are not types
-    in the Python sense).  Each box internally contains a regular object;
-    each bomb internally contains an exception object.  An operation
-    involving Tainted Boxes is performed on the objects contained in the
-    boxes, and gives a Tainted Box or a Tainted Bomb as a result (such an
-    operation does not let an exception be raised).  An operation called
-    with a Tainted Bomb argument immediately returns the same Tainted Bomb.
-
-    In a PyPy running with (or translated with) the Taint Object Space,
-    the ``__pypy__`` module exposes the following interface:
-
-    * ``taint(obj)``
-
-        Return a new Tainted Box wrapping ``obj``.  Return ``obj`` itself
-        if it is already tainted (a Box or a Bomb).
-
-    * ``is_tainted(obj)``
-
-        Check if ``obj`` is tainted (a Box or a Bomb).
-
-    * ``untaint(type, obj)``
-
-        Untaints ``obj`` if it is tainted.  Raise ``TaintError`` if the type
-        of the untainted object is not exactly ``type``, or if ``obj`` is a
-        Bomb.
-
-    * ``taint_atomic(func)``
-
-        Return a wrapper function around the callable ``func``.  The wrapper
-        behaves `like a built-in operation`_ with respect to untainting the
-        arguments, tainting the result, and returning a Bomb.
-
-    * ``TaintError``
-
-        Exception.  On purpose, it provides no attribute or error message.
-
-    * ``_taint_debug(level)``
-
-        Set the debugging level to ``level`` (0=off).  At level 1 or above,
-        all Taint Bombs print a diagnostic message to stderr when they are
-        created.
-
-    * ``_taint_look(obj)``
-
-        For debugging purposes: prints (to stderr) the type and address of
-        the object in a Tainted Box, or prints the exception if ``obj`` is
-        a Taint Bomb.
-
-
 .. _dump:
 
 The Dump Object Space
diff --git a/pypy/doc/project-ideas.rst b/pypy/doc/project-ideas.rst
--- a/pypy/doc/project-ideas.rst
+++ b/pypy/doc/project-ideas.rst
@@ -17,6 +17,12 @@
 projects, or anything else in PyPy, pop up on IRC or write to us on the
 `mailing list`_.
 
+Make big integers faster
+-------------------------
+
+PyPy's implementation of the Python ``long`` type is slower than CPython's.
+Find out why and optimize them.
+
 Numpy improvements
 ------------------
 
diff --git a/pypy/interpreter/astcompiler/ast.py b/pypy/interpreter/astcompiler/ast.py
--- a/pypy/interpreter/astcompiler/ast.py
+++ b/pypy/interpreter/astcompiler/ast.py
@@ -2,7 +2,7 @@
 from pypy.interpreter.baseobjspace import Wrappable
 from pypy.interpreter import typedef
 from pypy.interpreter.gateway import interp2app
-from pypy.interpreter.error import OperationError
+from pypy.interpreter.error import OperationError, operationerrfmt
 from pypy.rlib.unroll import unrolling_iterable
 from pypy.tool.pairtype import extendabletype
 from pypy.tool.sourcetools import func_with_new_name
@@ -2925,14 +2925,13 @@
 def Module_get_body(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -2968,14 +2967,13 @@
 def Interactive_get_body(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3015,8 +3013,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     return space.wrap(w_self.body)
 
 def Expression_set_body(space, w_self, w_new_value):
@@ -3057,14 +3054,13 @@
 def Suite_get_body(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3104,8 +3100,7 @@
             return w_obj
     if not w_self.initialization_state & w_self._lineno_mask:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'lineno'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'lineno')
     return space.wrap(w_self.lineno)
 
 def stmt_set_lineno(space, w_self, w_new_value):
@@ -3126,8 +3121,7 @@
             return w_obj
     if not w_self.initialization_state & w_self._col_offset_mask:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'col_offset'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'col_offset')
     return space.wrap(w_self.col_offset)
 
 def stmt_set_col_offset(space, w_self, w_new_value):
@@ -3157,8 +3151,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'name'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'name')
     return space.wrap(w_self.name)
 
 def FunctionDef_set_name(space, w_self, w_new_value):
@@ -3179,8 +3172,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'args'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'args')
     return space.wrap(w_self.args)
 
 def FunctionDef_set_args(space, w_self, w_new_value):
@@ -3197,14 +3189,13 @@
 def FunctionDef_get_body(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3215,14 +3206,13 @@
 def FunctionDef_get_decorator_list(space, w_self):
     if not w_self.initialization_state & 8:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'decorator_list'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'decorator_list')
     if w_self.w_decorator_list is None:
         if w_self.decorator_list is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.decorator_list]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_decorator_list = w_list
     return w_self.w_decorator_list
 
@@ -3266,8 +3256,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'name'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'name')
     return space.wrap(w_self.name)
 
 def ClassDef_set_name(space, w_self, w_new_value):
@@ -3284,14 +3273,13 @@
 def ClassDef_get_bases(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'bases'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'bases')
     if w_self.w_bases is None:
         if w_self.bases is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.bases]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_bases = w_list
     return w_self.w_bases
 
@@ -3302,14 +3290,13 @@
 def ClassDef_get_body(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3320,14 +3307,13 @@
 def ClassDef_get_decorator_list(space, w_self):
     if not w_self.initialization_state & 8:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'decorator_list'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'decorator_list')
     if w_self.w_decorator_list is None:
         if w_self.decorator_list is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.decorator_list]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_decorator_list = w_list
     return w_self.w_decorator_list
 
@@ -3372,8 +3358,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Return_set_value(space, w_self, w_new_value):
@@ -3414,14 +3399,13 @@
 def Delete_get_targets(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'targets'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'targets')
     if w_self.w_targets is None:
         if w_self.targets is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.targets]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_targets = w_list
     return w_self.w_targets
 
@@ -3457,14 +3441,13 @@
 def Assign_get_targets(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'targets'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'targets')
     if w_self.w_targets is None:
         if w_self.targets is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.targets]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_targets = w_list
     return w_self.w_targets
 
@@ -3479,8 +3462,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Assign_set_value(space, w_self, w_new_value):
@@ -3527,8 +3509,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'target'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'target')
     return space.wrap(w_self.target)
 
 def AugAssign_set_target(space, w_self, w_new_value):
@@ -3549,8 +3530,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'op'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'op')
     return operator_to_class[w_self.op - 1]()
 
 def AugAssign_set_op(space, w_self, w_new_value):
@@ -3573,8 +3553,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def AugAssign_set_value(space, w_self, w_new_value):
@@ -3621,8 +3600,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'dest'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'dest')
     return space.wrap(w_self.dest)
 
 def Print_set_dest(space, w_self, w_new_value):
@@ -3639,14 +3617,13 @@
 def Print_get_values(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'values'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'values')
     if w_self.w_values is None:
         if w_self.values is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.values]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_values = w_list
     return w_self.w_values
 
@@ -3661,8 +3638,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'nl'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'nl')
     return space.wrap(w_self.nl)
 
 def Print_set_nl(space, w_self, w_new_value):
@@ -3710,8 +3686,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'target'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'target')
     return space.wrap(w_self.target)
 
 def For_set_target(space, w_self, w_new_value):
@@ -3732,8 +3707,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'iter'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'iter')
     return space.wrap(w_self.iter)
 
 def For_set_iter(space, w_self, w_new_value):
@@ -3750,14 +3724,13 @@
 def For_get_body(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3768,14 +3741,13 @@
 def For_get_orelse(space, w_self):
     if not w_self.initialization_state & 8:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'orelse'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'orelse')
     if w_self.w_orelse is None:
         if w_self.orelse is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.orelse]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_orelse = w_list
     return w_self.w_orelse
 
@@ -3819,8 +3791,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'test'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'test')
     return space.wrap(w_self.test)
 
 def While_set_test(space, w_self, w_new_value):
@@ -3837,14 +3808,13 @@
 def While_get_body(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3855,14 +3825,13 @@
 def While_get_orelse(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'orelse'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'orelse')
     if w_self.w_orelse is None:
         if w_self.orelse is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.orelse]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_orelse = w_list
     return w_self.w_orelse
 
@@ -3905,8 +3874,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'test'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'test')
     return space.wrap(w_self.test)
 
 def If_set_test(space, w_self, w_new_value):
@@ -3923,14 +3891,13 @@
 def If_get_body(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -3941,14 +3908,13 @@
 def If_get_orelse(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'orelse'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'orelse')
     if w_self.w_orelse is None:
         if w_self.orelse is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.orelse]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_orelse = w_list
     return w_self.w_orelse
 
@@ -3991,8 +3957,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'context_expr'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'context_expr')
     return space.wrap(w_self.context_expr)
 
 def With_set_context_expr(space, w_self, w_new_value):
@@ -4013,8 +3978,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'optional_vars'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'optional_vars')
     return space.wrap(w_self.optional_vars)
 
 def With_set_optional_vars(space, w_self, w_new_value):
@@ -4031,14 +3995,13 @@
 def With_get_body(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -4080,8 +4043,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'type'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'type')
     return space.wrap(w_self.type)
 
 def Raise_set_type(space, w_self, w_new_value):
@@ -4102,8 +4064,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'inst'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'inst')
     return space.wrap(w_self.inst)
 
 def Raise_set_inst(space, w_self, w_new_value):
@@ -4124,8 +4085,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'tback'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'tback')
     return space.wrap(w_self.tback)
 
 def Raise_set_tback(space, w_self, w_new_value):
@@ -4168,14 +4128,13 @@
 def TryExcept_get_body(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -4186,14 +4145,13 @@
 def TryExcept_get_handlers(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'handlers'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'handlers')
     if w_self.w_handlers is None:
         if w_self.handlers is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.handlers]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_handlers = w_list
     return w_self.w_handlers
 
@@ -4204,14 +4162,13 @@
 def TryExcept_get_orelse(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'orelse'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'orelse')
     if w_self.w_orelse is None:
         if w_self.orelse is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.orelse]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_orelse = w_list
     return w_self.w_orelse
 
@@ -4251,14 +4208,13 @@
 def TryFinally_get_body(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -4269,14 +4225,13 @@
 def TryFinally_get_finalbody(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'finalbody'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'finalbody')
     if w_self.w_finalbody is None:
         if w_self.finalbody is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.finalbody]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_finalbody = w_list
     return w_self.w_finalbody
 
@@ -4318,8 +4273,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'test'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'test')
     return space.wrap(w_self.test)
 
 def Assert_set_test(space, w_self, w_new_value):
@@ -4340,8 +4294,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'msg'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'msg')
     return space.wrap(w_self.msg)
 
 def Assert_set_msg(space, w_self, w_new_value):
@@ -4383,14 +4336,13 @@
 def Import_get_names(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'names'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'names')
     if w_self.w_names is None:
         if w_self.names is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.names]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_names = w_list
     return w_self.w_names
 
@@ -4430,8 +4382,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'module'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'module')
     return space.wrap(w_self.module)
 
 def ImportFrom_set_module(space, w_self, w_new_value):
@@ -4451,14 +4402,13 @@
 def ImportFrom_get_names(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'names'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'names')
     if w_self.w_names is None:
         if w_self.names is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.names]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_names = w_list
     return w_self.w_names
 
@@ -4473,8 +4423,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'level'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'level')
     return space.wrap(w_self.level)
 
 def ImportFrom_set_level(space, w_self, w_new_value):
@@ -4522,8 +4471,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     return space.wrap(w_self.body)
 
 def Exec_set_body(space, w_self, w_new_value):
@@ -4544,8 +4492,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'globals'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'globals')
     return space.wrap(w_self.globals)
 
 def Exec_set_globals(space, w_self, w_new_value):
@@ -4566,8 +4513,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'locals'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'locals')
     return space.wrap(w_self.locals)
 
 def Exec_set_locals(space, w_self, w_new_value):
@@ -4610,14 +4556,13 @@
 def Global_get_names(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'names'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'names')
     if w_self.w_names is None:
         if w_self.names is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.names]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_names = w_list
     return w_self.w_names
 
@@ -4657,8 +4602,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Expr_set_value(space, w_self, w_new_value):
@@ -4754,8 +4698,7 @@
             return w_obj
     if not w_self.initialization_state & w_self._lineno_mask:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'lineno'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'lineno')
     return space.wrap(w_self.lineno)
 
 def expr_set_lineno(space, w_self, w_new_value):
@@ -4776,8 +4719,7 @@
             return w_obj
     if not w_self.initialization_state & w_self._col_offset_mask:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'col_offset'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'col_offset')
     return space.wrap(w_self.col_offset)
 
 def expr_set_col_offset(space, w_self, w_new_value):
@@ -4807,8 +4749,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'op'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'op')
     return boolop_to_class[w_self.op - 1]()
 
 def BoolOp_set_op(space, w_self, w_new_value):
@@ -4827,14 +4768,13 @@
 def BoolOp_get_values(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'values'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'values')
     if w_self.w_values is None:
         if w_self.values is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.values]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_values = w_list
     return w_self.w_values
 
@@ -4875,8 +4815,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'left'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'left')
     return space.wrap(w_self.left)
 
 def BinOp_set_left(space, w_self, w_new_value):
@@ -4897,8 +4836,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'op'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'op')
     return operator_to_class[w_self.op - 1]()
 
 def BinOp_set_op(space, w_self, w_new_value):
@@ -4921,8 +4859,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'right'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'right')
     return space.wrap(w_self.right)
 
 def BinOp_set_right(space, w_self, w_new_value):
@@ -4969,8 +4906,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'op'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'op')
     return unaryop_to_class[w_self.op - 1]()
 
 def UnaryOp_set_op(space, w_self, w_new_value):
@@ -4993,8 +4929,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'operand'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'operand')
     return space.wrap(w_self.operand)
 
 def UnaryOp_set_operand(space, w_self, w_new_value):
@@ -5040,8 +4975,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'args'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'args')
     return space.wrap(w_self.args)
 
 def Lambda_set_args(space, w_self, w_new_value):
@@ -5062,8 +4996,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     return space.wrap(w_self.body)
 
 def Lambda_set_body(space, w_self, w_new_value):
@@ -5109,8 +5042,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'test'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'test')
     return space.wrap(w_self.test)
 
 def IfExp_set_test(space, w_self, w_new_value):
@@ -5131,8 +5063,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     return space.wrap(w_self.body)
 
 def IfExp_set_body(space, w_self, w_new_value):
@@ -5153,8 +5084,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'orelse'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'orelse')
     return space.wrap(w_self.orelse)
 
 def IfExp_set_orelse(space, w_self, w_new_value):
@@ -5197,14 +5127,13 @@
 def Dict_get_keys(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'keys'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'keys')
     if w_self.w_keys is None:
         if w_self.keys is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.keys]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_keys = w_list
     return w_self.w_keys
 
@@ -5215,14 +5144,13 @@
 def Dict_get_values(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'values'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'values')
     if w_self.w_values is None:
         if w_self.values is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.values]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_values = w_list
     return w_self.w_values
 
@@ -5260,14 +5188,13 @@
 def Set_get_elts(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'elts'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'elts')
     if w_self.w_elts is None:
         if w_self.elts is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.elts]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_elts = w_list
     return w_self.w_elts
 
@@ -5307,8 +5234,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'elt'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'elt')
     return space.wrap(w_self.elt)
 
 def ListComp_set_elt(space, w_self, w_new_value):
@@ -5325,14 +5251,13 @@
 def ListComp_get_generators(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'generators'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'generators')
     if w_self.w_generators is None:
         if w_self.generators is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.generators]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_generators = w_list
     return w_self.w_generators
 
@@ -5373,8 +5298,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'elt'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'elt')
     return space.wrap(w_self.elt)
 
 def SetComp_set_elt(space, w_self, w_new_value):
@@ -5391,14 +5315,13 @@
 def SetComp_get_generators(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'generators'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'generators')
     if w_self.w_generators is None:
         if w_self.generators is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.generators]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_generators = w_list
     return w_self.w_generators
 
@@ -5439,8 +5362,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'key'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'key')
     return space.wrap(w_self.key)
 
 def DictComp_set_key(space, w_self, w_new_value):
@@ -5461,8 +5383,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def DictComp_set_value(space, w_self, w_new_value):
@@ -5479,14 +5400,13 @@
 def DictComp_get_generators(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'generators'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'generators')
     if w_self.w_generators is None:
         if w_self.generators is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.generators]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_generators = w_list
     return w_self.w_generators
 
@@ -5528,8 +5448,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'elt'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'elt')
     return space.wrap(w_self.elt)
 
 def GeneratorExp_set_elt(space, w_self, w_new_value):
@@ -5546,14 +5465,13 @@
 def GeneratorExp_get_generators(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'generators'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'generators')
     if w_self.w_generators is None:
         if w_self.generators is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.generators]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_generators = w_list
     return w_self.w_generators
 
@@ -5594,8 +5512,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Yield_set_value(space, w_self, w_new_value):
@@ -5640,8 +5557,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'left'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'left')
     return space.wrap(w_self.left)
 
 def Compare_set_left(space, w_self, w_new_value):
@@ -5658,14 +5574,13 @@
 def Compare_get_ops(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ops'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ops')
     if w_self.w_ops is None:
         if w_self.ops is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [cmpop_to_class[node - 1]() for node in w_self.ops]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_ops = w_list
     return w_self.w_ops
 
@@ -5676,14 +5591,13 @@
 def Compare_get_comparators(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'comparators'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'comparators')
     if w_self.w_comparators is None:
         if w_self.comparators is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.comparators]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_comparators = w_list
     return w_self.w_comparators
 
@@ -5726,8 +5640,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'func'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'func')
     return space.wrap(w_self.func)
 
 def Call_set_func(space, w_self, w_new_value):
@@ -5744,14 +5657,13 @@
 def Call_get_args(space, w_self):
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'args'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'args')
     if w_self.w_args is None:
         if w_self.args is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.args]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_args = w_list
     return w_self.w_args
 
@@ -5762,14 +5674,13 @@
 def Call_get_keywords(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'keywords'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'keywords')
     if w_self.w_keywords is None:
         if w_self.keywords is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.keywords]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_keywords = w_list
     return w_self.w_keywords
 
@@ -5784,8 +5695,7 @@
             return w_obj
     if not w_self.initialization_state & 8:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'starargs'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'starargs')
     return space.wrap(w_self.starargs)
 
 def Call_set_starargs(space, w_self, w_new_value):
@@ -5806,8 +5716,7 @@
             return w_obj
     if not w_self.initialization_state & 16:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'kwargs'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'kwargs')
     return space.wrap(w_self.kwargs)
 
 def Call_set_kwargs(space, w_self, w_new_value):
@@ -5858,8 +5767,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Repr_set_value(space, w_self, w_new_value):
@@ -5904,8 +5812,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'n'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'n')
     return w_self.n
 
 def Num_set_n(space, w_self, w_new_value):
@@ -5950,8 +5857,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 's'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 's')
     return w_self.s
 
 def Str_set_s(space, w_self, w_new_value):
@@ -5996,8 +5902,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Attribute_set_value(space, w_self, w_new_value):
@@ -6018,8 +5923,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'attr'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'attr')
     return space.wrap(w_self.attr)
 
 def Attribute_set_attr(space, w_self, w_new_value):
@@ -6040,8 +5944,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ctx'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ctx')
     return expr_context_to_class[w_self.ctx - 1]()
 
 def Attribute_set_ctx(space, w_self, w_new_value):
@@ -6090,8 +5993,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Subscript_set_value(space, w_self, w_new_value):
@@ -6112,8 +6014,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'slice'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'slice')
     return space.wrap(w_self.slice)
 
 def Subscript_set_slice(space, w_self, w_new_value):
@@ -6134,8 +6035,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ctx'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ctx')
     return expr_context_to_class[w_self.ctx - 1]()
 
 def Subscript_set_ctx(space, w_self, w_new_value):
@@ -6184,8 +6084,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'id'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'id')
     return space.wrap(w_self.id)
 
 def Name_set_id(space, w_self, w_new_value):
@@ -6206,8 +6105,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ctx'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ctx')
     return expr_context_to_class[w_self.ctx - 1]()
 
 def Name_set_ctx(space, w_self, w_new_value):
@@ -6251,14 +6149,13 @@
 def List_get_elts(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'elts'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'elts')
     if w_self.w_elts is None:
         if w_self.elts is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.elts]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_elts = w_list
     return w_self.w_elts
 
@@ -6273,8 +6170,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ctx'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ctx')
     return expr_context_to_class[w_self.ctx - 1]()
 
 def List_set_ctx(space, w_self, w_new_value):
@@ -6319,14 +6215,13 @@
 def Tuple_get_elts(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'elts'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'elts')
     if w_self.w_elts is None:
         if w_self.elts is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.elts]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_elts = w_list
     return w_self.w_elts
 
@@ -6341,8 +6236,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ctx'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ctx')
     return expr_context_to_class[w_self.ctx - 1]()
 
 def Tuple_set_ctx(space, w_self, w_new_value):
@@ -6391,8 +6285,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return w_self.value
 
 def Const_set_value(space, w_self, w_new_value):
@@ -6510,8 +6403,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'lower'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'lower')
     return space.wrap(w_self.lower)
 
 def Slice_set_lower(space, w_self, w_new_value):
@@ -6532,8 +6424,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'upper'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'upper')
     return space.wrap(w_self.upper)
 
 def Slice_set_upper(space, w_self, w_new_value):
@@ -6554,8 +6445,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'step'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'step')
     return space.wrap(w_self.step)
 
 def Slice_set_step(space, w_self, w_new_value):
@@ -6598,14 +6488,13 @@
 def ExtSlice_get_dims(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'dims'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'dims')
     if w_self.w_dims is None:
         if w_self.dims is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.dims]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_dims = w_list
     return w_self.w_dims
 
@@ -6645,8 +6534,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def Index_set_value(space, w_self, w_new_value):
@@ -6915,8 +6803,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'target'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'target')
     return space.wrap(w_self.target)
 
 def comprehension_set_target(space, w_self, w_new_value):
@@ -6937,8 +6824,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'iter'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'iter')
     return space.wrap(w_self.iter)
 
 def comprehension_set_iter(space, w_self, w_new_value):
@@ -6955,14 +6841,13 @@
 def comprehension_get_ifs(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'ifs'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'ifs')
     if w_self.w_ifs is None:
         if w_self.ifs is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.ifs]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_ifs = w_list
     return w_self.w_ifs
 
@@ -7004,8 +6889,7 @@
             return w_obj
     if not w_self.initialization_state & w_self._lineno_mask:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'lineno'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'lineno')
     return space.wrap(w_self.lineno)
 
 def excepthandler_set_lineno(space, w_self, w_new_value):
@@ -7026,8 +6910,7 @@
             return w_obj
     if not w_self.initialization_state & w_self._col_offset_mask:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'col_offset'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'col_offset')
     return space.wrap(w_self.col_offset)
 
 def excepthandler_set_col_offset(space, w_self, w_new_value):
@@ -7057,8 +6940,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'type'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'type')
     return space.wrap(w_self.type)
 
 def ExceptHandler_set_type(space, w_self, w_new_value):
@@ -7079,8 +6961,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'name'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'name')
     return space.wrap(w_self.name)
 
 def ExceptHandler_set_name(space, w_self, w_new_value):
@@ -7097,14 +6978,13 @@
 def ExceptHandler_get_body(space, w_self):
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'body'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'body')
     if w_self.w_body is None:
         if w_self.body is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.body]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_body = w_list
     return w_self.w_body
 
@@ -7142,14 +7022,13 @@
 def arguments_get_args(space, w_self):
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'args'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'args')
     if w_self.w_args is None:
         if w_self.args is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.args]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_args = w_list
     return w_self.w_args
 
@@ -7164,8 +7043,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'vararg'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'vararg')
     return space.wrap(w_self.vararg)
 
 def arguments_set_vararg(space, w_self, w_new_value):
@@ -7189,8 +7067,7 @@
             return w_obj
     if not w_self.initialization_state & 4:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'kwarg'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'kwarg')
     return space.wrap(w_self.kwarg)
 
 def arguments_set_kwarg(space, w_self, w_new_value):
@@ -7210,14 +7087,13 @@
 def arguments_get_defaults(space, w_self):
     if not w_self.initialization_state & 8:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'defaults'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'defaults')
     if w_self.w_defaults is None:
         if w_self.defaults is None:
-            w_list = space.newlist([])
+            list_w = []
         else:
             list_w = [space.wrap(node) for node in w_self.defaults]
-            w_list = space.newlist(list_w)
+        w_list = space.newlist(list_w)
         w_self.w_defaults = w_list
     return w_self.w_defaults
 
@@ -7261,8 +7137,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'arg'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'arg')
     return space.wrap(w_self.arg)
 
 def keyword_set_arg(space, w_self, w_new_value):
@@ -7283,8 +7158,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'value'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'value')
     return space.wrap(w_self.value)
 
 def keyword_set_value(space, w_self, w_new_value):
@@ -7330,8 +7204,7 @@
             return w_obj
     if not w_self.initialization_state & 1:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'name'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'name')
     return space.wrap(w_self.name)
 
 def alias_set_name(space, w_self, w_new_value):
@@ -7352,8 +7225,7 @@
             return w_obj
     if not w_self.initialization_state & 2:
         typename = space.type(w_self).getname(space)
-        w_err = space.wrap("'%s' object has no attribute 'asname'" % typename)
-        raise OperationError(space.w_AttributeError, w_err)
+        raise operationerrfmt(space.w_AttributeError, "'%s' object has no attribute '%s'", typename, 'asname')
     return space.wrap(w_self.asname)
 
 def alias_set_asname(space, w_self, w_new_value):
diff --git a/pypy/interpreter/astcompiler/tools/asdl_py.py b/pypy/interpreter/astcompiler/tools/asdl_py.py
--- a/pypy/interpreter/astcompiler/tools/asdl_py.py
+++ b/pypy/interpreter/astcompiler/tools/asdl_py.py
@@ -414,13 +414,12 @@
             self.emit("        return w_obj", 1)
         self.emit("if not w_self.initialization_state & %s:" % (flag,), 1)
         self.emit("typename = space.type(w_self).getname(space)", 2)
-        self.emit("w_err = space.wrap(\"'%%s' object has no attribute '%s'\" %% typename)" %
+        self.emit("raise operationerrfmt(space.w_AttributeError, \"'%%s' object has no attribute '%%s'\", typename, '%s')" %
                   (field.name,), 2)
-        self.emit("raise OperationError(space.w_AttributeError, w_err)", 2)
         if field.seq:
             self.emit("if w_self.w_%s is None:" % (field.name,), 1)
             self.emit("if w_self.%s is None:" % (field.name,), 2)
-            self.emit("w_list = space.newlist([])", 3)
+            self.emit("list_w = []", 3)
             self.emit("else:", 2)
             if field.type.value in self.data.simple_types:
                 wrapper = "%s_to_class[node - 1]()" % (field.type,)
@@ -428,7 +427,7 @@
                 wrapper = "space.wrap(node)"
             self.emit("list_w = [%s for node in w_self.%s]" %
                       (wrapper, field.name), 3)
-            self.emit("w_list = space.newlist(list_w)", 3)
+            self.emit("w_list = space.newlist(list_w)", 2)
             self.emit("w_self.w_%s = w_list" % (field.name,), 2)
             self.emit("return w_self.w_%s" % (field.name,), 1)
         elif field.type.value in self.data.simple_types:
@@ -540,7 +539,7 @@
 from pypy.interpreter.baseobjspace import Wrappable
 from pypy.interpreter import typedef
 from pypy.interpreter.gateway import interp2app
-from pypy.interpreter.error import OperationError
+from pypy.interpreter.error import OperationError, operationerrfmt
 from pypy.rlib.unroll import unrolling_iterable
 from pypy.tool.pairtype import extendabletype
 from pypy.tool.sourcetools import func_with_new_name
@@ -639,9 +638,7 @@
             missing = required[i]
             if missing is not None:
                  err = "required field \\"%s\\" missing from %s"
-                 err = err % (missing, host)
-                 w_err = space.wrap(err)
-                 raise OperationError(space.w_TypeError, w_err)
+                 raise operationerrfmt(space.w_TypeError, err, missing, host)
     raise AssertionError("should not reach here")
 
 
diff --git a/pypy/interpreter/baseobjspace.py b/pypy/interpreter/baseobjspace.py
--- a/pypy/interpreter/baseobjspace.py
+++ b/pypy/interpreter/baseobjspace.py
@@ -777,22 +777,63 @@
         """Unpack an iterable object into a real (interpreter-level) list.
         Raise an OperationError(w_ValueError) if the length is wrong."""
         w_iterator = self.iter(w_iterable)
-        # If we know the expected length we can preallocate.
         if expected_length == -1:
+            # xxx special hack for speed
+            from pypy.interpreter.generator import GeneratorIterator
+            if isinstance(w_iterator, GeneratorIterator):
+                lst_w = []
+                w_iterator.unpack_into(lst_w)
+                return lst_w
+            # /xxx
+            return self._unpackiterable_unknown_length(w_iterator, w_iterable)
+        else:
+            lst_w = self._unpackiterable_known_length(w_iterator,
+                                                      expected_length)
+            return lst_w[:]     # make the resulting list resizable
+
+    @jit.dont_look_inside
+    def _unpackiterable_unknown_length(self, w_iterator, w_iterable):
+        # Unpack a variable-size list of unknown length.
+        # The JIT does not look inside this function because it
+        # contains a loop (made explicit with the decorator above).
+        #
+        # If we can guess the expected length we can preallocate.
+        try:
+            lgt_estimate = self.len_w(w_iterable)
+        except OperationError, o:
+            if (not o.match(self, self.w_AttributeError) and
+                not o.match(self, self.w_TypeError)):
+                raise
+            items = []
+        else:
             try:
-                lgt_estimate = self.len_w(w_iterable)
-            except OperationError, o:
-                if (not o.match(self, self.w_AttributeError) and
-                    not o.match(self, self.w_TypeError)):
+                items = newlist(lgt_estimate)
+            except MemoryError:
+                items = [] # it might have lied
+        #
+        while True:
+            try:
+                w_item = self.next(w_iterator)
+            except OperationError, e:
+                if not e.match(self, self.w_StopIteration):
                     raise
-                items = []
-            else:
-                try:
-                    items = newlist(lgt_estimate)
-                except MemoryError:
-                    items = [] # it might have lied
-        else:
-            items = [None] * expected_length
+                break  # done
+            items.append(w_item)
+        #
+        return items
+
+    @jit.dont_look_inside
+    def _unpackiterable_known_length(self, w_iterator, expected_length):
+        # Unpack a known length list, without letting the JIT look inside.
+        # Implemented by just calling the @jit.unroll_safe version, but
+        # the JIT stopped looking inside already.
+        return self._unpackiterable_known_length_jitlook(w_iterator,
+                                                         expected_length)
+
+    @jit.unroll_safe
+    def _unpackiterable_known_length_jitlook(self, w_iterator,
+                                             expected_length):
+        items = [None] * expected_length
         idx = 0
         while True:
             try:
@@ -801,26 +842,29 @@
                 if not e.match(self, self.w_StopIteration):
                     raise
                 break  # done
-            if expected_length != -1 and idx == expected_length:
+            if idx == expected_length:
                 raise OperationError(self.w_ValueError,
-                                     self.wrap("too many values to unpack"))
-            if expected_length == -1:
-                items.append(w_item)
-            else:
-                items[idx] = w_item
+                                    self.wrap("too many values to unpack"))
+            items[idx] = w_item
             idx += 1
-        if expected_length != -1 and idx < expected_length:
+        if idx < expected_length:
             if idx == 1:
                 plural = ""
             else:
                 plural = "s"
-            raise OperationError(self.w_ValueError,
-                      self.wrap("need more than %d value%s to unpack" %
-                                (idx, plural)))
+            raise operationerrfmt(self.w_ValueError,
+                                  "need more than %d value%s to unpack",
+                                  idx, plural)
         return items
 
-    unpackiterable_unroll = jit.unroll_safe(func_with_new_name(unpackiterable,
-                                            'unpackiterable_unroll'))
+    def unpackiterable_unroll(self, w_iterable, expected_length):
+        # Like unpackiterable(), but for the cases where we have
+        # an expected_length and want to unroll when JITted.
+        # Returns a fixed-size list.
+        w_iterator = self.iter(w_iterable)
+        assert expected_length != -1
+        return self._unpackiterable_known_length_jitlook(w_iterator,
+                                                         expected_length)
 
     def fixedview(self, w_iterable, expected_length=-1):
         """ A fixed list view of w_iterable. Don't modify the result
diff --git a/pypy/interpreter/executioncontext.py b/pypy/interpreter/executioncontext.py
--- a/pypy/interpreter/executioncontext.py
+++ b/pypy/interpreter/executioncontext.py
@@ -175,6 +175,9 @@
             self.w_tracefunc = w_func
             self.space.frame_trace_action.fire()
 
+    def gettrace(self):
+        return self.w_tracefunc
+
     def setprofile(self, w_func):
         """Set the global trace function."""
         if self.space.is_w(w_func, self.space.w_None):
@@ -388,8 +391,11 @@
     def decrement_ticker(self, by):
         value = self._ticker
         if self.has_bytecode_counter:    # this 'if' is constant-folded
-            value -= by
-            self._ticker = value
+            if jit.isconstant(by) and by == 0:
+                pass     # normally constant-folded too
+            else:
+                value -= by
+                self._ticker = value
         return value
 
 
diff --git a/pypy/interpreter/generator.py b/pypy/interpreter/generator.py
--- a/pypy/interpreter/generator.py
+++ b/pypy/interpreter/generator.py
@@ -8,7 +8,7 @@
 class GeneratorIterator(Wrappable):
     "An iterator created by a generator."
     _immutable_fields_ = ['pycode']
-    
+
     def __init__(self, frame):
         self.space = frame.space
         self.frame = frame     # turned into None when frame_finished_execution
@@ -81,7 +81,7 @@
             # if the frame is now marked as finished, it was RETURNed from
             if frame.frame_finished_execution:
                 self.frame = None
-                raise OperationError(space.w_StopIteration, space.w_None) 
+                raise OperationError(space.w_StopIteration, space.w_None)
             else:
                 return w_result     # YIELDed
         finally:
@@ -97,21 +97,21 @@
     def throw(self, w_type, w_val, w_tb):
         from pypy.interpreter.pytraceback import check_traceback
         space = self.space
-        
+
         msg = "throw() third argument must be a traceback object"
         if space.is_w(w_tb, space.w_None):
             tb = None
         else:
             tb = check_traceback(space, w_tb, msg)
-       
+
         operr = OperationError(w_type, w_val, tb)
         operr.normalize_exception(space)
         return self.send_ex(space.w_None, operr)
-             
+
     def descr_next(self):
         """x.next() -> the next value, or raise StopIteration"""
         return self.send_ex(self.space.w_None)
- 
+
     def descr_close(self):
         """x.close(arg) -> raise GeneratorExit inside generator."""
         assert isinstance(self, GeneratorIterator)
@@ -124,7 +124,7 @@
                     e.match(space, space.w_GeneratorExit):
                 return space.w_None
             raise
-        
+
         if w_retval is not None:
             msg = "generator ignored GeneratorExit"
             raise OperationError(space.w_RuntimeError, space.wrap(msg))
@@ -155,3 +155,39 @@
                                                  "interrupting generator of ")
                     break
                 block = block.previous
+
+    def unpack_into(self, results_w):
+        """This is a hack for performance: runs the generator and collects
+        all produced items in a list."""
+        # XXX copied and simplified version of send_ex()
+        space = self.space
+        if self.running:
+            raise OperationError(space.w_ValueError,
+                                 space.wrap('generator already executing'))
+        frame = self.frame
+        if frame is None:    # already finished
+            return
+        self.running = True
+        try:
+            pycode = self.pycode
+            while True:
+                jitdriver.jit_merge_point(self=self, frame=frame,
+                                          results_w=results_w,
+                                          pycode=pycode)
+                try:
+                    w_result = frame.execute_frame(space.w_None)
+                except OperationError, e:
+                    if not e.match(space, space.w_StopIteration):
+                        raise
+                    break
+                # if the frame is now marked as finished, it was RETURNed from
+                if frame.frame_finished_execution:
+                    break
+                results_w.append(w_result)     # YIELDed
+        finally:
+            frame.f_backref = jit.vref_None
+            self.running = False
+            self.frame = None
+
+jitdriver = jit.JitDriver(greens=['pycode'],
+                          reds=['self', 'frame', 'results_w'])
diff --git a/pypy/interpreter/pyparser/pytokenizer.py b/pypy/interpreter/pyparser/pytokenizer.py
--- a/pypy/interpreter/pyparser/pytokenizer.py
+++ b/pypy/interpreter/pyparser/pytokenizer.py
@@ -226,7 +226,7 @@
                         parenlev = parenlev - 1
                         if parenlev < 0:
                             raise TokenError("unmatched '%s'" % initial, line,
-                                             lnum-1, 0, token_list)
+                                             lnum, start + 1, token_list)
                     if token in python_opmap:
                         punct = python_opmap[token]
                     else:
diff --git a/pypy/interpreter/pyparser/test/test_pyparse.py b/pypy/interpreter/pyparser/test/test_pyparse.py
--- a/pypy/interpreter/pyparser/test/test_pyparse.py
+++ b/pypy/interpreter/pyparser/test/test_pyparse.py
@@ -87,6 +87,10 @@
         assert exc.lineno == 1
         assert exc.offset == 5
         assert exc.lastlineno == 5
+        exc = py.test.raises(SyntaxError, parse, "abc)").value
+        assert exc.msg == "unmatched ')'"
+        assert exc.lineno == 1
+        assert exc.offset == 4
 
     def test_is(self):
         self.parse("x is y")
diff --git a/pypy/interpreter/test/test_generator.py b/pypy/interpreter/test/test_generator.py
--- a/pypy/interpreter/test/test_generator.py
+++ b/pypy/interpreter/test/test_generator.py
@@ -117,7 +117,7 @@
         g = f()
         raises(NameError, g.throw, NameError, "Error", None)
 
-    
+
     def test_throw_fail(self):
         def f():
             yield 1
@@ -129,7 +129,7 @@
             yield 1
         g = f()
         raises(TypeError, g.throw, list())
- 
+
     def test_throw_fail3(self):
         def f():
             yield 1
@@ -188,7 +188,7 @@
         g = f()
         g.next()
         raises(NameError, g.close)
- 
+
     def test_close_fail(self):
         def f():
             try:
@@ -267,3 +267,15 @@
         assert r.startswith("<generator object myFunc at 0x")
         assert list(g) == [1]
         assert repr(g) == r
+
+    def test_unpackiterable_gen(self):
+        g = (i*i for i in range(-5, 3))
+        assert set(g) == set([0, 1, 4, 9, 16, 25])
+        assert set(g) == set()
+        assert set(i for i in range(0)) == set()
+
+    def test_explicit_stop_iteration_unpackiterable(self):
+        def f():
+            yield 1
+            raise StopIteration
+        assert tuple(f()) == (1,)
\ No newline at end of file
diff --git a/pypy/jit/backend/llgraph/llimpl.py b/pypy/jit/backend/llgraph/llimpl.py
--- a/pypy/jit/backend/llgraph/llimpl.py
+++ b/pypy/jit/backend/llgraph/llimpl.py
@@ -7,9 +7,7 @@
 import weakref
 from pypy.objspace.flow.model import Variable, Constant
 from pypy.annotation import model as annmodel
-from pypy.jit.metainterp.history import (ConstInt, ConstPtr,
-                                         BoxInt, BoxPtr, BoxObj, BoxFloat,
-                                         REF, INT, FLOAT)
+from pypy.jit.metainterp.history import REF, INT, FLOAT
 from pypy.jit.codewriter import heaptracker
 from pypy.rpython.lltypesystem import lltype, llmemory, rclass, rstr, rffi
 from pypy.rpython.ootypesystem import ootype
@@ -17,7 +15,7 @@
 from pypy.rpython.llinterp import LLException
 from pypy.rpython.extregistry import ExtRegistryEntry
 
-from pypy.jit.metainterp import resoperation, executor
+from pypy.jit.metainterp import resoperation
 from pypy.jit.metainterp.resoperation import rop
 from pypy.jit.backend.llgraph import symbolic
 from pypy.jit.codewriter import longlong
@@ -165,6 +163,7 @@
     'unicodegetitem'  : (('ref', 'int'), 'int'),
     'unicodesetitem'  : (('ref', 'int', 'int'), 'int'),
     'cast_ptr_to_int' : (('ref',), 'int'),
+    'cast_int_to_ptr' : (('int',), 'ref'),
     'debug_merge_point': (('ref', 'int'), None),
     'force_token'     : ((), 'int'),
     'call_may_force'  : (('int', 'varargs'), 'intorptr'),
@@ -333,6 +332,13 @@
     assert isinstance(type, str) and len(type) == 1
     op.descr = Descr(ofs, type, arg_types=arg_types)
 
+def compile_add_descr_arg(loop, ofs, type, arg_types):
+    from pypy.jit.backend.llgraph.runner import Descr
+    loop = _from_opaque(loop)
+    op = loop.operations[-1]
+    assert isinstance(type, str) and len(type) == 1
+    op.args.append(Descr(ofs, type, arg_types=arg_types))
+
 def compile_add_loop_token(loop, descr):
     if we_are_translated():
         raise ValueError("CALL_ASSEMBLER not supported")
@@ -437,8 +443,11 @@
         self._may_force = -1
 
     def getenv(self, v):
+        from pypy.jit.backend.llgraph.runner import Descr
         if isinstance(v, Constant):
             return v.value
+        elif isinstance(v, Descr):
+            return v
         else:
             return self.env[v]
 
@@ -806,6 +815,29 @@
         else:
             raise NotImplementedError
 
+    def op_getinteriorfield_gc(self, descr, array, index):
+        if descr.typeinfo == REF:
+            return do_getinteriorfield_gc_ptr(array, index, descr.ofs)
+        elif descr.typeinfo == INT:
+            return do_getinteriorfield_gc_int(array, index, descr.ofs)
+        elif descr.typeinfo == FLOAT:
+            return do_getinteriorfield_gc_float(array, index, descr.ofs)
+        else:
+            raise NotImplementedError
+
+    def op_setinteriorfield_gc(self, descr, array, index, newvalue):
+        if descr.typeinfo == REF:
+            return do_setinteriorfield_gc_ptr(array, index, descr.ofs,
+                                              newvalue)
+        elif descr.typeinfo == INT:
+            return do_setinteriorfield_gc_int(array, index, descr.ofs,
+                                              newvalue)
+        elif descr.typeinfo == FLOAT:
+            return do_setinteriorfield_gc_float(array, index, descr.ofs,
+                                                newvalue)
+        else:
+            raise NotImplementedError
+
     def op_setfield_gc(self, fielddescr, struct, newvalue):
         if fielddescr.typeinfo == REF:
             do_setfield_gc_ptr(struct, fielddescr.ofs, newvalue)
@@ -875,9 +907,6 @@
     def op_new_array(self, arraydescr, count):
         return do_new_array(arraydescr.ofs, count)
 
-    def op_cast_ptr_to_int(self, descr, ptr):
-        return cast_to_int(ptr)
-
     def op_force_token(self, descr):
         opaque_frame = _to_opaque(self)
         return llmemory.cast_ptr_to_adr(opaque_frame)
@@ -1085,7 +1114,9 @@
     if isinstance(TYPE, lltype.Ptr):
         if isinstance(x, (int, long, llmemory.AddressAsInt)):
             x = llmemory.cast_int_to_adr(x)
-        if TYPE is rffi.VOIDP or TYPE.TO._hints.get("uncast_on_llgraph"):
+        if TYPE is rffi.VOIDP or (
+                hasattr(TYPE.TO, '_hints') and
+                TYPE.TO._hints.get("uncast_on_llgraph")):
             # assume that we want a "C-style" cast, without typechecking the value
             return rffi.cast(TYPE, x)
         return llmemory.cast_adr_to_ptr(x, TYPE)
@@ -1356,6 +1387,22 @@
 def do_getfield_gc_ptr(struct, fieldnum):
     return cast_to_ptr(_getfield_gc(struct, fieldnum))
 
+def _getinteriorfield_gc(struct, fieldnum):
+    STRUCT, fieldname = symbolic.TokenToField[fieldnum]
+    return getattr(struct, fieldname)
+
+def do_getinteriorfield_gc_int(array, index, fieldnum):
+    struct = array._obj.container.getitem(index)
+    return cast_to_int(_getinteriorfield_gc(struct, fieldnum))
+
+def do_getinteriorfield_gc_float(array, index, fieldnum):
+    struct = array._obj.container.getitem(index)
+    return cast_to_floatstorage(_getinteriorfield_gc(struct, fieldnum))
+
+def do_getinteriorfield_gc_ptr(array, index, fieldnum):
+    struct = array._obj.container.getitem(index)
+    return cast_to_ptr(_getinteriorfield_gc(struct, fieldnum))
+
 def _getfield_raw(struct, fieldnum):
     STRUCT, fieldname = symbolic.TokenToField[fieldnum]
     ptr = cast_from_int(lltype.Ptr(STRUCT), struct)
@@ -1411,26 +1458,28 @@
     newvalue = cast_from_ptr(ITEMTYPE, newvalue)
     array.setitem(index, newvalue)
 
-def do_setfield_gc_int(struct, fieldnum, newvalue):
-    STRUCT, fieldname = symbolic.TokenToField[fieldnum]
-    ptr = lltype.cast_opaque_ptr(lltype.Ptr(STRUCT), struct)
-    FIELDTYPE = getattr(STRUCT, fieldname)
-    newvalue = cast_from_int(FIELDTYPE, newvalue)
-    setattr(ptr, fieldname, newvalue)
+def new_setfield_gc(cast_func):
+    def do_setfield_gc(struct, fieldnum, newvalue):
+        STRUCT, fieldname = symbolic.TokenToField[fieldnum]
+        ptr = lltype.cast_opaque_ptr(lltype.Ptr(STRUCT), struct)
+        FIELDTYPE = getattr(STRUCT, fieldname)
+        newvalue = cast_func(FIELDTYPE, newvalue)
+        setattr(ptr, fieldname, newvalue)
+    return do_setfield_gc
+do_setfield_gc_int = new_setfield_gc(cast_from_int)
+do_setfield_gc_float = new_setfield_gc(cast_from_floatstorage)
+do_setfield_gc_ptr = new_setfield_gc(cast_from_ptr)
 
-def do_setfield_gc_float(struct, fieldnum, newvalue):
-    STRUCT, fieldname = symbolic.TokenToField[fieldnum]
-    ptr = lltype.cast_opaque_ptr(lltype.Ptr(STRUCT), struct)
-    FIELDTYPE = getattr(STRUCT, fieldname)
-    newvalue = cast_from_floatstorage(FIELDTYPE, newvalue)
-    setattr(ptr, fieldname, newvalue)
-
-def do_setfield_gc_ptr(struct, fieldnum, newvalue):
-    STRUCT, fieldname = symbolic.TokenToField[fieldnum]
-    ptr = lltype.cast_opaque_ptr(lltype.Ptr(STRUCT), struct)
-    FIELDTYPE = getattr(STRUCT, fieldname)
-    newvalue = cast_from_ptr(FIELDTYPE, newvalue)
-    setattr(ptr, fieldname, newvalue)
+def new_setinteriorfield_gc(cast_func):
+    def do_setinteriorfield_gc(array, index, fieldnum, newvalue):
+        STRUCT, fieldname = symbolic.TokenToField[fieldnum]
+        struct = array._obj.container.getitem(index)
+        FIELDTYPE = getattr(STRUCT, fieldname)
+        setattr(struct, fieldname, cast_func(FIELDTYPE, newvalue))
+    return do_setinteriorfield_gc
+do_setinteriorfield_gc_int = new_setinteriorfield_gc(cast_from_int)
+do_setinteriorfield_gc_float = new_setinteriorfield_gc(cast_from_floatstorage)
+do_setinteriorfield_gc_ptr = new_setinteriorfield_gc(cast_from_ptr)        
 
 def do_setfield_raw_int(struct, fieldnum, newvalue):
     STRUCT, fieldname = symbolic.TokenToField[fieldnum]
@@ -1696,6 +1745,7 @@
 setannotation(compile_start_float_var, annmodel.SomeInteger())
 setannotation(compile_add, annmodel.s_None)
 setannotation(compile_add_descr, annmodel.s_None)
+setannotation(compile_add_descr_arg, annmodel.s_None)
 setannotation(compile_add_var, annmodel.s_None)
 setannotation(compile_add_int_const, annmodel.s_None)
 setannotation(compile_add_ref_const, annmodel.s_None)
@@ -1743,6 +1793,9 @@
 setannotation(do_getfield_raw_int, annmodel.SomeInteger())
 setannotation(do_getfield_raw_ptr, annmodel.SomePtr(llmemory.GCREF))
 setannotation(do_getfield_raw_float, s_FloatStorage)
+setannotation(do_getinteriorfield_gc_int, annmodel.SomeInteger())
+setannotation(do_getinteriorfield_gc_ptr, annmodel.SomePtr(llmemory.GCREF))
+setannotation(do_getinteriorfield_gc_float, s_FloatStorage)
 setannotation(do_new, annmodel.SomePtr(llmemory.GCREF))
 setannotation(do_new_array, annmodel.SomePtr(llmemory.GCREF))
 setannotation(do_setarrayitem_gc_int, annmodel.s_None)
@@ -1756,6 +1809,9 @@
 setannotation(do_setfield_raw_int, annmodel.s_None)
 setannotation(do_setfield_raw_ptr, annmodel.s_None)
 setannotation(do_setfield_raw_float, annmodel.s_None)
+setannotation(do_setinteriorfield_gc_int, annmodel.s_None)
+setannotation(do_setinteriorfield_gc_ptr, annmodel.s_None)
+setannotation(do_setinteriorfield_gc_float, annmodel.s_None)
 setannotation(do_newstr, annmodel.SomePtr(llmemory.GCREF))
 setannotation(do_strsetitem, annmodel.s_None)
 setannotation(do_newunicode, annmodel.SomePtr(llmemory.GCREF))
diff --git a/pypy/jit/backend/llgraph/runner.py b/pypy/jit/backend/llgraph/runner.py
--- a/pypy/jit/backend/llgraph/runner.py
+++ b/pypy/jit/backend/llgraph/runner.py
@@ -2,21 +2,19 @@
 Minimal-API wrapper around the llinterpreter to run operations.
 """
 
-import sys
 from pypy.rlib.unroll import unrolling_iterable
 from pypy.rlib.objectmodel import we_are_translated
 from pypy.rpython.lltypesystem import lltype, llmemory, rclass
 from pypy.rpython.ootypesystem import ootype
 from pypy.rpython.llinterp import LLInterpreter
 from pypy.jit.metainterp import history
-from pypy.jit.metainterp.history import REF, INT, FLOAT
+from pypy.jit.metainterp.history import REF, INT, FLOAT, STRUCT
 from pypy.jit.metainterp.warmstate import unwrap
-from pypy.jit.metainterp.resoperation import ResOperation, rop
+from pypy.jit.metainterp.resoperation import rop
 from pypy.jit.backend import model
 from pypy.jit.backend.llgraph import llimpl, symbolic
 from pypy.jit.metainterp.typesystem import llhelper, oohelper
 from pypy.jit.codewriter import heaptracker, longlong
-from pypy.rlib import rgc
 
 class MiniStats:
     pass
@@ -62,6 +60,9 @@
     def is_array_of_floats(self):
         return self.typeinfo == FLOAT
 
+    def is_array_of_structs(self):
+        return self.typeinfo == STRUCT
+
     def as_vtable_size_descr(self):
         return self
 
@@ -177,8 +178,10 @@
             llimpl.compile_add(c, op.getopnum())
             descr = op.getdescr()
             if isinstance(descr, Descr):
-                llimpl.compile_add_descr(c, descr.ofs, descr.typeinfo, descr.arg_types)
-            if isinstance(descr, history.LoopToken) and op.getopnum() != rop.JUMP:
+                llimpl.compile_add_descr(c, descr.ofs, descr.typeinfo,
+                                         descr.arg_types)
+            if (isinstance(descr, history.LoopToken) and
+                op.getopnum() != rop.JUMP):
                 llimpl.compile_add_loop_token(c, descr)
             if self.is_oo and isinstance(descr, (OODescr, MethDescr)):
                 # hack hack, not rpython
@@ -193,6 +196,9 @@
                     llimpl.compile_add_ref_const(c, x.value, self.ts.BASETYPE)
                 elif isinstance(x, history.ConstFloat):
                     llimpl.compile_add_float_const(c, x.value)
+                elif isinstance(x, Descr):
+                    llimpl.compile_add_descr_arg(c, x.ofs, x.typeinfo,
+                                                 x.arg_types)
                 else:
                     raise Exception("'%s' args contain: %r" % (op.getopname(),
                                                                x))
@@ -316,6 +322,13 @@
         token = history.getkind(getattr(S, fieldname))
         return self.getdescr(ofs, token[0], name=fieldname)
 
+    def interiorfielddescrof(self, A, fieldname):
+        S = A.OF
+        ofs2 = symbolic.get_size(A)
+        ofs, size = symbolic.get_field_token(S, fieldname)
+        token = history.getkind(getattr(S, fieldname))
+        return self.getdescr(ofs, token[0], name=fieldname, extrainfo=ofs2)
+
     def calldescrof(self, FUNC, ARGS, RESULT, extrainfo):
         arg_types = []
         for ARG in ARGS:
@@ -352,9 +365,15 @@
 
     def arraydescrof(self, A):
         assert A.OF != lltype.Void
+        assert isinstance(A, lltype.GcArray) or A._hints.get('nolength', False)
         size = symbolic.get_size(A)
-        token = history.getkind(A.OF)
-        return self.getdescr(size, token[0])
+        if isinstance(A.OF, lltype.Ptr) or isinstance(A.OF, lltype.Primitive):
+            token = history.getkind(A.OF)[0]
+        elif isinstance(A.OF, lltype.Struct):
+            token = 's'
+        else:
+            token = '?'
+        return self.getdescr(size, token)
 
     # ---------- the backend-dependent operations ----------
 
@@ -406,6 +425,29 @@
         assert isinstance(fielddescr, Descr)
         return llimpl.do_getfield_raw_float(struct, fielddescr.ofs)
 
+    def bh_getinteriorfield_gc_i(self, array, index, descr):
+        assert isinstance(descr, Descr)
+        return llimpl.do_getinteriorfield_gc_int(array, index, descr.ofs)
+    def bh_getinteriorfield_gc_r(self, array, index, descr):
+        assert isinstance(descr, Descr)
+        return llimpl.do_getinteriorfield_gc_ptr(array, index, descr.ofs)
+    def bh_getinteriorfield_gc_f(self, array, index, descr):
+        assert isinstance(descr, Descr)
+        return llimpl.do_getinteriorfield_gc_float(array, index, descr.ofs)
+
+    def bh_setinteriorfield_gc_i(self, array, index, descr, value):
+        assert isinstance(descr, Descr)
+        return llimpl.do_setinteriorfield_gc_int(array, index, descr.ofs,
+                                                 value)
+    def bh_setinteriorfield_gc_r(self, array, index, descr, value):
+        assert isinstance(descr, Descr)
+        return llimpl.do_setinteriorfield_gc_ptr(array, index, descr.ofs,
+                                                 value)
+    def bh_setinteriorfield_gc_f(self, array, index, descr, value):
+        assert isinstance(descr, Descr)
+        return llimpl.do_setinteriorfield_gc_float(array, index, descr.ofs,
+                                                   value)
+
     def bh_new(self, sizedescr):
         assert isinstance(sizedescr, Descr)
         return llimpl.do_new(sizedescr.ofs)
@@ -418,7 +460,6 @@
 
     def bh_classof(self, struct):
         struct = lltype.cast_opaque_ptr(rclass.OBJECTPTR, struct)
-        result = struct.typeptr
         result_adr = llmemory.cast_ptr_to_adr(struct.typeptr)
         return heaptracker.adr2int(result_adr)
 
diff --git a/pypy/jit/backend/llsupport/descr.py b/pypy/jit/backend/llsupport/descr.py
--- a/pypy/jit/backend/llsupport/descr.py
+++ b/pypy/jit/backend/llsupport/descr.py
@@ -1,13 +1,10 @@
 import py
-from pypy.rpython.lltypesystem import lltype, rffi, llmemory, rclass
+from pypy.rpython.lltypesystem import lltype, rffi, llmemory
 from pypy.rpython.lltypesystem.lloperation import llop
 from pypy.jit.backend.llsupport import symbolic, support
-from pypy.jit.metainterp.history import AbstractDescr, getkind, BoxInt, BoxPtr
-from pypy.jit.metainterp.history import BasicFailDescr, LoopToken, BoxFloat
+from pypy.jit.metainterp.history import AbstractDescr, getkind
 from pypy.jit.metainterp import history
-from pypy.jit.metainterp.resoperation import ResOperation, rop
 from pypy.jit.codewriter import heaptracker, longlong
-from pypy.rlib.rarithmetic import r_longlong, r_ulonglong
 
 # The point of the class organization in this file is to make instances
 # as compact as possible.  This is done by not storing the field size or
@@ -23,6 +20,7 @@
         self._cache_field = {}
         self._cache_array = {}
         self._cache_call = {}
+        self._cache_interiorfield = {}
 
     def init_size_descr(self, STRUCT, sizedescr):
         assert isinstance(STRUCT, lltype.GcStruct)
@@ -142,7 +140,6 @@
         cachedict[fieldname] = fielddescr
         return fielddescr
 
-
 # ____________________________________________________________
 # ArrayDescrs
 
@@ -167,6 +164,7 @@
 
     _is_array_of_pointers = False      # unless overridden by GcPtrArrayDescr
     _is_array_of_floats   = False      # unless overridden by FloatArrayDescr
+    _is_array_of_structs  = False      # unless overridden by StructArrayDescr
     _is_item_signed       = False      # unless overridden by XxxArrayDescr
 
     def is_array_of_pointers(self):
@@ -175,6 +173,9 @@
     def is_array_of_floats(self):
         return self._is_array_of_floats
 
+    def is_array_of_structs(self):
+        return self._is_array_of_structs
+
     def is_item_signed(self):
         return self._is_item_signed
 
@@ -199,6 +200,10 @@
     def get_item_size(self, translate_support_code):
         return symbolic.get_size(lltype.Float, translate_support_code)
 
+class StructArrayDescr(BaseArrayDescr):
+    _clsname = 'StructArrayDescr'
+    _is_array_of_structs = True
+
 class BaseArrayNoLengthDescr(BaseArrayDescr):
     def get_base_size(self, translate_support_code):
         return 0
@@ -218,6 +223,13 @@
 def getArrayDescrClass(ARRAY):
     if ARRAY.OF is lltype.Float:
         return FloatArrayDescr
+    elif isinstance(ARRAY.OF, lltype.Struct):
+        class Descr(StructArrayDescr):
+            _clsname = '%sArrayDescr' % ARRAY.OF._name
+            def get_item_size(self, translate_support_code):
+                return symbolic.get_size(ARRAY.OF, translate_support_code)
+        Descr.__name__ = Descr._clsname
+        return Descr
     return getDescrClass(ARRAY.OF, BaseArrayDescr, GcPtrArrayDescr,
                          NonGcPtrArrayDescr, 'Array', 'get_item_size',
                          '_is_array_of_floats', '_is_item_signed')
@@ -252,6 +264,39 @@
         cache[ARRAY] = arraydescr
         return arraydescr
 
+# ____________________________________________________________
+# InteriorFieldDescr
+
+class InteriorFieldDescr(AbstractDescr):
+    arraydescr = BaseArrayDescr()     # workaround for the annotator
+    fielddescr = BaseFieldDescr('', 0)
+
+    def __init__(self, arraydescr, fielddescr):
+        self.arraydescr = arraydescr
+        self.fielddescr = fielddescr
+
+    def is_pointer_field(self):
+        return self.fielddescr.is_pointer_field()
+
+    def is_float_field(self):
+        return self.fielddescr.is_float_field()
+
+    def sort_key(self):
+        return self.fielddescr.sort_key()
+
+    def repr_of_descr(self):
+        return '<InteriorFieldDescr %s>' % self.fielddescr.repr_of_descr()
+
+def get_interiorfield_descr(gc_ll_descr, ARRAY, FIELDTP, name):
+    cache = gc_ll_descr._cache_interiorfield
+    try:
+        return cache[(ARRAY, FIELDTP, name)]
+    except KeyError:
+        arraydescr = get_array_descr(gc_ll_descr, ARRAY)
+        fielddescr = get_field_descr(gc_ll_descr, FIELDTP, name)
+        descr = InteriorFieldDescr(arraydescr, fielddescr)
+        cache[(ARRAY, FIELDTP, name)] = descr
+        return descr
 
 # ____________________________________________________________
 # CallDescrs
@@ -525,7 +570,8 @@
         #
         if TYPE is lltype.Float or is_longlong(TYPE):
             setattr(Descr, floatattrname, True)
-        elif TYPE is not lltype.Bool and rffi.cast(TYPE, -1) == -1:
+        elif (TYPE is not lltype.Bool and isinstance(TYPE, lltype.Number) and
+              rffi.cast(TYPE, -1) == -1):
             setattr(Descr, signedattrname, True)
         #
         _cache[nameprefix, TYPE] = Descr
diff --git a/pypy/jit/backend/llsupport/gc.py b/pypy/jit/backend/llsupport/gc.py
--- a/pypy/jit/backend/llsupport/gc.py
+++ b/pypy/jit/backend/llsupport/gc.py
@@ -45,6 +45,22 @@
     def freeing_block(self, start, stop):
         pass
 
+    def get_funcptr_for_newarray(self):
+        return llhelper(self.GC_MALLOC_ARRAY, self.malloc_array)
+    def get_funcptr_for_newstr(self):
+        return llhelper(self.GC_MALLOC_STR_UNICODE, self.malloc_str)
+    def get_funcptr_for_newunicode(self):
+        return llhelper(self.GC_MALLOC_STR_UNICODE, self.malloc_unicode)
+
+
+    def record_constptrs(self, op, gcrefs_output_list):
+        for i in range(op.numargs()):
+            v = op.getarg(i)
+            if isinstance(v, ConstPtr) and bool(v.value):
+                p = v.value
+                rgc._make_sure_does_not_move(p)
+                gcrefs_output_list.append(p)
+
 # ____________________________________________________________
 
 class GcLLDescr_boehm(GcLLDescription):
@@ -88,6 +104,39 @@
         malloc_fn_ptr = self.configure_boehm_once()
         self.funcptr_for_new = malloc_fn_ptr
 
+        def malloc_array(basesize, itemsize, ofs_length, num_elem):
+            try:
+                size = ovfcheck(basesize + ovfcheck(itemsize * num_elem))
+            except OverflowError:
+                return lltype.nullptr(llmemory.GCREF.TO)
+            res = self.funcptr_for_new(size)
+            if not res:
+                return res
+            rffi.cast(rffi.CArrayPtr(lltype.Signed), res)[ofs_length/WORD] = num_elem
+            return res
+        self.malloc_array = malloc_array
+        self.GC_MALLOC_ARRAY = lltype.Ptr(lltype.FuncType(
+            [lltype.Signed] * 4, llmemory.GCREF))
+
+
+        (str_basesize, str_itemsize, str_ofs_length
+         ) = symbolic.get_array_token(rstr.STR, self.translate_support_code)
+        (unicode_basesize, unicode_itemsize, unicode_ofs_length
+         ) = symbolic.get_array_token(rstr.UNICODE, self.translate_support_code)
+        def malloc_str(length):
+            return self.malloc_array(
+                str_basesize, str_itemsize, str_ofs_length, length
+            )
+        def malloc_unicode(length):
+            return self.malloc_array(
+                unicode_basesize, unicode_itemsize, unicode_ofs_length, length
+            )
+        self.malloc_str = malloc_str
+        self.malloc_unicode = malloc_unicode
+        self.GC_MALLOC_STR_UNICODE = lltype.Ptr(lltype.FuncType(
+            [lltype.Signed], llmemory.GCREF))
+
+
         # on some platform GC_init is required before any other
         # GC_* functions, call it here for the benefit of tests
         # XXX move this to tests
@@ -108,38 +157,34 @@
         ofs_length = arraydescr.get_ofs_length(self.translate_support_code)
         basesize = arraydescr.get_base_size(self.translate_support_code)
         itemsize = arraydescr.get_item_size(self.translate_support_code)
-        size = basesize + itemsize * num_elem
-        res = self.funcptr_for_new(size)
-        rffi.cast(rffi.CArrayPtr(lltype.Signed), res)[ofs_length/WORD] = num_elem
-        return res
+        return self.malloc_array(basesize, itemsize, ofs_length, num_elem)
 
     def gc_malloc_str(self, num_elem):
-        basesize, itemsize, ofs_length = symbolic.get_array_token(rstr.STR,
-                                                   self.translate_support_code)
-        assert itemsize == 1
-        size = basesize + num_elem
-        res = self.funcptr_for_new(size)
-        rffi.cast(rffi.CArrayPtr(lltype.Signed), res)[ofs_length/WORD] = num_elem
-        return res
+        return self.malloc_str(num_elem)
 
     def gc_malloc_unicode(self, num_elem):
-        basesize, itemsize, ofs_length = symbolic.get_array_token(rstr.UNICODE,
-                                                   self.translate_support_code)
-        size = basesize + num_elem * itemsize
-        res = self.funcptr_for_new(size)
-        rffi.cast(rffi.CArrayPtr(lltype.Signed), res)[ofs_length/WORD] = num_elem
-        return res
+        return self.malloc_unicode(num_elem)
 
     def args_for_new(self, sizedescr):
         assert isinstance(sizedescr, BaseSizeDescr)
         return [sizedescr.size]
 
+    def args_for_new_array(self, arraydescr):
+        ofs_length = arraydescr.get_ofs_length(self.translate_support_code)
+        basesize = arraydescr.get_base_size(self.translate_support_code)
+        itemsize = arraydescr.get_item_size(self.translate_support_code)
+        return [basesize, itemsize, ofs_length]
+
     def get_funcptr_for_new(self):
         return self.funcptr_for_new
 
-    get_funcptr_for_newarray = None
-    get_funcptr_for_newstr = None
-    get_funcptr_for_newunicode = None
+    def rewrite_assembler(self, cpu, operations, gcrefs_output_list):
+        # record all GCREFs too, because Boehm cannot see them and keep them
+        # alive if they end up as constants in the assembler
+        for op in operations:
+            self.record_constptrs(op, gcrefs_output_list)
+        return GcLLDescription.rewrite_assembler(self, cpu, operations,
+                                                 gcrefs_output_list)
 
 
 # ____________________________________________________________
@@ -604,10 +649,13 @@
         def malloc_basic(size, tid):
             type_id = llop.extract_ushort(llgroup.HALFWORD, tid)
             has_finalizer = bool(tid & (1<<llgroup.HALFSHIFT))
+            has_light_finalizer = bool(tid & (1<<(llgroup.HALFSHIFT + 1)))
             check_typeid(type_id)
             res = llop1.do_malloc_fixedsize_clear(llmemory.GCREF,
                                                   type_id, size,
-                                                  has_finalizer, False)
+                                                  has_finalizer,
+                                                  has_light_finalizer,
+                                                  False)
             # In case the operation above failed, we are returning NULL
             # from this function to assembler.  There is also an RPython
             # exception set, typically MemoryError; but it's easier and
@@ -678,7 +726,7 @@
             # also use it to allocate varsized objects.  The tid
             # and possibly the length are both set afterward.
             gcref = llop1.do_malloc_fixedsize_clear(llmemory.GCREF,
-                                        0, size, False, False)
+                                        0, size, False, False, False)
             return rffi.cast(lltype.Signed, gcref)
         self.malloc_slowpath = malloc_slowpath
         self.MALLOC_SLOWPATH = lltype.FuncType([lltype.Signed], lltype.Signed)
@@ -702,7 +750,9 @@
         type_id = self.layoutbuilder.get_type_id(S)
         assert not self.layoutbuilder.is_weakref_type(S)
         has_finalizer = bool(self.layoutbuilder.has_finalizer(S))
-        flags = int(has_finalizer) << llgroup.HALFSHIFT
+        has_light_finalizer = bool(self.layoutbuilder.has_light_finalizer(S))
+        flags = (int(has_finalizer) << llgroup.HALFSHIFT |
+                 int(has_light_finalizer) << (llgroup.HALFSHIFT + 1))
         descr.tid = llop.combine_ushort(lltype.Signed, type_id, flags)
 
     def init_array_descr(self, A, descr):
@@ -736,15 +786,6 @@
     def get_funcptr_for_new(self):
         return llhelper(self.GC_MALLOC_BASIC, self.malloc_basic)
 
-    def get_funcptr_for_newarray(self):
-        return llhelper(self.GC_MALLOC_ARRAY, self.malloc_array)
-
-    def get_funcptr_for_newstr(self):
-        return llhelper(self.GC_MALLOC_STR_UNICODE, self.malloc_str)
-
-    def get_funcptr_for_newunicode(self):
-        return llhelper(self.GC_MALLOC_STR_UNICODE, self.malloc_unicode)
-
     def do_write_barrier(self, gcref_struct, gcref_newptr):
         hdr_addr = llmemory.cast_ptr_to_adr(gcref_struct)
         hdr_addr -= self.gcheaderbuilder.size_gc_header
@@ -757,14 +798,6 @@
             funcptr(llmemory.cast_ptr_to_adr(gcref_struct),
                     llmemory.cast_ptr_to_adr(gcref_newptr))
 
-    def record_constptrs(self, op, gcrefs_output_list):
-        for i in range(op.numargs()):
-            v = op.getarg(i)
-            if isinstance(v, ConstPtr) and bool(v.value):
-                p = v.value
-                rgc._make_sure_does_not_move(p)
-                gcrefs_output_list.append(p)
-
     def rewrite_assembler(self, cpu, operations, gcrefs_output_list):
         # Perform two kinds of rewrites in parallel:
         #
diff --git a/pypy/jit/backend/llsupport/llmodel.py b/pypy/jit/backend/llsupport/llmodel.py
--- a/pypy/jit/backend/llsupport/llmodel.py
+++ b/pypy/jit/backend/llsupport/llmodel.py
@@ -1,24 +1,18 @@
-import sys
 from pypy.rpython.lltypesystem import lltype, llmemory, rffi, rclass, rstr
 from pypy.rpython.lltypesystem.lloperation import llop
-from pypy.rpython.llinterp import LLInterpreter, LLException
+from pypy.rpython.llinterp import LLInterpreter
 from pypy.rpython.annlowlevel import llhelper
 from pypy.rlib.objectmodel import we_are_translated, specialize
-from pypy.jit.metainterp.history import BoxInt, BoxPtr, set_future_values,\
-     BoxFloat
 from pypy.jit.metainterp import history
 from pypy.jit.codewriter import heaptracker, longlong
 from pypy.jit.backend.model import AbstractCPU
 from pypy.jit.backend.llsupport import symbolic
 from pypy.jit.backend.llsupport.symbolic import WORD, unroll_basic_sizes
-from pypy.jit.backend.llsupport.descr import get_size_descr,  BaseSizeDescr
-from pypy.jit.backend.llsupport.descr import get_field_descr, BaseFieldDescr
-from pypy.jit.backend.llsupport.descr import get_array_descr, BaseArrayDescr
-from pypy.jit.backend.llsupport.descr import get_call_descr
-from pypy.jit.backend.llsupport.descr import BaseIntCallDescr, GcPtrCallDescr
-from pypy.jit.backend.llsupport.descr import FloatCallDescr, VoidCallDescr
+from pypy.jit.backend.llsupport.descr import (get_size_descr,
+     get_field_descr, BaseFieldDescr, get_array_descr, BaseArrayDescr,
+     get_call_descr, BaseIntCallDescr, GcPtrCallDescr, FloatCallDescr,
+     VoidCallDescr, InteriorFieldDescr, get_interiorfield_descr)
 from pypy.jit.backend.llsupport.asmmemmgr import AsmMemoryManager
-from pypy.rpython.annlowlevel import cast_instance_to_base_ptr
 
 
 class AbstractLLCPU(AbstractCPU):
@@ -241,6 +235,9 @@
     def arraydescrof(self, A):
         return get_array_descr(self.gc_ll_descr, A)
 
+    def interiorfielddescrof(self, A, fieldname):
+        return get_interiorfield_descr(self.gc_ll_descr, A, A.OF, fieldname)
+
     def unpack_arraydescr(self, arraydescr):
         assert isinstance(arraydescr, BaseArrayDescr)
         return arraydescr.get_base_size(self.translate_support_code)
@@ -358,6 +355,100 @@
     bh_getarrayitem_raw_i = bh_getarrayitem_gc_i
     bh_getarrayitem_raw_f = bh_getarrayitem_gc_f
 
+    def bh_getinteriorfield_gc_i(self, gcref, itemindex, descr):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs, size, _ = self.unpack_arraydescr_size(arraydescr)
+        ofs += descr.fielddescr.offset
+        fieldsize = descr.fielddescr.get_field_size(self.translate_support_code)
+        sign = descr.fielddescr.is_field_signed()
+        fullofs = itemindex * size + ofs
+        # --- start of GC unsafe code (no GC operation!) ---
+        items = rffi.ptradd(rffi.cast(rffi.CCHARP, gcref), fullofs)
+        for STYPE, UTYPE, itemsize in unroll_basic_sizes:
+            if fieldsize == itemsize:
+                if sign:
+                    item = rffi.cast(rffi.CArrayPtr(STYPE), items)
+                    val = item[0]
+                    val = rffi.cast(lltype.Signed, val)
+                else:
+                    item = rffi.cast(rffi.CArrayPtr(UTYPE), items)
+                    val = item[0]
+                    val = rffi.cast(lltype.Signed, val)
+                # --- end of GC unsafe code ---
+                return val
+        else:
+            raise NotImplementedError("size = %d" % fieldsize)
+
+    def bh_getinteriorfield_gc_r(self, gcref, itemindex, descr):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs, size, _ = self.unpack_arraydescr_size(arraydescr)
+        ofs += descr.fielddescr.offset
+        # --- start of GC unsafe code (no GC operation!) ---
+        items = rffi.ptradd(rffi.cast(rffi.CCHARP, gcref), ofs +
+                            size * itemindex)
+        items = rffi.cast(rffi.CArrayPtr(lltype.Signed), items)
+        pval = self._cast_int_to_gcref(items[0])
+        # --- end of GC unsafe code ---
+        return pval
+
+    def bh_getinteriorfield_gc_f(self, gcref, itemindex, descr):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs, size, _ = self.unpack_arraydescr_size(arraydescr)
+        ofs += descr.fielddescr.offset
+        # --- start of GC unsafe code (no GC operation!) ---
+        items = rffi.ptradd(rffi.cast(rffi.CCHARP, gcref), ofs +
+                            size * itemindex)
+        items = rffi.cast(rffi.CArrayPtr(longlong.FLOATSTORAGE), items)
+        fval = items[0]
+        # --- end of GC unsafe code ---
+        return fval
+
+    def bh_setinteriorfield_gc_i(self, gcref, itemindex, descr, value):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs, size, _ = self.unpack_arraydescr_size(arraydescr)
+        ofs += descr.fielddescr.offset
+        fieldsize = descr.fielddescr.get_field_size(self.translate_support_code)
+        ofs = itemindex * size + ofs
+        # --- start of GC unsafe code (no GC operation!) ---
+        items = rffi.ptradd(rffi.cast(rffi.CCHARP, gcref), ofs)
+        for TYPE, _, itemsize in unroll_basic_sizes:
+            if fieldsize == itemsize:
+                items = rffi.cast(rffi.CArrayPtr(TYPE), items)
+                items[0] = rffi.cast(TYPE, value)
+                # --- end of GC unsafe code ---
+                return
+        else:
+            raise NotImplementedError("size = %d" % fieldsize)
+
+    def bh_setinteriorfield_gc_r(self, gcref, itemindex, descr, newvalue):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs, size, _ = self.unpack_arraydescr_size(arraydescr)
+        ofs += descr.fielddescr.offset
+        self.gc_ll_descr.do_write_barrier(gcref, newvalue)
+        # --- start of GC unsafe code (no GC operation!) ---
+        items = rffi.ptradd(rffi.cast(rffi.CCHARP, gcref),
+                            ofs + size * itemindex)
+        items = rffi.cast(rffi.CArrayPtr(lltype.Signed), items)
+        items[0] = self.cast_gcref_to_int(newvalue)
+        # --- end of GC unsafe code ---
+
+    def bh_setinteriorfield_gc_f(self, gcref, itemindex, descr, newvalue):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs, size, _ = self.unpack_arraydescr_size(arraydescr)
+        ofs += descr.fielddescr.offset
+        # --- start of GC unsafe code (no GC operation!) ---
+        items = rffi.ptradd(rffi.cast(rffi.CCHARP, gcref),
+                            ofs + size * itemindex)
+        items = rffi.cast(rffi.CArrayPtr(longlong.FLOATSTORAGE), items)
+        items[0] = newvalue
+        # --- end of GC unsafe code ---
+
     def bh_strlen(self, string):
         s = lltype.cast_opaque_ptr(lltype.Ptr(rstr.STR), string)
         return len(s.chars)
@@ -475,7 +566,6 @@
 
     def bh_classof(self, struct):
         struct = lltype.cast_opaque_ptr(rclass.OBJECTPTR, struct)
-        result = struct.typeptr
         result_adr = llmemory.cast_ptr_to_adr(struct.typeptr)
         return heaptracker.adr2int(result_adr)
 
diff --git a/pypy/jit/backend/llsupport/regalloc.py b/pypy/jit/backend/llsupport/regalloc.py
--- a/pypy/jit/backend/llsupport/regalloc.py
+++ b/pypy/jit/backend/llsupport/regalloc.py
@@ -287,7 +287,6 @@
         self.reg_bindings[to_v] = reg
 
     def _move_variable_away(self, v, prev_loc):
-        reg = None
         if self.free_regs:
             loc = self.free_regs.pop()
             self.reg_bindings[v] = loc
diff --git a/pypy/jit/backend/llsupport/test/test_asmmemmgr.py b/pypy/jit/backend/llsupport/test/test_asmmemmgr.py
--- a/pypy/jit/backend/llsupport/test/test_asmmemmgr.py
+++ b/pypy/jit/backend/llsupport/test/test_asmmemmgr.py
@@ -211,14 +211,14 @@
     debug._log = debug.DebugLog()
     try:
         mc._dump(addr, 'test-logname-section')
-        log = list(debug._log)
+        log = list(debug._log) 
     finally:
         debug._log = None
     encoded = ''.join(writtencode).encode('hex').upper()
     ataddr = '@%x' % addr
     assert log == [('test-logname-section',
                     [('debug_print', 'CODE_DUMP', ataddr, '+0 ', encoded)])]
-    #
+    
     lltype.free(p, flavor='raw')
 
 def test_blockbuildermixin2():
diff --git a/pypy/jit/backend/llsupport/test/test_descr.py b/pypy/jit/backend/llsupport/test/test_descr.py
--- a/pypy/jit/backend/llsupport/test/test_descr.py
+++ b/pypy/jit/backend/llsupport/test/test_descr.py
@@ -3,7 +3,6 @@
 from pypy.jit.backend.llsupport import symbolic
 from pypy.rlib.objectmodel import Symbolic
 from pypy.rpython.annlowlevel import llhelper
-from pypy.jit.metainterp.history import BoxInt, BoxFloat, BoxPtr
 from pypy.jit.metainterp import history
 from pypy.jit.codewriter import longlong
 import sys, struct, py
@@ -149,7 +148,9 @@
     A2 = lltype.GcArray(lltype.Ptr(T))
     A3 = lltype.GcArray(lltype.Ptr(U))
     A4 = lltype.GcArray(lltype.Float)
-    A5 = lltype.GcArray(lltype.SingleFloat)
+    A5 = lltype.GcArray(lltype.Struct('x', ('v', lltype.Signed),
+                                      ('k', lltype.Signed)))
+    A6 = lltype.GcArray(lltype.SingleFloat)
     assert getArrayDescrClass(A2) is GcPtrArrayDescr
     assert getArrayDescrClass(A3) is NonGcPtrArrayDescr
     cls = getArrayDescrClass(A1)
@@ -158,7 +159,7 @@
     clsf = getArrayDescrClass(A4)
     assert clsf != cls
     assert clsf == getArrayDescrClass(lltype.GcArray(lltype.Float))
-    clss = getArrayDescrClass(A5)
+    clss = getArrayDescrClass(A6)
     assert clss not in (clsf, cls)
     assert clss == getArrayDescrClass(lltype.GcArray(rffi.UINT))
     #
@@ -168,11 +169,12 @@
     descr3 = get_array_descr(c0, A3)
     descr4 = get_array_descr(c0, A4)
     descr5 = get_array_descr(c0, A5)
+    descr6 = get_array_descr(c0, A6)
     assert descr1.__class__ is cls
     assert descr2.__class__ is GcPtrArrayDescr
     assert descr3.__class__ is NonGcPtrArrayDescr
     assert descr4.__class__ is clsf
-    assert descr5.__class__ is clss
+    assert descr6.__class__ is clss
     assert descr1 == get_array_descr(c0, lltype.GcArray(lltype.Char))
     assert not descr1.is_array_of_pointers()
     assert     descr2.is_array_of_pointers()
@@ -202,7 +204,8 @@
     assert descr2.get_item_size(False) == rffi.sizeof(lltype.Ptr(T))
     assert descr3.get_item_size(False) == rffi.sizeof(lltype.Ptr(U))
     assert descr4.get_item_size(False) == rffi.sizeof(lltype.Float)
-    assert descr5.get_item_size(False) == rffi.sizeof(lltype.SingleFloat)
+    assert descr5.get_item_size(False) == rffi.sizeof(lltype.Signed) * 2
+    assert descr6.get_item_size(False) == rffi.sizeof(lltype.SingleFloat)
     #
     assert isinstance(descr1.get_base_size(True), Symbolic)
     assert isinstance(descr2.get_base_size(True), Symbolic)
@@ -348,7 +351,6 @@
                             (rffi.SHORT,      True), (rffi.USHORT, False),
                             (rffi.INT,        True), (rffi.UINT,   False),
                             (rffi.LONG,       True), (rffi.ULONG,  False)]:
-        A = lltype.GcArray(RESTYPE)
         for tsc in [False, True]:
             c2 = GcCache(tsc)
             descr1 = get_call_descr(c2, [], RESTYPE)
@@ -379,7 +381,6 @@
     descr3i = get_array_descr(c0, lltype.GcArray(lltype.Char))
     assert descr3i.repr_of_descr() == '<CharArrayDescr>'
     #
-    cache = {}
     descr4 = get_call_descr(c0, [lltype.Char, lltype.Ptr(S)], lltype.Ptr(S))
     assert 'GcPtrCallDescr' in descr4.repr_of_descr()
     #
@@ -412,10 +413,10 @@
     ARGS = [lltype.Float, lltype.Ptr(ARRAY)]
     RES = lltype.Float
 
-    def f(a, b):
+    def f2(a, b):
         return float(b[0]) + a
 
-    fnptr = llhelper(lltype.Ptr(lltype.FuncType(ARGS, RES)), f)
+    fnptr = llhelper(lltype.Ptr(lltype.FuncType(ARGS, RES)), f2)
     descr2 = get_call_descr(c0, ARGS, RES)
     a = lltype.malloc(ARRAY, 3)
     opaquea = lltype.cast_opaque_ptr(llmemory.GCREF, a)
diff --git a/pypy/jit/backend/llsupport/test/test_gc.py b/pypy/jit/backend/llsupport/test/test_gc.py
--- a/pypy/jit/backend/llsupport/test/test_gc.py
+++ b/pypy/jit/backend/llsupport/test/test_gc.py
@@ -247,12 +247,14 @@
         self.record = []
 
     def do_malloc_fixedsize_clear(self, RESTYPE, type_id, size,
-                                  has_finalizer, contains_weakptr):
+                                  has_finalizer, has_light_finalizer,
+                                  contains_weakptr):
         assert not contains_weakptr
+        assert not has_finalizer           # in these tests
+        assert not has_light_finalizer     # in these tests
         p = llmemory.raw_malloc(size)
         p = llmemory.cast_adr_to_ptr(p, RESTYPE)
-        flags = int(has_finalizer) << 16
-        tid = llop.combine_ushort(lltype.Signed, type_id, flags)
+        tid = llop.combine_ushort(lltype.Signed, type_id, 0)
         self.record.append(("fixedsize", repr(size), tid, p))
         return p
 
diff --git a/pypy/jit/backend/model.py b/pypy/jit/backend/model.py
--- a/pypy/jit/backend/model.py
+++ b/pypy/jit/backend/model.py
@@ -1,5 +1,5 @@
 from pypy.rlib.debug import debug_start, debug_print, debug_stop
-from pypy.jit.metainterp import history, compile
+from pypy.jit.metainterp import history
 
 
 class AbstractCPU(object):
@@ -213,6 +213,10 @@
     def typedescrof(TYPE):
         raise NotImplementedError
 
+    @staticmethod
+    def interiorfielddescrof(A, fieldname):
+        raise NotImplementedError
+
     # ---------- the backend-dependent operations ----------
 
     # lltype specific operations
diff --git a/pypy/jit/backend/test/runner_test.py b/pypy/jit/backend/test/runner_test.py
--- a/pypy/jit/backend/test/runner_test.py
+++ b/pypy/jit/backend/test/runner_test.py
@@ -5,7 +5,7 @@
                                          BoxInt, Box, BoxPtr,
                                          LoopToken,
                                          ConstInt, ConstPtr,
-                                         BoxObj, Const,
+                                         BoxObj,
                                          ConstObj, BoxFloat, ConstFloat)
 from pypy.jit.metainterp.resoperation import ResOperation, rop
 from pypy.jit.metainterp.typesystem import deref
@@ -111,7 +111,7 @@
         self.cpu.set_future_value_int(0, 2)
         fail = self.cpu.execute_token(looptoken)
         res = self.cpu.get_latest_value_int(0)
-        assert res == 3        
+        assert res == 3
         assert fail.identifier == 1
 
     def test_compile_loop(self):
@@ -127,7 +127,7 @@
             ]
         inputargs = [i0]
         operations[2].setfailargs([i1])
-        
+
         self.cpu.compile_loop(inputargs, operations, looptoken)
         self.cpu.set_future_value_int(0, 2)
         fail = self.cpu.execute_token(looptoken)
@@ -148,7 +148,7 @@
             ]
         inputargs = [i0]
         operations[2].setfailargs([None, None, i1, None])
-        
+
         self.cpu.compile_loop(inputargs, operations, looptoken)
         self.cpu.set_future_value_int(0, 2)
         fail = self.cpu.execute_token(looptoken)
@@ -372,7 +372,7 @@
         for opnum, boxargs, retvalue in get_int_tests():
             res = self.execute_operation(opnum, boxargs, 'int')
             assert res.value == retvalue
-        
+
     def test_float_operations(self):
         from pypy.jit.metainterp.test.test_executor import get_float_tests
         for opnum, boxargs, rettype, retvalue in get_float_tests(self.cpu):
@@ -438,7 +438,7 @@
 
     def test_ovf_operations_reversed(self):
         self.test_ovf_operations(reversed=True)
-        
+
     def test_bh_call(self):
         cpu = self.cpu
         #
@@ -503,7 +503,7 @@
                                          [funcbox, BoxInt(num), BoxInt(num)],
                                          'int', descr=dyn_calldescr)
             assert res.value == 2 * num
-            
+
 
         if cpu.supports_floats:
             def func(f0, f1, f2, f3, f4, f5, f6, i0, i1, f7, f8, f9):
@@ -543,7 +543,7 @@
         funcbox = self.get_funcbox(self.cpu, func_ptr)
         res = self.execute_operation(rop.CALL, [funcbox] + map(BoxInt, args), 'int', descr=calldescr)
         assert res.value == func(*args)
-        
+
     def test_call_stack_alignment(self):
         # test stack alignment issues, notably for Mac OS/X.
         # also test the ordering of the arguments.
@@ -615,7 +615,7 @@
         res = self.execute_operation(rop.GETFIELD_GC, [t_box],
                                      'int', descr=shortdescr)
         assert res.value == 1331
-        
+
         #
         u_box, U_box = self.alloc_instance(self.U)
         fielddescr2 = self.cpu.fielddescrof(self.S, 'next')
@@ -695,7 +695,7 @@
 
     def test_failing_guard_class(self):
         t_box, T_box = self.alloc_instance(self.T)
-        u_box, U_box = self.alloc_instance(self.U)        
+        u_box, U_box = self.alloc_instance(self.U)
         null_box = self.null_instance()
         for opname, args in [(rop.GUARD_CLASS, [t_box, U_box]),
                              (rop.GUARD_CLASS, [u_box, T_box]),
@@ -787,7 +787,7 @@
         r = self.execute_operation(rop.GETARRAYITEM_GC, [a_box, BoxInt(3)],
                                    'int', descr=arraydescr)
         assert r.value == 160
-        
+
         #
         if isinstance(A, lltype.GcArray):
             A = lltype.Ptr(A)
@@ -880,6 +880,73 @@
                                    'int', descr=arraydescr)
         assert r.value == 7441
 
+    def test_array_of_structs(self):
+        TP = lltype.GcStruct('x')
+        ITEM = lltype.Struct('x',
+                             ('vs', lltype.Signed),
+                             ('vu', lltype.Unsigned),
+                             ('vsc', rffi.SIGNEDCHAR),
+                             ('vuc', rffi.UCHAR),
+                             ('vss', rffi.SHORT),
+                             ('vus', rffi.USHORT),
+                             ('vsi', rffi.INT),
+                             ('vui', rffi.UINT),
+                             ('k', lltype.Float),
+                             ('p', lltype.Ptr(TP)))
+        a_box, A = self.alloc_array_of(ITEM, 15)
+        s_box, S = self.alloc_instance(TP)
+        kdescr = self.cpu.interiorfielddescrof(A, 'k')
+        pdescr = self.cpu.interiorfielddescrof(A, 'p')
+        self.execute_operation(rop.SETINTERIORFIELD_GC, [a_box, BoxInt(3),
+                                                         boxfloat(1.5)],
+                               'void', descr=kdescr)
+        f = self.cpu.bh_getinteriorfield_gc_f(a_box.getref_base(), 3, kdescr)
+        assert longlong.getrealfloat(f) == 1.5
+        self.cpu.bh_setinteriorfield_gc_f(a_box.getref_base(), 3, kdescr, longlong.getfloatstorage(2.5))
+        r = self.execute_operation(rop.GETINTERIORFIELD_GC, [a_box, BoxInt(3)],
+                                   'float', descr=kdescr)
+        assert r.getfloat() == 2.5
+        #
+        NUMBER_FIELDS = [('vs', lltype.Signed),
+                         ('vu', lltype.Unsigned),
+                         ('vsc', rffi.SIGNEDCHAR),
+                         ('vuc', rffi.UCHAR),
+                         ('vss', rffi.SHORT),
+                         ('vus', rffi.USHORT),
+                         ('vsi', rffi.INT),
+                         ('vui', rffi.UINT)]
+        for name, TYPE in NUMBER_FIELDS[::-1]:
+            vdescr = self.cpu.interiorfielddescrof(A, name)
+            self.execute_operation(rop.SETINTERIORFIELD_GC, [a_box, BoxInt(3),
+                                                             BoxInt(-15)],
+                                   'void', descr=vdescr)
+        for name, TYPE in NUMBER_FIELDS:
+            vdescr = self.cpu.interiorfielddescrof(A, name)
+            i = self.cpu.bh_getinteriorfield_gc_i(a_box.getref_base(), 3,
+                                                  vdescr)
+            assert i == rffi.cast(lltype.Signed, rffi.cast(TYPE, -15))
+        for name, TYPE in NUMBER_FIELDS[::-1]:
+            vdescr = self.cpu.interiorfielddescrof(A, name)
+            self.cpu.bh_setinteriorfield_gc_i(a_box.getref_base(), 3,
+                                              vdescr, -25)
+        for name, TYPE in NUMBER_FIELDS:
+            vdescr = self.cpu.interiorfielddescrof(A, name)
+            r = self.execute_operation(rop.GETINTERIORFIELD_GC,
+                                       [a_box, BoxInt(3)],
+                                       'int', descr=vdescr)
+            assert r.getint() == rffi.cast(lltype.Signed, rffi.cast(TYPE, -25))
+        #
+        self.execute_operation(rop.SETINTERIORFIELD_GC, [a_box, BoxInt(4),
+                                                         s_box],
+                               'void', descr=pdescr)
+        r = self.cpu.bh_getinteriorfield_gc_r(a_box.getref_base(), 4, pdescr)
+        assert r == s_box.getref_base()
+        self.cpu.bh_setinteriorfield_gc_r(a_box.getref_base(), 3, pdescr,
+                                          s_box.getref_base())
+        r = self.execute_operation(rop.GETINTERIORFIELD_GC, [a_box, BoxInt(3)],
+                                   'ref', descr=pdescr)
+        assert r.getref_base() == s_box.getref_base()
+
     def test_string_basic(self):
         s_box = self.alloc_string("hello\xfe")
         r = self.execute_operation(rop.STRLEN, [s_box], 'int')
@@ -1402,7 +1469,7 @@
         addr = llmemory.cast_ptr_to_adr(func_ptr)
         return ConstInt(heaptracker.adr2int(addr))
 
-    
+
     MY_VTABLE = rclass.OBJECT_VTABLE    # for tests only
 
     S = lltype.GcForwardReference()
@@ -1439,7 +1506,6 @@
         return BoxPtr(lltype.nullptr(llmemory.GCREF.TO))
 
     def alloc_array_of(self, ITEM, length):
-        cpu = self.cpu
         A = lltype.GcArray(ITEM)
         a = lltype.malloc(A, length)
         a_box = BoxPtr(lltype.cast_opaque_ptr(llmemory.GCREF, a))
@@ -1468,20 +1534,16 @@
         return u''.join(u.chars)
 
 
-    def test_casts(self):
-        py.test.skip("xxx fix or kill")
-        from pypy.rpython.lltypesystem import lltype, llmemory
-        TP = lltype.GcStruct('x')
-        x = lltype.malloc(TP)        
-        x = lltype.cast_opaque_ptr(llmemory.GCREF, x)
+    def test_cast_int_to_ptr(self):
+        res = self.execute_operation(rop.CAST_INT_TO_PTR,
+                                     [BoxInt(-17)],  'ref').value
+        assert lltype.cast_ptr_to_int(res) == -17
+
+    def test_cast_ptr_to_int(self):
+        x = lltype.cast_int_to_ptr(llmemory.GCREF, -19)
         res = self.execute_operation(rop.CAST_PTR_TO_INT,
-                                     [BoxPtr(x)],  'int').value
-        expected = self.cpu.cast_adr_to_int(llmemory.cast_ptr_to_adr(x))
-        assert rffi.get_real_int(res) == rffi.get_real_int(expected)
-        res = self.execute_operation(rop.CAST_PTR_TO_INT,
-                                     [ConstPtr(x)],  'int').value
-        expected = self.cpu.cast_adr_to_int(llmemory.cast_ptr_to_adr(x))
-        assert rffi.get_real_int(res) == rffi.get_real_int(expected)
+                                     [BoxPtr(x)], 'int').value
+        assert res == -19
 
     def test_ooops_non_gc(self):
         x = lltype.malloc(lltype.Struct('x'), flavor='raw')
@@ -2299,13 +2361,6 @@
         #
         cpu.bh_strsetitem(x, 4, ord('/'))
         assert str.chars[4] == '/'
-        #
-##        x = cpu.bh_newstr(5)
-##        y = cpu.bh_cast_ptr_to_int(x)
-##        z = cpu.bh_cast_ptr_to_int(x)
-##        y = rffi.get_real_int(y)
-##        z = rffi.get_real_int(z)
-##        assert type(y) == type(z) == int and y == z
 
     def test_sorting_of_fields(self):
         S = self.S
@@ -2329,7 +2384,7 @@
         for opname, arg, res in ops:
             self.execute_operation(opname, [arg], 'void')
             assert self.guard_failed == res
-        
+
         lltype.free(x, flavor='raw')
 
     def test_assembler_call(self):
@@ -2409,7 +2464,7 @@
         FakeJitDriverSD.portal_calldescr = self.cpu.calldescrof(
             lltype.Ptr(lltype.FuncType(ARGS, RES)), ARGS, RES,
             EffectInfo.MOST_GENERAL)
-        
+
         ops = '''
         [f0, f1]
         f2 = float_add(f0, f1)
@@ -2500,7 +2555,7 @@
         FakeJitDriverSD.portal_calldescr = self.cpu.calldescrof(
             lltype.Ptr(lltype.FuncType(ARGS, RES)), ARGS, RES,
             EffectInfo.MOST_GENERAL)
-        
+
         ops = '''
         [f0, f1]
         f2 = float_add(f0, f1)
@@ -2951,4 +3006,4 @@
 
     def alloc_unicode(self, unicode):
         py.test.skip("implement me")
-    
+
diff --git a/pypy/jit/backend/test/test_ll_random.py b/pypy/jit/backend/test/test_ll_random.py
--- a/pypy/jit/backend/test/test_ll_random.py
+++ b/pypy/jit/backend/test/test_ll_random.py
@@ -28,16 +28,27 @@
         fork.structure_types_and_vtables = self.structure_types_and_vtables
         return fork
 
-    def get_structptr_var(self, r, must_have_vtable=False, type=lltype.Struct):
+    def _choose_ptr_vars(self, from_, type, array_of_structs):
+        ptrvars = []
+        for i in range(len(from_)):
+            v, S = from_[i][:2]
+            if not isinstance(S, type):
+                continue
+            if ((isinstance(S, lltype.Array) and
+                 isinstance(S.OF, lltype.Struct)) == array_of_structs):
+                ptrvars.append((v, S))
+        return ptrvars
+
+    def get_structptr_var(self, r, must_have_vtable=False, type=lltype.Struct,
+                          array_of_structs=False):
         while True:
-            ptrvars = [(v, S) for (v, S) in self.ptrvars
-                              if isinstance(S, type)]
+            ptrvars = self._choose_ptr_vars(self.ptrvars, type,
+                                            array_of_structs)
             if ptrvars and r.random() < 0.8:
                 v, S = r.choice(ptrvars)
             else:
-                prebuilt_ptr_consts = [(v, S)
-                                 for (v, S, _) in self.prebuilt_ptr_consts
-                                 if isinstance(S, type)]
+                prebuilt_ptr_consts = self._choose_ptr_vars(
+                    self.prebuilt_ptr_consts, type, array_of_structs)
                 if prebuilt_ptr_consts and r.random() < 0.7:
                     v, S = r.choice(prebuilt_ptr_consts)
                 else:
@@ -48,7 +59,8 @@
                                                 has_vtable=must_have_vtable)
                     else:
                         # create a new constant array
-                        p = self.get_random_array(r)
+                        p = self.get_random_array(r,
+                                    must_be_array_of_structs=array_of_structs)
                     S = lltype.typeOf(p).TO
                     v = ConstPtr(lltype.cast_opaque_ptr(llmemory.GCREF, p))
                     self.prebuilt_ptr_consts.append((v, S,
@@ -74,7 +86,8 @@
                 TYPE = lltype.Signed
         return TYPE
 
-    def get_random_structure_type(self, r, with_vtable=None, cache=True):
+    def get_random_structure_type(self, r, with_vtable=None, cache=True,
+                                  type=lltype.GcStruct):
         if cache and self.structure_types and r.random() < 0.5:
             return r.choice(self.structure_types)
         fields = []
@@ -85,7 +98,7 @@
         for i in range(r.randrange(1, 5)):
             TYPE = self.get_random_primitive_type(r)
             fields.append(('f%d' % i, TYPE))
-        S = lltype.GcStruct('S%d' % self.counter, *fields, **kwds)
+        S = type('S%d' % self.counter, *fields, **kwds)
         self.counter += 1
         if cache:
             self.structure_types.append(S)
@@ -125,17 +138,29 @@
                 setattr(p, fieldname, rffi.cast(TYPE, r.random_integer()))
         return p
 
-    def get_random_array_type(self, r):
-        TYPE = self.get_random_primitive_type(r)
+    def get_random_array_type(self, r, can_be_array_of_struct=False,
+                              must_be_array_of_structs=False):
+        if ((can_be_array_of_struct and r.random() < 0.1) or
+            must_be_array_of_structs):
+            TYPE = self.get_random_structure_type(r, cache=False,
+                                                  type=lltype.Struct)
+        else:
+            TYPE = self.get_random_primitive_type(r)
         return lltype.GcArray(TYPE)
 
-    def get_random_array(self, r):
-        A = self.get_random_array_type(r)
+    def get_random_array(self, r, must_be_array_of_structs=False):
+        A = self.get_random_array_type(r,
+                           must_be_array_of_structs=must_be_array_of_structs)
         length = (r.random_integer() // 15) % 300  # length: between 0 and 299
                                                    # likely to be small
         p = lltype.malloc(A, length)
-        for i in range(length):
-            p[i] = rffi.cast(A.OF, r.random_integer())
+        if isinstance(A.OF, lltype.Primitive):
+            for i in range(length):
+                p[i] = rffi.cast(A.OF, r.random_integer())
+        else:
+            for i in range(length):
+                for fname, TP in A.OF._flds.iteritems():
+                    setattr(p[i], fname, rffi.cast(TP, r.random_integer()))
         return p
 
     def get_index(self, length, r):
@@ -155,8 +180,16 @@
                     dic[fieldname] = getattr(p, fieldname)
         else:
             assert isinstance(S, lltype.Array)
-            for i in range(len(p)):
-                dic[i] = p[i]
+            if isinstance(S.OF, lltype.Struct):
+                for i in range(len(p)):
+                    item = p[i]
+                    s1 = {}
+                    for fieldname in S.OF._names:
+                        s1[fieldname] = getattr(item, fieldname)
+                    dic[i] = s1
+            else:
+                for i in range(len(p)):
+                    dic[i] = p[i]
         return dic
 
     def print_loop_prebuilt(self, names, writevar, s):
@@ -220,7 +253,7 @@
 
 class GetFieldOperation(test_random.AbstractOperation):
     def field_descr(self, builder, r):
-        v, S = builder.get_structptr_var(r)
+        v, S = builder.get_structptr_var(r, )
         names = S._names
         if names[0] == 'parent':
             names = names[1:]
@@ -239,6 +272,28 @@
                 continue
             break
 
+class GetInteriorFieldOperation(test_random.AbstractOperation):
+    def field_descr(self, builder, r):
+        v, A = builder.get_structptr_var(r, type=lltype.Array,
+                                         array_of_structs=True)
+        array = v.getref(lltype.Ptr(A))
+        v_index = builder.get_index(len(array), r)
+        name = r.choice(A.OF._names)
+        descr = builder.cpu.interiorfielddescrof(A, name)
+        descr._random_info = 'cpu.interiorfielddescrof(%s, %r)' % (A.OF._name,
+                                                                   name)
+        TYPE = getattr(A.OF, name)
+        return v, v_index, descr, TYPE
+
+    def produce_into(self, builder, r):
+        while True:
+            try:
+                v, v_index, descr, _ = self.field_descr(builder, r)
+                self.put(builder, [v, v_index], descr)
+            except lltype.UninitializedMemoryAccess:
+                continue
+            break
+
 class SetFieldOperation(GetFieldOperation):
     def produce_into(self, builder, r):
         v, descr, TYPE = self.field_descr(builder, r)
@@ -251,6 +306,18 @@
                 break
         builder.do(self.opnum, [v, w], descr)
 
+class SetInteriorFieldOperation(GetInteriorFieldOperation):
+    def produce_into(self, builder, r):
+        v, v_index, descr, TYPE = self.field_descr(builder, r)
+        while True:
+            if r.random() < 0.3:
+                w = ConstInt(r.random_integer())
+            else:
+                w = r.choice(builder.intvars)
+            if rffi.cast(lltype.Signed, rffi.cast(TYPE, w.value)) == w.value:
+                break
+        builder.do(self.opnum, [v, v_index, w], descr)
+
 class NewOperation(test_random.AbstractOperation):
     def size_descr(self, builder, S):
         descr = builder.cpu.sizeof(S)
@@ -306,7 +373,7 @@
 
 class NewArrayOperation(ArrayOperation):
     def produce_into(self, builder, r):
-        A = builder.get_random_array_type(r)
+        A = builder.get_random_array_type(r, can_be_array_of_struct=True)
         v_size = builder.get_index(300, r)
         v_ptr = builder.do(self.opnum, [v_size], self.array_descr(builder, A))
         builder.ptrvars.append((v_ptr, A))
@@ -586,7 +653,9 @@
 for i in range(4):      # make more common
     OPERATIONS.append(GetFieldOperation(rop.GETFIELD_GC))
     OPERATIONS.append(GetFieldOperation(rop.GETFIELD_GC))
+    OPERATIONS.append(GetInteriorFieldOperation(rop.GETINTERIORFIELD_GC))
     OPERATIONS.append(SetFieldOperation(rop.SETFIELD_GC))
+    OPERATIONS.append(SetInteriorFieldOperation(rop.SETINTERIORFIELD_GC))
     OPERATIONS.append(NewOperation(rop.NEW))
     OPERATIONS.append(NewOperation(rop.NEW_WITH_VTABLE))
 
diff --git a/pypy/jit/backend/test/test_random.py b/pypy/jit/backend/test/test_random.py
--- a/pypy/jit/backend/test/test_random.py
+++ b/pypy/jit/backend/test/test_random.py
@@ -595,6 +595,10 @@
             for name, value in fields.items():
                 if isinstance(name, str):
                     setattr(container, name, value)
+                elif isinstance(value, dict):
+                    item = container.getitem(name)
+                    for key1, value1 in value.items():
+                        setattr(item, key1, value1)
                 else:
                     container.setitem(name, value)
 
diff --git a/pypy/jit/backend/x86/assembler.py b/pypy/jit/backend/x86/assembler.py
--- a/pypy/jit/backend/x86/assembler.py
+++ b/pypy/jit/backend/x86/assembler.py
@@ -1,7 +1,7 @@
 import sys, os
 from pypy.jit.backend.llsupport import symbolic
 from pypy.jit.backend.llsupport.asmmemmgr import MachineDataBlockWrapper
-from pypy.jit.metainterp.history import Const, Box, BoxInt, BoxPtr, BoxFloat
+from pypy.jit.metainterp.history import Const, Box, BoxInt, ConstInt
 from pypy.jit.metainterp.history import (AbstractFailDescr, INT, REF, FLOAT,
                                          LoopToken)
 from pypy.rpython.lltypesystem import lltype, rffi, rstr, llmemory
@@ -36,7 +36,6 @@
 from pypy.rlib import rgc
 from pypy.rlib.clibffi import FFI_DEFAULT_ABI
 from pypy.jit.backend.x86.jump import remap_frame_layout
-from pypy.jit.metainterp.history import ConstInt, BoxInt
 from pypy.jit.codewriter.effectinfo import EffectInfo
 from pypy.jit.codewriter import longlong
 
@@ -729,8 +728,8 @@
         # Also, make sure this is consistent with FRAME_FIXED_SIZE.
         self.mc.PUSH_r(ebp.value)
         self.mc.MOV_rr(ebp.value, esp.value)
-        for regloc in self.cpu.CALLEE_SAVE_REGISTERS:
-            self.mc.PUSH_r(regloc.value)
+        for loc in self.cpu.CALLEE_SAVE_REGISTERS:
+            self.mc.PUSH_r(loc.value)
 
         gcrootmap = self.cpu.gc_ll_descr.gcrootmap
         if gcrootmap and gcrootmap.is_shadow_stack:
@@ -994,7 +993,7 @@
         effectinfo = op.getdescr().get_extra_info()
         oopspecindex = effectinfo.oopspecindex
         genop_llong_list[oopspecindex](self, op, arglocs, resloc)
-        
+
     def regalloc_perform_math(self, op, arglocs, resloc):
         effectinfo = op.getdescr().get_extra_info()
         oopspecindex = effectinfo.oopspecindex
@@ -1277,8 +1276,8 @@
     genop_int_ne = _cmpop("NE", "NE")
     genop_int_gt = _cmpop("G", "L")
     genop_int_ge = _cmpop("GE", "LE")
-    genop_ptr_eq = genop_int_eq
-    genop_ptr_ne = genop_int_ne
+    genop_ptr_eq = genop_instance_ptr_eq = genop_int_eq
+    genop_ptr_ne = genop_instance_ptr_ne = genop_int_ne
 
     genop_float_lt = _cmpop_float('B', 'A')
     genop_float_le = _cmpop_float('BE', 'AE')
@@ -1298,8 +1297,8 @@
     genop_guard_int_ne = _cmpop_guard("NE", "NE", "E", "E")
     genop_guard_int_gt = _cmpop_guard("G", "L", "LE", "GE")
     genop_guard_int_ge = _cmpop_guard("GE", "LE", "L", "G")
-    genop_guard_ptr_eq = genop_guard_int_eq
-    genop_guard_ptr_ne = genop_guard_int_ne
+    genop_guard_ptr_eq = genop_guard_instance_ptr_eq = genop_guard_int_eq
+    genop_guard_ptr_ne = genop_guard_instance_ptr_ne = genop_guard_int_ne
 
     genop_guard_uint_gt = _cmpop_guard("A", "B", "BE", "AE")
     genop_guard_uint_lt = _cmpop_guard("B", "A", "AE", "BE")
@@ -1311,7 +1310,7 @@
     genop_guard_float_eq = _cmpop_guard_float("E", "E", "NE","NE")
     genop_guard_float_gt = _cmpop_guard_float("A", "B", "BE","AE")
     genop_guard_float_ge = _cmpop_guard_float("AE","BE", "B", "A")
-    
+
     def genop_math_sqrt(self, op, arglocs, resloc):
         self.mc.SQRTSD(arglocs[0], resloc)
 
@@ -1387,7 +1386,8 @@
 
     def genop_same_as(self, op, arglocs, resloc):
         self.mov(arglocs[0], resloc)
-    #genop_cast_ptr_to_int = genop_same_as
+    genop_cast_ptr_to_int = genop_same_as
+    genop_cast_int_to_ptr = genop_same_as
 
     def genop_int_mod(self, op, arglocs, resloc):
         if IS_X86_32:
@@ -1596,12 +1596,44 @@
     genop_getarrayitem_gc_pure = genop_getarrayitem_gc
     genop_getarrayitem_raw = genop_getarrayitem_gc
 
+    def _get_interiorfield_addr(self, temp_loc, index_loc, itemsize_loc,
+                                base_loc, ofs_loc):
+        assert isinstance(itemsize_loc, ImmedLoc)
+        if isinstance(index_loc, ImmedLoc):
+            temp_loc = imm(index_loc.value * itemsize_loc.value)
+        else:
+            # XXX should not use IMUL in most cases
+            assert isinstance(temp_loc, RegLoc)
+            assert isinstance(index_loc, RegLoc)
+            assert not temp_loc.is_xmm
+            self.mc.IMUL_rri(temp_loc.value, index_loc.value,
+                             itemsize_loc.value)
+        assert isinstance(ofs_loc, ImmedLoc)
+        return AddressLoc(base_loc, temp_loc, 0, ofs_loc.value)
+
+    def genop_getinteriorfield_gc(self, op, arglocs, resloc):
+        (base_loc, ofs_loc, itemsize_loc, fieldsize_loc,
+            index_loc, temp_loc, sign_loc) = arglocs
+        src_addr = self._get_interiorfield_addr(temp_loc, index_loc,
+                                                itemsize_loc, base_loc,
+                                                ofs_loc)
+        self.load_from_mem(resloc, src_addr, fieldsize_loc, sign_loc)
+
+
     def genop_discard_setfield_gc(self, op, arglocs):
         base_loc, ofs_loc, size_loc, value_loc = arglocs
         assert isinstance(size_loc, ImmedLoc)
         dest_addr = AddressLoc(base_loc, ofs_loc)
         self.save_into_mem(dest_addr, value_loc, size_loc)
 
+    def genop_discard_setinteriorfield_gc(self, op, arglocs):
+        (base_loc, ofs_loc, itemsize_loc, fieldsize_loc,
+            index_loc, temp_loc, value_loc) = arglocs
+        dest_addr = self._get_interiorfield_addr(temp_loc, index_loc,
+                                                 itemsize_loc, base_loc,
+                                                 ofs_loc)
+        self.save_into_mem(dest_addr, value_loc, fieldsize_loc)
+
     def genop_discard_setarrayitem_gc(self, op, arglocs):
         base_loc, ofs_loc, value_loc, size_loc, baseofs = arglocs
         assert isinstance(baseofs, ImmedLoc)
diff --git a/pypy/jit/backend/x86/regalloc.py b/pypy/jit/backend/x86/regalloc.py
--- a/pypy/jit/backend/x86/regalloc.py
+++ b/pypy/jit/backend/x86/regalloc.py
@@ -7,7 +7,7 @@
                                          ResOperation, BoxPtr, ConstFloat,
                                          BoxFloat, LoopToken, INT, REF, FLOAT)
 from pypy.jit.backend.x86.regloc import *
-from pypy.rpython.lltypesystem import lltype, ll2ctypes, rffi, rstr
+from pypy.rpython.lltypesystem import lltype, rffi, rstr
 from pypy.rlib.objectmodel import we_are_translated
 from pypy.rlib import rgc
 from pypy.jit.backend.llsupport import symbolic
@@ -17,11 +17,12 @@
 from pypy.jit.metainterp.resoperation import rop
 from pypy.jit.backend.llsupport.descr import BaseFieldDescr, BaseArrayDescr
 from pypy.jit.backend.llsupport.descr import BaseCallDescr, BaseSizeDescr
+from pypy.jit.backend.llsupport.descr import InteriorFieldDescr
 from pypy.jit.backend.llsupport.regalloc import FrameManager, RegisterManager,\
      TempBox
 from pypy.jit.backend.x86.arch import WORD, FRAME_FIXED_SIZE
 from pypy.jit.backend.x86.arch import IS_X86_32, IS_X86_64, MY_COPY_OF_REGS
-from pypy.rlib.rarithmetic import r_longlong, r_uint
+from pypy.rlib.rarithmetic import r_longlong
 
 class X86RegisterManager(RegisterManager):
 
@@ -433,7 +434,7 @@
             if self.can_merge_with_next_guard(op, i, operations):
                 oplist_with_guard[op.getopnum()](self, op, operations[i + 1])
                 i += 1
-            elif not we_are_translated() and op.getopnum() == -124: 
+            elif not we_are_translated() and op.getopnum() == -124:
                 self._consider_force_spill(op)
             else:
                 oplist[op.getopnum()](self, op)
@@ -650,8 +651,8 @@
     consider_uint_lt = _consider_compop
     consider_uint_le = _consider_compop
     consider_uint_ge = _consider_compop
-    consider_ptr_eq = _consider_compop
-    consider_ptr_ne = _consider_compop
+    consider_ptr_eq = consider_instance_ptr_eq = _consider_compop
+    consider_ptr_ne = consider_instance_ptr_ne = _consider_compop
 
     def _consider_float_op(self, op):
         loc1 = self.xrm.loc(op.getarg(1))
@@ -815,7 +816,7 @@
         save_all_regs = guard_not_forced_op is not None
         self.xrm.before_call(force_store, save_all_regs=save_all_regs)
         if not save_all_regs:
-            gcrootmap = gc_ll_descr = self.assembler.cpu.gc_ll_descr.gcrootmap
+            gcrootmap = self.assembler.cpu.gc_ll_descr.gcrootmap
             if gcrootmap and gcrootmap.is_shadow_stack:
                 save_all_regs = 2
         self.rm.before_call(force_store, save_all_regs=save_all_regs)
@@ -972,74 +973,27 @@
             return self._call(op, arglocs)
 
     def consider_newstr(self, op):
-        gc_ll_descr = self.assembler.cpu.gc_ll_descr
-        if gc_ll_descr.get_funcptr_for_newstr is not None:
-            # framework GC
-            loc = self.loc(op.getarg(0))
-            return self._call(op, [loc])
-        # boehm GC (XXX kill the following code at some point)
-        ofs_items, itemsize, ofs = symbolic.get_array_token(rstr.STR, self.translate_support_code)
-        assert itemsize == 1
-        return self._malloc_varsize(ofs_items, ofs, 0, op.getarg(0),
-                                    op.result)
+        loc = self.loc(op.getarg(0))
+        return self._call(op, [loc])
 
     def consider_newunicode(self, op):
-        gc_ll_descr = self.assembler.cpu.gc_ll_descr
-        if gc_ll_descr.get_funcptr_for_newunicode is not None:
-            # framework GC
-            loc = self.loc(op.getarg(0))
-            return self._call(op, [loc])
-        # boehm GC (XXX kill the following code at some point)
-        ofs_items, _, ofs = symbolic.get_array_token(rstr.UNICODE,
-                                                   self.translate_support_code)
-        scale = self._get_unicode_item_scale()
-        return self._malloc_varsize(ofs_items, ofs, scale, op.getarg(0),
-                                    op.result)
-
-    def _malloc_varsize(self, ofs_items, ofs_length, scale, v, res_v):
-        # XXX kill this function at some point
-        if isinstance(v, Box):
-            loc = self.rm.make_sure_var_in_reg(v, [v])
-            tempbox = TempBox()
-            other_loc = self.rm.force_allocate_reg(tempbox, [v])
-            self.assembler.load_effective_addr(loc, ofs_items,scale, other_loc)
-        else:
-            tempbox = None
-            other_loc = imm(ofs_items + (v.getint() << scale))
-        self._call(ResOperation(rop.NEW, [], res_v),
-                   [other_loc], [v])
-        loc = self.rm.make_sure_var_in_reg(v, [res_v])
-        assert self.loc(res_v) == eax
-        # now we have to reload length to some reasonable place
-        self.rm.possibly_free_var(v)
-        if tempbox is not None:
-            self.rm.possibly_free_var(tempbox)
-        self.PerformDiscard(ResOperation(rop.SETFIELD_GC, [None, None], None),
-                            [eax, imm(ofs_length), imm(WORD), loc])
+        loc = self.loc(op.getarg(0))
+        return self._call(op, [loc])
 
     def consider_new_array(self, op):
         gc_ll_descr = self.assembler.cpu.gc_ll_descr
-        if gc_ll_descr.get_funcptr_for_newarray is not None:
-            # framework GC
-            box_num_elem = op.getarg(0)
-            if isinstance(box_num_elem, ConstInt):
-                num_elem = box_num_elem.value
-                if gc_ll_descr.can_inline_malloc_varsize(op.getdescr(),
-                                                         num_elem):
-                    self.fastpath_malloc_varsize(op, op.getdescr(), num_elem)
-                    return
-            args = self.assembler.cpu.gc_ll_descr.args_for_new_array(
-                op.getdescr())
-            arglocs = [imm(x) for x in args]
-            arglocs.append(self.loc(box_num_elem))
-            self._call(op, arglocs)
-            return
-        # boehm GC (XXX kill the following code at some point)
-        itemsize, basesize, ofs_length, _, _ = (
-            self._unpack_arraydescr(op.getdescr()))
-        scale_of_field = _get_scale(itemsize)
-        self._malloc_varsize(basesize, ofs_length, scale_of_field,
-                             op.getarg(0), op.result)
+        box_num_elem = op.getarg(0)
+        if isinstance(box_num_elem, ConstInt):
+            num_elem = box_num_elem.value
+            if gc_ll_descr.can_inline_malloc_varsize(op.getdescr(),
+                                                     num_elem):
+                self.fastpath_malloc_varsize(op, op.getdescr(), num_elem)
+                return
+        args = self.assembler.cpu.gc_ll_descr.args_for_new_array(
+            op.getdescr())
+        arglocs = [imm(x) for x in args]
+        arglocs.append(self.loc(box_num_elem))
+        self._call(op, arglocs)
 
     def _unpack_arraydescr(self, arraydescr):
         assert isinstance(arraydescr, BaseArrayDescr)
@@ -1058,6 +1012,16 @@
         sign = fielddescr.is_field_signed()
         return imm(ofs), imm(size), ptr, sign
 
+    def _unpack_interiorfielddescr(self, descr):
+        assert isinstance(descr, InteriorFieldDescr)
+        arraydescr = descr.arraydescr
+        ofs = arraydescr.get_base_size(self.translate_support_code)
+        itemsize = arraydescr.get_item_size(self.translate_support_code)
+        fieldsize = descr.fielddescr.get_field_size(self.translate_support_code)
+        sign = descr.fielddescr.is_field_signed()
+        ofs += descr.fielddescr.offset
+        return imm(ofs), imm(itemsize), imm(fieldsize), sign
+
     def consider_setfield_gc(self, op):
         ofs_loc, size_loc, _, _ = self._unpack_fielddescr(op.getdescr())
         assert isinstance(size_loc, ImmedLoc)
@@ -1074,6 +1038,35 @@
 
     consider_setfield_raw = consider_setfield_gc
 
+    def consider_setinteriorfield_gc(self, op):
+        t = self._unpack_interiorfielddescr(op.getdescr())
+        ofs, itemsize, fieldsize, _ = t
+        args = op.getarglist()
+        if fieldsize.value == 1:
+            need_lower_byte = True
+        else:
+            need_lower_byte = False
+        box_base, box_index, box_value = args
+        base_loc = self.rm.make_sure_var_in_reg(box_base, args)
+        index_loc = self.rm.make_sure_var_in_reg(box_index, args)
+        value_loc = self.make_sure_var_in_reg(box_value, args,
+                                              need_lower_byte=need_lower_byte)
+        # If 'index_loc' is not an immediate, then we need a 'temp_loc' that
+        # is a register whose value will be destroyed.  It's fine to destroy
+        # the same register as 'index_loc', but not the other ones.
+        self.rm.possibly_free_var(box_index)
+        if not isinstance(index_loc, ImmedLoc):
+            tempvar = TempBox()
+            temp_loc = self.rm.force_allocate_reg(tempvar, [box_base,
+                                                            box_value])
+            self.rm.possibly_free_var(tempvar)
+        else:
+            temp_loc = None
+        self.rm.possibly_free_var(box_base)
+        self.possibly_free_var(box_value)
+        self.PerformDiscard(op, [base_loc, ofs, itemsize, fieldsize,
+                                 index_loc, temp_loc, value_loc])
+
     def consider_strsetitem(self, op):
         args = op.getarglist()
         base_loc = self.rm.make_sure_var_in_reg(op.getarg(0), args)
@@ -1135,6 +1128,36 @@
     consider_getarrayitem_raw = consider_getarrayitem_gc
     consider_getarrayitem_gc_pure = consider_getarrayitem_gc
 
+    def consider_getinteriorfield_gc(self, op):
+        t = self._unpack_interiorfielddescr(op.getdescr())
+        ofs, itemsize, fieldsize, sign = t
+        if sign:
+            sign_loc = imm1
+        else:
+            sign_loc = imm0
+        args = op.getarglist()
+        base_loc = self.rm.make_sure_var_in_reg(op.getarg(0), args)
+        index_loc = self.rm.make_sure_var_in_reg(op.getarg(1), args)
+        # 'base' and 'index' are put in two registers (or one if 'index'
+        # is an immediate).  'result' can be in the same register as
+        # 'index' but must be in a different register than 'base'.
+        self.rm.possibly_free_var(op.getarg(1))
+        result_loc = self.force_allocate_reg(op.result, [op.getarg(0)])
+        assert isinstance(result_loc, RegLoc)
+        # two cases: 1) if result_loc is a normal register, use it as temp_loc
+        if not result_loc.is_xmm:
+            temp_loc = result_loc
+        else:
+            # 2) if result_loc is an xmm register, we (likely) need another
+            # temp_loc that is a normal register.  It can be in the same
+            # register as 'index' but not 'base'.
+            tempvar = TempBox()
+            temp_loc = self.rm.force_allocate_reg(tempvar, [op.getarg(0)])
+            self.rm.possibly_free_var(tempvar)
+        self.rm.possibly_free_var(op.getarg(0))
+        self.Perform(op, [base_loc, ofs, itemsize, fieldsize,
+                          index_loc, temp_loc, sign_loc], result_loc)
+
     def consider_int_is_true(self, op, guard_op):
         # doesn't need arg to be in a register
         argloc = self.loc(op.getarg(0))
@@ -1152,7 +1175,8 @@
         self.possibly_free_var(op.getarg(0))
         resloc = self.force_allocate_reg(op.result)
         self.Perform(op, [argloc], resloc)
-    #consider_cast_ptr_to_int = consider_same_as
+    consider_cast_ptr_to_int = consider_same_as
+    consider_cast_int_to_ptr = consider_same_as
 
     def consider_strlen(self, op):
         args = op.getarglist()
@@ -1240,7 +1264,6 @@
         self.rm.possibly_free_var(srcaddr_box)
 
     def _gen_address_inside_string(self, baseloc, ofsloc, resloc, is_unicode):
-        cpu = self.assembler.cpu
         if is_unicode:
             ofs_items, _, _ = symbolic.get_array_token(rstr.UNICODE,
                                                   self.translate_support_code)
@@ -1299,7 +1322,7 @@
         tmpreg = X86RegisterManager.all_regs[0]
         tmploc = self.rm.force_allocate_reg(box, selected_reg=tmpreg)
         xmmtmp = X86XMMRegisterManager.all_regs[0]
-        xmmtmploc = self.xrm.force_allocate_reg(box1, selected_reg=xmmtmp)
+        self.xrm.force_allocate_reg(box1, selected_reg=xmmtmp)
         # Part about non-floats
         # XXX we don't need a copy, we only just the original list
         src_locations1 = [self.loc(op.getarg(i)) for i in range(op.numargs())
@@ -1379,7 +1402,7 @@
     return lambda self, op: fn(self, op, None)
 
 def is_comparison_or_ovf_op(opnum):
-    from pypy.jit.metainterp.resoperation import opclasses, AbstractResOp
+    from pypy.jit.metainterp.resoperation import opclasses
     cls = opclasses[opnum]
     # hack hack: in theory they are instance method, but they don't use
     # any instance field, we can use a fake object
diff --git a/pypy/jit/backend/x86/test/test_del.py b/pypy/jit/backend/x86/test/test_del.py
--- a/pypy/jit/backend/x86/test/test_del.py
+++ b/pypy/jit/backend/x86/test/test_del.py
@@ -1,5 +1,4 @@
 
-import py
 from pypy.jit.backend.x86.test.test_basic import Jit386Mixin
 from pypy.jit.metainterp.test.test_del import DelTests
 
diff --git a/pypy/jit/backend/x86/test/test_dict.py b/pypy/jit/backend/x86/test/test_dict.py
new file mode 100644
--- /dev/null
+++ b/pypy/jit/backend/x86/test/test_dict.py
@@ -0,0 +1,9 @@
+
+from pypy.jit.backend.x86.test.test_basic import Jit386Mixin
+from pypy.jit.metainterp.test.test_dict import DictTests
+
+
+class TestDict(Jit386Mixin, DictTests):
+    # for the individual tests see
+    # ====> ../../../metainterp/test/test_dict.py
+    pass
diff --git a/pypy/jit/backend/x86/test/test_runner.py b/pypy/jit/backend/x86/test/test_runner.py
--- a/pypy/jit/backend/x86/test/test_runner.py
+++ b/pypy/jit/backend/x86/test/test_runner.py
@@ -31,7 +31,7 @@
 
     # for the individual tests see
     # ====> ../../test/runner_test.py
-    
+
     def setup_method(self, meth):
         self.cpu = CPU(rtyper=None, stats=FakeStats())
         self.cpu.setup_once()
@@ -69,22 +69,16 @@
 
     def test_allocations(self):
         from pypy.rpython.lltypesystem import rstr
-        
+
         allocs = [None]
         all = []
+        orig_new = self.cpu.gc_ll_descr.funcptr_for_new
         def f(size):
             allocs.insert(0, size)
-            buf = ctypes.create_string_buffer(size)
-            all.append(buf)
-            return ctypes.cast(buf, ctypes.c_void_p).value
-        func = ctypes.CFUNCTYPE(ctypes.c_int, ctypes.c_int)(f)
-        addr = ctypes.cast(func, ctypes.c_void_p).value
-        # ctypes produces an unsigned value. We need it to be signed for, eg,
-        # relative addressing to work properly.
-        addr = rffi.cast(lltype.Signed, addr)
-        
+            return orig_new(size)
+
         self.cpu.assembler.setup_once()
-        self.cpu.assembler.malloc_func_addr = addr
+        self.cpu.gc_ll_descr.funcptr_for_new = f
         ofs = symbolic.get_field_token(rstr.STR, 'chars', False)[0]
 
         res = self.execute_operation(rop.NEWSTR, [ConstInt(7)], 'ref')
@@ -108,7 +102,7 @@
         res = self.execute_operation(rop.NEW_ARRAY, [ConstInt(10)],
                                          'ref', descr)
         assert allocs[0] == 10*WORD + ofs + WORD
-        resbuf = self._resbuf(res)            
+        resbuf = self._resbuf(res)
         assert resbuf[ofs/WORD] == 10
 
         # ------------------------------------------------------------
@@ -116,7 +110,7 @@
         res = self.execute_operation(rop.NEW_ARRAY, [BoxInt(10)],
                                          'ref', descr)
         assert allocs[0] == 10*WORD + ofs + WORD
-        resbuf = self._resbuf(res)                        
+        resbuf = self._resbuf(res)
         assert resbuf[ofs/WORD] == 10
 
     def test_stringitems(self):
@@ -146,7 +140,7 @@
                                                      ConstInt(2), BoxInt(38)],
                                'void', descr)
         assert resbuf[itemsofs/WORD + 2] == 38
-        
+
         self.execute_operation(rop.SETARRAYITEM_GC, [res,
                                                      BoxInt(3), BoxInt(42)],
                                'void', descr)
@@ -167,7 +161,7 @@
                                                          BoxInt(2)],
                                    'int', descr)
         assert r.value == 38
-        
+
         r = self.execute_operation(rop.GETARRAYITEM_GC, [res, BoxInt(3)],
                                    'int', descr)
         assert r.value == 42
@@ -226,7 +220,7 @@
         self.execute_operation(rop.SETFIELD_GC, [res, BoxInt(1234)], 'void', ofs_i)
         i = self.execute_operation(rop.GETFIELD_GC, [res], 'int', ofs_i)
         assert i.value == 1234
-        
+
         #u = self.execute_operation(rop.GETFIELD_GC, [res, ofs_u], 'int')
         #assert u.value == 5
         self.execute_operation(rop.SETFIELD_GC, [res, ConstInt(1)], 'void',
@@ -299,7 +293,7 @@
                     else:
                         assert result != execute(self.cpu, None,
                                                  op, None, b).value
-                    
+
 
     def test_stuff_followed_by_guard(self):
         boxes = [(BoxInt(1), BoxInt(0)),
@@ -523,7 +517,7 @@
     def test_debugger_on(self):
         from pypy.tool.logparser import parse_log_file, extract_category
         from pypy.rlib import debug
-        
+
         loop = """
         [i0]
         debug_merge_point('xyz', 0)
diff --git a/pypy/jit/backend/x86/tool/viewcode.py b/pypy/jit/backend/x86/tool/viewcode.py
--- a/pypy/jit/backend/x86/tool/viewcode.py
+++ b/pypy/jit/backend/x86/tool/viewcode.py
@@ -9,7 +9,12 @@
 """
 
 import autopath
-import operator, sys, os, re, py, new
+import new
+import operator
+import py
+import re
+import sys
+import subprocess
 from bisect import bisect_left
 
 # don't use pypy.tool.udir here to avoid removing old usessions which
@@ -44,14 +49,16 @@
     f = open(tmpfile, 'wb')
     f.write(data)
     f.close()
-    g = os.popen(objdump % {
+    p = subprocess.Popen(objdump % {
         'file': tmpfile,
         'origin': originaddr,
         'backend': objdump_backend_option[backend_name],
-    }, 'r')
-    result = g.readlines()
-    g.close()
-    lines = result[6:]   # drop some objdump cruft
+    }, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    stdout, stderr = p.communicate()
+    assert not p.returncode, ('Encountered an error running objdump: %s' %
+                              stderr)
+    # drop some objdump cruft
+    lines = stdout.splitlines(True)[6:]     # drop some objdump cruft
     return format_code_dump_with_labels(originaddr, lines, label_list)
 
 def format_code_dump_with_labels(originaddr, lines, label_list):
@@ -85,8 +92,12 @@
     #
     print 'loading symbols from %s...' % (filename,)
     symbols = {}
-    g = os.popen(symbollister % filename, "r")
-    for line in g:
+    p = subprocess.Popen(symbollister % filename, shell=True,
+                         stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+    stdout, stderr = p.communicate()
+    assert not p.returncode, ('Encountered an error running nm: %s' %
+                              stderr)
+    for line in stdout.splitlines(True):
         match = re_symbolentry.match(line)
         if match:
             addr = long(match.group(1), 16)
@@ -94,7 +105,6 @@
             if name.startswith('pypy_g_'):
                 name = '\xb7' + name[7:]
             symbols[addr] = name
-    g.close()
     print '%d symbols found' % (len(symbols),)
     return symbols
 
diff --git a/pypy/jit/codewriter/jtransform.py b/pypy/jit/codewriter/jtransform.py
--- a/pypy/jit/codewriter/jtransform.py
+++ b/pypy/jit/codewriter/jtransform.py
@@ -52,9 +52,11 @@
         newoperations = []
         #
         def do_rename(var, var_or_const):
+            if var.concretetype is lltype.Void:
+                renamings[var] = Constant(None, lltype.Void)
+                return
             renamings[var] = var_or_const
-            if (isinstance(var_or_const, Constant)
-                and var.concretetype != lltype.Void):
+            if isinstance(var_or_const, Constant):
                 value = var_or_const.value
                 value = lltype._cast_whatever(var.concretetype, value)
                 renamings_constants[var] = Constant(value, var.concretetype)
@@ -441,6 +443,8 @@
     rewrite_op_gc_identityhash = _do_builtin_call
     rewrite_op_gc_id           = _do_builtin_call
     rewrite_op_uint_mod        = _do_builtin_call
+    rewrite_op_cast_float_to_uint = _do_builtin_call
+    rewrite_op_cast_uint_to_float = _do_builtin_call
 
     # ----------
     # getfield/setfield/mallocs etc.
@@ -455,6 +459,23 @@
             # the special return value None forces op.result to be considered
             # equal to op.args[0]
             return [op0, op1, None]
+        if (hints.get('promote_string') and
+            op.args[0].concretetype is not lltype.Void):
+            S = lltype.Ptr(rstr.STR)
+            assert op.args[0].concretetype == S
+            self._register_extra_helper(EffectInfo.OS_STREQ_NONNULL,
+                                        "str.eq_nonnull",
+                                        [S, S],
+                                        lltype.Signed,
+                                        EffectInfo.EF_ELIDABLE_CANNOT_RAISE)
+            descr, p = self.callcontrol.callinfocollection.callinfo_for_oopspec(
+                EffectInfo.OS_STREQ_NONNULL)
+            # XXX this is fairly ugly way of creating a constant,
+            #     however, callinfocollection has no better interface
+            c = Constant(p.adr.ptr, lltype.typeOf(p.adr.ptr))
+            op1 = SpaceOperation('str_guard_value', [op.args[0], c, descr],
+                                 op.result)
+            return [SpaceOperation('-live-', [], None), op1, None]
         else:
             log.WARNING('ignoring hint %r at %r' % (hints, self.graph))
 
@@ -718,29 +739,54 @@
         return SpaceOperation(opname, [op.args[0]], op.result)
 
     def rewrite_op_getinteriorfield(self, op):
-        # only supports strings and unicodes
         assert len(op.args) == 3
-        assert op.args[1].value == 'chars'
         optype = op.args[0].concretetype
         if optype == lltype.Ptr(rstr.STR):
             opname = "strgetitem"
+            return SpaceOperation(opname, [op.args[0], op.args[2]], op.result)
+        elif optype == lltype.Ptr(rstr.UNICODE):
+            opname = "unicodegetitem"
+            return SpaceOperation(opname, [op.args[0], op.args[2]], op.result)
         else:
-            assert optype == lltype.Ptr(rstr.UNICODE)
-            opname = "unicodegetitem"
-        return SpaceOperation(opname, [op.args[0], op.args[2]], op.result)
+            v_inst, v_index, c_field = op.args
+            if op.result.concretetype is lltype.Void:
+                return
+            # only GcArray of Struct supported
+            assert isinstance(v_inst.concretetype.TO, lltype.GcArray)
+            STRUCT = v_inst.concretetype.TO.OF
+            assert isinstance(STRUCT, lltype.Struct)
+            descr = self.cpu.interiorfielddescrof(v_inst.concretetype.TO,
+                                                  c_field.value)
+            args = [v_inst, v_index, descr]
+            kind = getkind(op.result.concretetype)[0]
+            return SpaceOperation('getinteriorfield_gc_%s' % kind, args,
+                                  op.result)
 
     def rewrite_op_setinteriorfield(self, op):
-        # only supports strings and unicodes
         assert len(op.args) == 4
-        assert op.args[1].value == 'chars'
         optype = op.args[0].concretetype
         if optype == lltype.Ptr(rstr.STR):
             opname = "strsetitem"
+            return SpaceOperation(opname, [op.args[0], op.args[2], op.args[3]],
+                                  op.result)
+        elif optype == lltype.Ptr(rstr.UNICODE):
+            opname = "unicodesetitem"
+            return SpaceOperation(opname, [op.args[0], op.args[2], op.args[3]],
+                                  op.result)
         else:
-            assert optype == lltype.Ptr(rstr.UNICODE)
-            opname = "unicodesetitem"
-        return SpaceOperation(opname, [op.args[0], op.args[2], op.args[3]],
-                              op.result)
+            v_inst, v_index, c_field, v_value = op.args
+            if v_value.concretetype is lltype.Void:
+                return
+            # only GcArray of Struct supported
+            assert isinstance(v_inst.concretetype.TO, lltype.GcArray)
+            STRUCT = v_inst.concretetype.TO.OF
+            assert isinstance(STRUCT, lltype.Struct)
+            descr = self.cpu.interiorfielddescrof(v_inst.concretetype.TO,
+                                                  c_field.value)
+            kind = getkind(v_value.concretetype)[0]
+            args = [v_inst, v_index, v_value, descr]
+            return SpaceOperation('setinteriorfield_gc_%s' % kind, args,
+                                  op.result)
 
     def _rewrite_equality(self, op, opname):
         arg0, arg1 = op.args
@@ -754,6 +800,9 @@
     def _is_gc(self, v):
         return getattr(getattr(v.concretetype, "TO", None), "_gckind", "?") == 'gc'
 
+    def _is_rclass_instance(self, v):
+        return lltype._castdepth(v.concretetype.TO, rclass.OBJECT) >= 0
+
     def _rewrite_cmp_ptrs(self, op):
         if self._is_gc(op.args[0]):
             return op
@@ -771,11 +820,21 @@
         return self._rewrite_equality(op, 'int_is_true')
 
     def rewrite_op_ptr_eq(self, op):
-        op1 = self._rewrite_equality(op, 'ptr_iszero')
+        prefix = ''
+        if self._is_rclass_instance(op.args[0]):
+            assert self._is_rclass_instance(op.args[1])
+            op = SpaceOperation('instance_ptr_eq', op.args, op.result)
+            prefix = 'instance_'
+        op1 = self._rewrite_equality(op, prefix + 'ptr_iszero')
         return self._rewrite_cmp_ptrs(op1)
 
     def rewrite_op_ptr_ne(self, op):
-        op1 = self._rewrite_equality(op, 'ptr_nonzero')
+        prefix = ''
+        if self._is_rclass_instance(op.args[0]):
+            assert self._is_rclass_instance(op.args[1])
+            op = SpaceOperation('instance_ptr_ne', op.args, op.result)
+            prefix = 'instance_'
+        op1 = self._rewrite_equality(op, prefix + 'ptr_nonzero')
         return self._rewrite_cmp_ptrs(op1)
 
     rewrite_op_ptr_iszero = _rewrite_cmp_ptrs
@@ -783,8 +842,11 @@
 
     def rewrite_op_cast_ptr_to_int(self, op):
         if self._is_gc(op.args[0]):
-            #return op
-            raise NotImplementedError("cast_ptr_to_int")
+            return op
+
+    def rewrite_op_cast_opaque_ptr(self, op):
+        # None causes the result of this op to get aliased to op.args[0]
+        return [SpaceOperation('mark_opaque_ptr', op.args, None), None]
 
     def rewrite_op_force_cast(self, op):
         v_arg = op.args[0]
@@ -805,26 +867,44 @@
         elif not float_arg and float_res:
             # some int -> some float
             ops = []
-            v1 = varoftype(lltype.Signed)
-            oplist = self.rewrite_operation(
-                SpaceOperation('force_cast', [v_arg], v1)
-            )
-            if oplist:
-                ops.extend(oplist)
+            v2 = varoftype(lltype.Float)
+            sizesign = rffi.size_and_sign(v_arg.concretetype)
+            if sizesign <= rffi.size_and_sign(lltype.Signed):
+                # cast from a type that fits in an int: either the size is
+                # smaller, or it is equal and it is not unsigned
+                v1 = varoftype(lltype.Signed)
+                oplist = self.rewrite_operation(
+                    SpaceOperation('force_cast', [v_arg], v1)
+                )
+                if oplist:
+                    ops.extend(oplist)
+                else:
+                    v1 = v_arg
+                op = self.rewrite_operation(
+                    SpaceOperation('cast_int_to_float', [v1], v2)
+                )
+                ops.append(op)
             else:
-                v1 = v_arg
-            v2 = varoftype(lltype.Float)
-            op = self.rewrite_operation(
-                SpaceOperation('cast_int_to_float', [v1], v2)
-            )
-            ops.append(op)
+                if sizesign == rffi.size_and_sign(lltype.Unsigned):
+                    opname = 'cast_uint_to_float'
+                elif sizesign == rffi.size_and_sign(lltype.SignedLongLong):
+                    opname = 'cast_longlong_to_float'
+                elif sizesign == rffi.size_and_sign(lltype.UnsignedLongLong):
+                    opname = 'cast_ulonglong_to_float'
+                else:
+                    raise AssertionError('cast_x_to_float: %r' % (sizesign,))
+                ops1 = self.rewrite_operation(
+                    SpaceOperation(opname, [v_arg], v2)
+                )
+                if not isinstance(ops1, list): ops1 = [ops1]
+                ops.extend(ops1)
             op2 = self.rewrite_operation(
                 SpaceOperation('force_cast', [v2], v_result)
             )
             if op2:
                 ops.append(op2)
             else:
-                op.result = v_result
+                ops[-1].result = v_result
             return ops
         elif float_arg and not float_res:
             # some float -> some int
@@ -837,18 +917,36 @@
                 ops.append(op1)
             else:
                 v1 = v_arg
-            v2 = varoftype(lltype.Signed)
-            op = self.rewrite_operation(
-                SpaceOperation('cast_float_to_int', [v1], v2)
-            )
-            ops.append(op)
-            oplist = self.rewrite_operation(
-                SpaceOperation('force_cast', [v2], v_result)
-            )
-            if oplist:
-                ops.extend(oplist)
+            sizesign = rffi.size_and_sign(v_result.concretetype)
+            if sizesign <= rffi.size_and_sign(lltype.Signed):
+                # cast to a type that fits in an int: either the size is
+                # smaller, or it is equal and it is not unsigned
+                v2 = varoftype(lltype.Signed)
+                op = self.rewrite_operation(
+                    SpaceOperation('cast_float_to_int', [v1], v2)
+                )
+                ops.append(op)
+                oplist = self.rewrite_operation(
+                    SpaceOperation('force_cast', [v2], v_result)
+                )
+                if oplist:
+                    ops.extend(oplist)
+                else:
+                    op.result = v_result
             else:
-                op.result = v_result
+                if sizesign == rffi.size_and_sign(lltype.Unsigned):
+                    opname = 'cast_float_to_uint'
+                elif sizesign == rffi.size_and_sign(lltype.SignedLongLong):
+                    opname = 'cast_float_to_longlong'
+                elif sizesign == rffi.size_and_sign(lltype.UnsignedLongLong):
+                    opname = 'cast_float_to_ulonglong'
+                else:
+                    raise AssertionError('cast_float_to_x: %r' % (sizesign,))
+                ops1 = self.rewrite_operation(
+                    SpaceOperation(opname, [v1], v_result)
+                )
+                if not isinstance(ops1, list): ops1 = [ops1]
+                ops.extend(ops1)
             return ops
         else:
             assert False
@@ -1054,8 +1152,6 @@
     # The new operation is optionally further processed by rewrite_operation().
     for _old, _new in [('bool_not', 'int_is_zero'),
                        ('cast_bool_to_float', 'cast_int_to_float'),
-                       ('cast_uint_to_float', 'cast_int_to_float'),
-                       ('cast_float_to_uint', 'cast_float_to_int'),
 
                        ('int_add_nonneg_ovf', 'int_add_ovf'),
                        ('keepalive', '-live-'),
@@ -1526,6 +1622,10 @@
     def rewrite_op_jit_force_virtual(self, op):
         return self._do_builtin_call(op)
 
+    def rewrite_op_jit_is_virtual(self, op):
+        raise Exception, (
+            "'vref.virtual' should not be used from jit-visible code")
+
     def rewrite_op_jit_force_virtualizable(self, op):
         # this one is for virtualizables
         vinfo = self.get_vinfo(op.args[0])
diff --git a/pypy/jit/codewriter/support.py b/pypy/jit/codewriter/support.py
--- a/pypy/jit/codewriter/support.py
+++ b/pypy/jit/codewriter/support.py
@@ -13,7 +13,6 @@
 from pypy.translator.simplify import get_funcobj
 from pypy.translator.unsimplify import split_block
 from pypy.objspace.flow.model import Constant
-from pypy import conftest
 from pypy.translator.translator import TranslationContext
 from pypy.annotation.policy import AnnotatorPolicy
 from pypy.annotation import model as annmodel
@@ -38,9 +37,11 @@
     return a.typeannotation(t)
 
 def annotate(func, values, inline=None, backendoptimize=True,
-             type_system="lltype"):
+             type_system="lltype", translationoptions={}):
     # build the normal ll graphs for ll_function
     t = TranslationContext()
+    for key, value in translationoptions.items():
+        setattr(t.config.translation, key, value)
     annpolicy = AnnotatorPolicy()
     annpolicy.allow_someobjects = False
     a = t.buildannotator(policy=annpolicy)
@@ -48,15 +49,13 @@
     a.build_types(func, argtypes, main_entry_point=True)
     rtyper = t.buildrtyper(type_system = type_system)
     rtyper.specialize()
-    if inline:
-        auto_inlining(t, threshold=inline)
+    #if inline:
+    #    auto_inlining(t, threshold=inline)
     if backendoptimize:
         from pypy.translator.backendopt.all import backend_optimizations
         backend_optimizations(t, inline_threshold=inline or 0,
                 remove_asserts=True, really_remove_asserts=True)
 
-    #if conftest.option.view:
-    #    t.view()
     return rtyper
 
 def getgraph(func, values):
@@ -232,6 +231,17 @@
     else:
         return x
 
+def _ll_1_cast_uint_to_float(x):
+    # XXX on 32-bit platforms, this should be done using cast_longlong_to_float
+    # (which is a residual call right now in the x86 backend)
+    return llop.cast_uint_to_float(lltype.Float, x)
+
+def _ll_1_cast_float_to_uint(x):
+    # XXX on 32-bit platforms, this should be done using cast_float_to_longlong
+    # (which is a residual call right now in the x86 backend)
+    return llop.cast_float_to_uint(lltype.Unsigned, x)
+
+
 # math support
 # ------------
 
@@ -456,6 +466,8 @@
         return LLtypeHelpers._dictnext_items(lltype.Ptr(RES), iter)
     _ll_1_dictiter_nextitems.need_result_type = True
 
+    _ll_1_dict_resize = ll_rdict.ll_dict_resize
+
     # ---------- strings and unicode ----------
 
     _ll_1_str_str2unicode = ll_rstr.LLHelpers.ll_str2unicode
diff --git a/pypy/jit/codewriter/test/test_flatten.py b/pypy/jit/codewriter/test/test_flatten.py
--- a/pypy/jit/codewriter/test/test_flatten.py
+++ b/pypy/jit/codewriter/test/test_flatten.py
@@ -8,7 +8,7 @@
 from pypy.rpython.lltypesystem import lltype, rclass, rstr
 from pypy.objspace.flow.model import SpaceOperation, Variable, Constant
 from pypy.translator.unsimplify import varoftype
-from pypy.rlib.rarithmetic import ovfcheck, r_uint
+from pypy.rlib.rarithmetic import ovfcheck, r_uint, r_longlong, r_ulonglong
 from pypy.rlib.jit import dont_look_inside, _we_are_jitted, JitDriver
 from pypy.rlib.objectmodel import keepalive_until_here
 from pypy.rlib import jit
@@ -70,7 +70,8 @@
         return 'residual'
     def getcalldescr(self, op, oopspecindex=None, extraeffect=None):
         try:
-            if 'cannot_raise' in op.args[0].value._obj.graph.name:
+            name = op.args[0].value._obj._name
+            if 'cannot_raise' in name or name.startswith('cast_'):
                 return self._descr_cannot_raise
         except AttributeError:
             pass
@@ -900,6 +901,67 @@
             int_return %i4
         """, transform=True)
 
+        def f(dbl):
+            return rffi.cast(rffi.UCHAR, dbl)
+        self.encoding_test(f, [12.456], """
+            cast_float_to_int %f0 -> %i0
+            int_and %i0, $255 -> %i1
+            int_return %i1
+        """, transform=True)
+
+        def f(dbl):
+            return rffi.cast(lltype.Unsigned, dbl)
+        self.encoding_test(f, [12.456], """
+            residual_call_irf_i $<* fn cast_float_to_uint>, <Descr>, I[], R[], F[%f0] -> %i0
+            int_return %i0
+        """, transform=True)
+
+        def f(i):
+            return rffi.cast(lltype.Float, chr(i))    # "char -> float"
+        self.encoding_test(f, [12], """
+            cast_int_to_float %i0 -> %f0
+            float_return %f0
+        """, transform=True)
+
+        def f(i):
+            return rffi.cast(lltype.Float, r_uint(i))    # "uint -> float"
+        self.encoding_test(f, [12], """
+            residual_call_irf_f $<* fn cast_uint_to_float>, <Descr>, I[%i0], R[], F[] -> %f0
+            float_return %f0
+        """, transform=True)
+
+        if not longlong.is_64_bit:
+            def f(dbl):
+                return rffi.cast(lltype.SignedLongLong, dbl)
+            self.encoding_test(f, [12.3], """
+                residual_call_irf_f $<* fn llong_from_float>, <Descr>, I[], R[], F[%f0] -> %f1
+                float_return %f1
+            """, transform=True)
+
+            def f(dbl):
+                return rffi.cast(lltype.UnsignedLongLong, dbl)
+            self.encoding_test(f, [12.3], """
+                residual_call_irf_f $<* fn ullong_from_float>, <Descr>, I[], R[], F[%f0] -> %f1
+                float_return %f1
+            """, transform=True)
+
+            def f(x):
+                ll = r_longlong(x)
+                return rffi.cast(lltype.Float, ll)
+            self.encoding_test(f, [12], """
+                residual_call_irf_f $<* fn llong_from_int>, <Descr>, I[%i0], R[], F[] -> %f0
+                residual_call_irf_f $<* fn llong_to_float>, <Descr>, I[], R[], F[%f0] -> %f1
+                float_return %f1
+            """, transform=True)
+
+            def f(x):
+                ll = r_ulonglong(x)
+                return rffi.cast(lltype.Float, ll)
+            self.encoding_test(f, [12], """
+                residual_call_irf_f $<* fn ullong_from_int>, <Descr>, I[%i0], R[], F[] -> %f0
+                residual_call_irf_f $<* fn ullong_u_to_float>, <Descr>, I[], R[], F[%f0] -> %f1
+                float_return %f1
+            """, transform=True)
 
     def test_direct_ptradd(self):
         from pypy.rpython.lltypesystem import rffi
diff --git a/pypy/jit/codewriter/test/test_jtransform.py b/pypy/jit/codewriter/test/test_jtransform.py
--- a/pypy/jit/codewriter/test/test_jtransform.py
+++ b/pypy/jit/codewriter/test/test_jtransform.py
@@ -1,14 +1,27 @@
-import py
 import random
+try:
+    from itertools import product
+except ImportError:
+    # Python 2.5, this is taken from the CPython docs, but simplified.
+    def product(*args):
+        # product('ABCD', 'xy') --> Ax Ay Bx By Cx Cy Dx Dy
+        # product(range(2), repeat=3) --> 000 001 010 011 100 101 110 111
+        pools = map(tuple, args)
+        result = [[]]
+        for pool in pools:
+            result = [x+[y] for x in result for y in pool]
+        for prod in result:
+            yield tuple(prod)
+
 from pypy.objspace.flow.model import FunctionGraph, Block, Link
 from pypy.objspace.flow.model import SpaceOperation, Variable, Constant
-from pypy.jit.codewriter.jtransform import Transformer
-from pypy.jit.metainterp.history import getkind
-from pypy.rpython.lltypesystem import lltype, llmemory, rclass, rstr, rlist
+from pypy.rpython.lltypesystem import lltype, llmemory, rclass, rstr
 from pypy.rpython.lltypesystem.module import ll_math
 from pypy.translator.unsimplify import varoftype
 from pypy.jit.codewriter import heaptracker, effectinfo
 from pypy.jit.codewriter.flatten import ListOfKind
+from pypy.jit.codewriter.jtransform import Transformer
+from pypy.jit.metainterp.history import getkind
 
 def const(x):
     return Constant(x, lltype.typeOf(x))
@@ -23,6 +36,8 @@
         return ('calldescr', FUNC, ARGS, RESULT)
     def fielddescrof(self, STRUCT, name):
         return ('fielddescr', STRUCT, name)
+    def interiorfielddescrof(self, ARRAY, name):
+        return ('interiorfielddescr', ARRAY, name)
     def arraydescrof(self, ARRAY):
         return FakeDescr(('arraydescr', ARRAY))
     def sizeof(self, STRUCT):
@@ -85,6 +100,12 @@
             if i == oopspecindex:
                 return True
         return False
+    def callinfo_for_oopspec(self, oopspecindex):
+        assert oopspecindex == effectinfo.EffectInfo.OS_STREQ_NONNULL
+        class c:
+            class adr:
+                ptr = 1
+        return ('calldescr', c)
 
 class FakeBuiltinCallControl:
     def __init__(self):
@@ -105,6 +126,7 @@
              EI.OS_STR2UNICODE:([PSTR], PUNICODE),
              EI.OS_STR_CONCAT: ([PSTR, PSTR], PSTR),
              EI.OS_STR_SLICE:  ([PSTR, INT, INT], PSTR),
+             EI.OS_STREQ_NONNULL:  ([PSTR, PSTR], INT),
              EI.OS_UNI_CONCAT: ([PUNICODE, PUNICODE], PUNICODE),
              EI.OS_UNI_SLICE:  ([PUNICODE, INT, INT], PUNICODE),
              EI.OS_UNI_EQUAL:  ([PUNICODE, PUNICODE], lltype.Bool),
@@ -254,26 +276,35 @@
             assert op1.result is None
 
 def test_calls():
-    for RESTYPE in [lltype.Signed, rclass.OBJECTPTR,
-                    lltype.Float, lltype.Void]:
-      for with_void in [False, True]:
-        for with_i in [False, True]:
-          for with_r in [False, True]:
-            for with_f in [False, True]:
-              ARGS = []
-              if with_void: ARGS += [lltype.Void, lltype.Void]
-              if with_i: ARGS += [lltype.Signed, lltype.Char]
-              if with_r: ARGS += [rclass.OBJECTPTR, lltype.Ptr(rstr.STR)]
-              if with_f: ARGS += [lltype.Float, lltype.Float]
-              random.shuffle(ARGS)
-              if RESTYPE == lltype.Float: with_f = True
-              if with_f: expectedkind = 'irf'   # all kinds
-              elif with_i: expectedkind = 'ir'  # integers and references
-              else: expectedkind = 'r'          # only references
-              yield residual_call_test, ARGS, RESTYPE, expectedkind
-              yield direct_call_test, ARGS, RESTYPE, expectedkind
-              yield indirect_residual_call_test, ARGS, RESTYPE, expectedkind
-              yield indirect_regular_call_test, ARGS, RESTYPE, expectedkind
+    for RESTYPE, with_void, with_i, with_r, with_f in product(
+        [lltype.Signed, rclass.OBJECTPTR, lltype.Float, lltype.Void],
+        [False, True],
+        [False, True],
+        [False, True],
+        [False, True],
+    ):
+        ARGS = []
+        if with_void:
+            ARGS += [lltype.Void, lltype.Void]
+        if with_i:
+            ARGS += [lltype.Signed, lltype.Char]
+        if with_r:
+            ARGS += [rclass.OBJECTPTR, lltype.Ptr(rstr.STR)]
+        if with_f:
+            ARGS += [lltype.Float, lltype.Float]
+        random.shuffle(ARGS)
+        if RESTYPE == lltype.Float:
+            with_f = True
+        if with_f:
+            expectedkind = 'irf'   # all kinds
+        elif with_i:
+            expectedkind = 'ir'  # integers and references
+        else:
+            expectedkind = 'r'          # only references
+        yield residual_call_test, ARGS, RESTYPE, expectedkind
+        yield direct_call_test, ARGS, RESTYPE, expectedkind
+        yield indirect_residual_call_test, ARGS, RESTYPE, expectedkind
+        yield indirect_regular_call_test, ARGS, RESTYPE, expectedkind
 
 def get_direct_call_op(argtypes, restype):
     FUNC = lltype.FuncType(argtypes, restype)
@@ -509,7 +540,7 @@
 
 def test_rename_on_links():
     v1 = Variable()
-    v2 = Variable()
+    v2 = Variable(); v2.concretetype = llmemory.Address
     v3 = Variable()
     block = Block([v1])
     block.operations = [SpaceOperation('cast_pointer', [v1], v2)]
@@ -545,10 +576,10 @@
         assert op1.args == [v2]
 
 def test_ptr_eq():
-    v1 = varoftype(rclass.OBJECTPTR)
-    v2 = varoftype(rclass.OBJECTPTR)
+    v1 = varoftype(lltype.Ptr(rstr.STR))
+    v2 = varoftype(lltype.Ptr(rstr.STR))
     v3 = varoftype(lltype.Bool)
-    c0 = const(lltype.nullptr(rclass.OBJECT))
+    c0 = const(lltype.nullptr(rstr.STR))
     #
     for opname, reducedname in [('ptr_eq', 'ptr_iszero'),
                                 ('ptr_ne', 'ptr_nonzero')]:
@@ -567,6 +598,31 @@
         assert op1.opname == reducedname
         assert op1.args == [v2]
 
+def test_instance_ptr_eq():
+    v1 = varoftype(rclass.OBJECTPTR)
+    v2 = varoftype(rclass.OBJECTPTR)
+    v3 = varoftype(lltype.Bool)
+    c0 = const(lltype.nullptr(rclass.OBJECT))
+
+    for opname, newopname, reducedname in [
+        ('ptr_eq', 'instance_ptr_eq', 'instance_ptr_iszero'),
+        ('ptr_ne', 'instance_ptr_ne', 'instance_ptr_nonzero')
+    ]:
+        op = SpaceOperation(opname, [v1, v2], v3)
+        op1 = Transformer().rewrite_operation(op)
+        assert op1.opname == newopname
+        assert op1.args == [v1, v2]
+
+        op = SpaceOperation(opname, [v1, c0], v3)
+        op1 = Transformer().rewrite_operation(op)
+        assert op1.opname == reducedname
+        assert op1.args == [v1]
+
+        op = SpaceOperation(opname, [c0, v1], v3)
+        op1 = Transformer().rewrite_operation(op)
+        assert op1.opname == reducedname
+        assert op1.args == [v1]
+
 def test_nongc_ptr_eq():
     v1 = varoftype(rclass.NONGCOBJECTPTR)
     v2 = varoftype(rclass.NONGCOBJECTPTR)
@@ -646,6 +702,22 @@
     assert op1.args == [v, v_index]
     assert op1.result == v_result
 
+def test_dict_getinteriorfield():
+    DICT = lltype.GcArray(lltype.Struct('ENTRY', ('v', lltype.Signed),
+                                        ('k', lltype.Signed)))
+    v = varoftype(lltype.Ptr(DICT))
+    i = varoftype(lltype.Signed)
+    v_result = varoftype(lltype.Signed)
+    op = SpaceOperation('getinteriorfield', [v, i, Constant('v', lltype.Void)],
+                        v_result)
+    op1 = Transformer(FakeCPU()).rewrite_operation(op)
+    assert op1.opname == 'getinteriorfield_gc_i'
+    assert op1.args == [v, i, ('interiorfielddescr', DICT, 'v')]
+    op = SpaceOperation('getinteriorfield', [v, i, Constant('v', lltype.Void)],
+                        Constant(None, lltype.Void))
+    op1 = Transformer(FakeCPU()).rewrite_operation(op)
+    assert op1 is None
+
 def test_str_setinteriorfield():
     v = varoftype(lltype.Ptr(rstr.STR))
     v_index = varoftype(lltype.Signed)
@@ -672,6 +744,23 @@
     assert op1.args == [v, v_index, v_newchr]
     assert op1.result == v_void
 
+def test_dict_setinteriorfield():
+    DICT = lltype.GcArray(lltype.Struct('ENTRY', ('v', lltype.Signed),
+                                        ('k', lltype.Signed)))
+    v = varoftype(lltype.Ptr(DICT))
+    i = varoftype(lltype.Signed)
+    v_void = varoftype(lltype.Void)
+    op = SpaceOperation('setinteriorfield', [v, i, Constant('v', lltype.Void),
+                                             i],
+                        v_void)
+    op1 = Transformer(FakeCPU()).rewrite_operation(op)
+    assert op1.opname == 'setinteriorfield_gc_i'
+    assert op1.args == [v, i, i, ('interiorfielddescr', DICT, 'v')]
+    op = SpaceOperation('setinteriorfield', [v, i, Constant('v', lltype.Void),
+                                             v_void], v_void)
+    op1 = Transformer(FakeCPU()).rewrite_operation(op)
+    assert not op1
+
 def test_promote_1():
     v1 = varoftype(lltype.Signed)
     v2 = varoftype(lltype.Signed)
@@ -821,6 +910,21 @@
     assert op1.args[2] == ListOfKind('ref', [v1, v2])
     assert op1.result == v3
 
+def test_str_promote():
+    PSTR = lltype.Ptr(rstr.STR)
+    v1 = varoftype(PSTR)
+    v2 = varoftype(PSTR)
+    op = SpaceOperation('hint',
+                        [v1, Constant({'promote_string': True}, lltype.Void)],
+                        v2)
+    tr = Transformer(FakeCPU(), FakeBuiltinCallControl())
+    op0, op1, _ = tr.rewrite_operation(op)
+    assert op1.opname == 'str_guard_value'
+    assert op1.args[0] == v1
+    assert op1.args[2] == 'calldescr'
+    assert op1.result == v2
+    assert op0.opname == '-live-'
+
 def test_unicode_concat():
     # test that the oopspec is present and correctly transformed
     PSTR = lltype.Ptr(rstr.UNICODE)
@@ -1024,3 +1128,16 @@
                         varoftype(lltype.Signed))
     tr = Transformer(None, None)
     raises(NotImplementedError, tr.rewrite_operation, op)
+
+def test_cast_opaque_ptr():
+    S = lltype.GcStruct("S", ("x", lltype.Signed))
+    v1 = varoftype(lltype.Ptr(S))
+    v2 = varoftype(lltype.Ptr(rclass.OBJECT))
+
+    op = SpaceOperation('cast_opaque_ptr', [v1], v2)
+    tr = Transformer()
+    [op1, op2] = tr.rewrite_operation(op)
+    assert op1.opname == 'mark_opaque_ptr'
+    assert op1.args == [v1]
+    assert op1.result is None
+    assert op2 is None
\ No newline at end of file
diff --git a/pypy/jit/metainterp/blackhole.py b/pypy/jit/metainterp/blackhole.py
--- a/pypy/jit/metainterp/blackhole.py
+++ b/pypy/jit/metainterp/blackhole.py
@@ -2,11 +2,10 @@
 from pypy.rlib.rtimer import read_timestamp
 from pypy.rlib.rarithmetic import intmask, LONG_BIT, r_uint, ovfcheck
 from pypy.rlib.objectmodel import we_are_translated
-from pypy.rlib.debug import debug_start, debug_stop
+from pypy.rlib.debug import debug_start, debug_stop, ll_assert
 from pypy.rlib.debug import make_sure_not_resized
 from pypy.rpython.lltypesystem import lltype, llmemory, rclass
 from pypy.rpython.lltypesystem.lloperation import llop
-from pypy.rpython.llinterp import LLException
 from pypy.jit.codewriter.jitcode import JitCode, SwitchDictDescr
 from pypy.jit.codewriter import heaptracker, longlong
 from pypy.jit.metainterp.jitexc import JitException, get_llexception, reraise
@@ -500,9 +499,25 @@
     @arguments("r", returns="i")
     def bhimpl_ptr_nonzero(a):
         return bool(a)
-    @arguments("r", returns="r")
-    def bhimpl_cast_opaque_ptr(a):
-        return a
+    @arguments("r", "r", returns="i")
+    def bhimpl_instance_ptr_eq(a, b):
+        return a == b
+    @arguments("r", "r", returns="i")
+    def bhimpl_instance_ptr_ne(a, b):
+        return a != b
+    @arguments("r", returns="i")
+    def bhimpl_cast_ptr_to_int(a):
+        i = lltype.cast_ptr_to_int(a)
+        ll_assert((i & 1) == 1, "bhimpl_cast_ptr_to_int: not an odd int")
+        return i
+    @arguments("i", returns="r")
+    def bhimpl_cast_int_to_ptr(i):
+        ll_assert((i & 1) == 1, "bhimpl_cast_int_to_ptr: not an odd int")
+        return lltype.cast_int_to_ptr(llmemory.GCREF, i)
+
+    @arguments("r")
+    def bhimpl_mark_opaque_ptr(a):
+        pass
 
     @arguments("i", returns="i")
     def bhimpl_int_copy(a):
@@ -523,6 +538,9 @@
     @arguments("f")
     def bhimpl_float_guard_value(a):
         pass
+    @arguments("r", "i", "d", returns="r")
+    def bhimpl_str_guard_value(a, i, d):
+        return a
 
     @arguments("self", "i")
     def bhimpl_int_push(self, a):
@@ -619,6 +637,9 @@
         a = longlong.getrealfloat(a)
         # note: we need to call int() twice to care for the fact that
         # int(-2147483648.0) returns a long :-(
+        # we could also call intmask() instead of the outermost int(), but
+        # it's probably better to explicitly crash (by getting a long) if a
+        # non-translated version tries to cast a too large float to an int.
         return int(int(a))
 
     @arguments("i", returns="f")
@@ -1142,6 +1163,26 @@
         array = cpu.bh_getfield_gc_r(vable, fdescr)
         return cpu.bh_arraylen_gc(adescr, array)
 
+    @arguments("cpu", "r", "i", "d", returns="i")
+    def bhimpl_getinteriorfield_gc_i(cpu, array, index, descr):
+        return cpu.bh_getinteriorfield_gc_i(array, index, descr)
+    @arguments("cpu", "r", "i", "d", returns="r")
+    def bhimpl_getinteriorfield_gc_r(cpu, array, index, descr):
+        return cpu.bh_getinteriorfield_gc_r(array, index, descr)
+    @arguments("cpu", "r", "i", "d", returns="f")
+    def bhimpl_getinteriorfield_gc_f(cpu, array, index, descr):
+        return cpu.bh_getinteriorfield_gc_f(array, index, descr)
+
+    @arguments("cpu", "r", "i", "d", "i")
+    def bhimpl_setinteriorfield_gc_i(cpu, array, index, descr, value):
+        cpu.bh_setinteriorfield_gc_i(array, index, descr, value)
+    @arguments("cpu", "r", "i", "d", "r")
+    def bhimpl_setinteriorfield_gc_r(cpu, array, index, descr, value):
+        cpu.bh_setinteriorfield_gc_r(array, index, descr, value)
+    @arguments("cpu", "r", "i", "d", "f")
+    def bhimpl_setinteriorfield_gc_f(cpu, array, index, descr, value):
+        cpu.bh_setinteriorfield_gc_f(array, index, descr, value)
+
     @arguments("cpu", "r", "d", returns="i")
     def bhimpl_getfield_gc_i(cpu, struct, fielddescr):
         return cpu.bh_getfield_gc_i(struct, fielddescr)
diff --git a/pypy/jit/metainterp/executor.py b/pypy/jit/metainterp/executor.py
--- a/pypy/jit/metainterp/executor.py
+++ b/pypy/jit/metainterp/executor.py
@@ -1,11 +1,8 @@
 """This implements pyjitpl's execution of operations.
 """
 
-import py
-from pypy.rpython.lltypesystem import lltype, llmemory, rstr
-from pypy.rpython.ootypesystem import ootype
-from pypy.rpython.lltypesystem.lloperation import llop
-from pypy.rlib.rarithmetic import ovfcheck, r_uint, intmask, r_longlong
+from pypy.rpython.lltypesystem import lltype, rstr
+from pypy.rlib.rarithmetic import ovfcheck, r_longlong
 from pypy.rlib.rtimer import read_timestamp
 from pypy.rlib.unroll import unrolling_iterable
 from pypy.jit.metainterp.history import BoxInt, BoxPtr, BoxFloat, check_descr
@@ -123,6 +120,29 @@
     else:
         cpu.bh_setarrayitem_raw_i(arraydescr, array, index, itembox.getint())
 
+def do_getinteriorfield_gc(cpu, _, arraybox, indexbox, descr):
+    array = arraybox.getref_base()
+    index = indexbox.getint()
+    if descr.is_pointer_field():
+        return BoxPtr(cpu.bh_getinteriorfield_gc_r(array, index, descr))
+    elif descr.is_float_field():
+        return BoxFloat(cpu.bh_getinteriorfield_gc_f(array, index, descr))
+    else:
+        return BoxInt(cpu.bh_getinteriorfield_gc_i(array, index, descr))
+
+def do_setinteriorfield_gc(cpu, _, arraybox, indexbox, valuebox, descr):
+    array = arraybox.getref_base()
+    index = indexbox.getint()
+    if descr.is_pointer_field():
+        cpu.bh_setinteriorfield_gc_r(array, index, descr,
+                                     valuebox.getref_base())
+    elif descr.is_float_field():
+        cpu.bh_setinteriorfield_gc_f(array, index, descr,
+                                     valuebox.getfloatstorage())
+    else:
+        cpu.bh_setinteriorfield_gc_i(array, index, descr,
+                                     valuebox.getint())
+
 def do_getfield_gc(cpu, _, structbox, fielddescr):
     struct = structbox.getref_base()
     if fielddescr.is_pointer_field():
diff --git a/pypy/jit/metainterp/graphpage.py b/pypy/jit/metainterp/graphpage.py
--- a/pypy/jit/metainterp/graphpage.py
+++ b/pypy/jit/metainterp/graphpage.py
@@ -12,8 +12,8 @@
     def get_display_text(self):
         return None
 
-def display_loops(loops, errmsg=None, highlight_loops=()):
-    graphs = [(loop, loop in highlight_loops) for loop in loops]    
+def display_loops(loops, errmsg=None, highlight_loops={}):
+    graphs = [(loop, highlight_loops.get(loop, 0)) for loop in loops]    
     for graph, highlight in graphs:
         for op in graph.get_operations():
             if is_interesting_guard(op):
@@ -65,8 +65,7 @@
     def add_graph(self, graph, highlight=False):
         graphindex = len(self.graphs)
         self.graphs.append(graph)
-        if highlight:
-            self.highlight_graphs[graph] = True
+        self.highlight_graphs[graph] = highlight
         for i, op in enumerate(graph.get_operations()):
             self.all_operations[op] = graphindex, i
 
@@ -126,10 +125,13 @@
             self.dotgen.emit('subgraph cluster%d {' % graphindex)
         label = graph.get_display_text()
         if label is not None:
-            if self.highlight_graphs.get(graph):
-                fillcolor = '#f084c2'
+            colorindex = self.highlight_graphs.get(graph, 0)
+            if colorindex == 1:
+                fillcolor = '#f084c2'    # highlighted graph
+            elif colorindex == 2:
+                fillcolor = '#808080'    # invalidated graph
             else:
-                fillcolor = '#84f0c2'
+                fillcolor = '#84f0c2'    # normal color
             self.dotgen.emit_node(graphname, shape="octagon",
                                   label=label, fillcolor=fillcolor)
             self.pendingedges.append((graphname,
diff --git a/pypy/jit/metainterp/heapcache.py b/pypy/jit/metainterp/heapcache.py
--- a/pypy/jit/metainterp/heapcache.py
+++ b/pypy/jit/metainterp/heapcache.py
@@ -29,29 +29,11 @@
         # cache the length of arrays
         self.length_cache = {}
 
-        # equivalences between boxes
-        self.equivalent = {}
-
-    def get_repr(self, box):
-        res = self.equivalent.get(box, box)
-        if res is not box:
-            res2 = self.get_repr(res)
-            # path compression
-            if res2 is not res:
-                self.equivalent[box] = res2
-                res = res2
-        return res
-
-    def same_boxes(self, box1, box2):
-        assert box1 not in self.equivalent
-        self.equivalent[box1] = self.get_repr(box2)
-
     def invalidate_caches(self, opnum, descr, argboxes):
         self.mark_escaped(opnum, argboxes)
         self.clear_caches(opnum, descr, argboxes)
 
     def mark_escaped(self, opnum, argboxes):
-        idx = 0
         if opnum == rop.SETFIELD_GC:
             assert len(argboxes) == 2
             box, valuebox = argboxes
@@ -59,8 +41,20 @@
                 self.dependencies.setdefault(box, []).append(valuebox)
             else:
                 self._escape(valuebox)
-        # GETFIELD_GC doesn't escape it's argument
-        elif opnum != rop.GETFIELD_GC:
+        elif opnum == rop.SETARRAYITEM_GC:
+            assert len(argboxes) == 3
+            box, indexbox, valuebox = argboxes
+            if self.is_unescaped(box) and self.is_unescaped(valuebox):
+                self.dependencies.setdefault(box, []).append(valuebox)
+            else:
+                self._escape(valuebox)
+        # GETFIELD_GC, MARK_OPAQUE_PTR, PTR_EQ, and PTR_NE don't escape their
+        # arguments
+        elif (opnum != rop.GETFIELD_GC and
+              opnum != rop.MARK_OPAQUE_PTR and
+              opnum != rop.PTR_EQ and
+              opnum != rop.PTR_NE):
+            idx = 0
             for box in argboxes:
                 # setarrayitem_gc don't escape its first argument
                 if not (idx == 0 and opnum in [rop.SETARRAYITEM_GC]):
@@ -71,18 +65,19 @@
         if box in self.new_boxes:
             self.new_boxes[box] = False
         if box in self.dependencies:
-            for dep in self.dependencies[box]:
+            deps = self.dependencies[box]
+            del self.dependencies[box]
+            for dep in deps:
                 self._escape(dep)
-            del self.dependencies[box]
 
     def clear_caches(self, opnum, descr, argboxes):
-        if opnum == rop.SETFIELD_GC:
-            return
-        if opnum == rop.SETARRAYITEM_GC:
-            return
-        if opnum == rop.SETFIELD_RAW:
-            return
-        if opnum == rop.SETARRAYITEM_RAW:
+        if (opnum == rop.SETFIELD_GC or
+            opnum == rop.SETARRAYITEM_GC or
+            opnum == rop.SETFIELD_RAW or
+            opnum == rop.SETARRAYITEM_RAW or
+            opnum == rop.SETINTERIORFIELD_GC or
+            opnum == rop.COPYSTRCONTENT or
+            opnum == rop.COPYUNICODECONTENT):
             return
         if rop._OVF_FIRST <= opnum <= rop._OVF_LAST:
             return
@@ -91,9 +86,9 @@
         if opnum == rop.CALL or opnum == rop.CALL_LOOPINVARIANT:
             effectinfo = descr.get_extra_info()
             ef = effectinfo.extraeffect
-            if ef == effectinfo.EF_LOOPINVARIANT or \
-               ef == effectinfo.EF_ELIDABLE_CANNOT_RAISE or \
-               ef == effectinfo.EF_ELIDABLE_CAN_RAISE:
+            if (ef == effectinfo.EF_LOOPINVARIANT or
+                ef == effectinfo.EF_ELIDABLE_CANNOT_RAISE or
+                ef == effectinfo.EF_ELIDABLE_CAN_RAISE):
                 return
             # A special case for ll_arraycopy, because it is so common, and its
             # effects are so well defined.
@@ -115,23 +110,18 @@
         self.heap_array_cache.clear()
 
     def is_class_known(self, box):
-        box = self.get_repr(box)
         return box in self.known_class_boxes
 
     def class_now_known(self, box):
-        box = self.get_repr(box)
         self.known_class_boxes[box] = None
 
     def is_nonstandard_virtualizable(self, box):
-        box = self.get_repr(box)
         return box in self.nonstandard_virtualizables
 
     def nonstandard_virtualizables_now_known(self, box):
-        box = self.get_repr(box)
         self.nonstandard_virtualizables[box] = None
 
     def is_unescaped(self, box):
-        box = self.get_repr(box)
         return self.new_boxes.get(box, False)
 
     def new(self, box):
@@ -142,7 +132,6 @@
         self.arraylen_now_known(box, lengthbox)
 
     def getfield(self, box, descr):
-        box = self.get_repr(box)
         d = self.heap_cache.get(descr, None)
         if d:
             tobox = d.get(box, None)
@@ -151,11 +140,9 @@
         return None
 
     def getfield_now_known(self, box, descr, fieldbox):
-        box = self.get_repr(box)
         self.heap_cache.setdefault(descr, {})[box] = fieldbox
 
     def setfield(self, box, descr, fieldbox):
-        box = self.get_repr(box)
         d = self.heap_cache.get(descr, None)
         new_d = self._do_write_with_aliasing(d, box, fieldbox)
         self.heap_cache[descr] = new_d
@@ -179,7 +166,6 @@
         return new_d
 
     def getarrayitem(self, box, descr, indexbox):
-        box = self.get_repr(box)
         if not isinstance(indexbox, ConstInt):
             return
         index = indexbox.getint()
@@ -190,7 +176,6 @@
                 return indexcache.get(box, None)
 
     def getarrayitem_now_known(self, box, descr, indexbox, valuebox):
-        box = self.get_repr(box)
         if not isinstance(indexbox, ConstInt):
             return
         index = indexbox.getint()
@@ -202,7 +187,6 @@
             cache[index] = {box: valuebox}
 
     def setarrayitem(self, box, descr, indexbox, valuebox):
-        box = self.get_repr(box)
         if not isinstance(indexbox, ConstInt):
             cache = self.heap_array_cache.get(descr, None)
             if cache is not None:
@@ -214,11 +198,9 @@
         cache[index] = self._do_write_with_aliasing(indexcache, box, valuebox)
 
     def arraylen(self, box):
-        box = self.get_repr(box)
         return self.length_cache.get(box, None)
 
     def arraylen_now_known(self, box, lengthbox):
-        box = self.get_repr(box)
         self.length_cache[box] = lengthbox
 
     def _replace_box(self, d, oldbox, newbox):
@@ -232,7 +214,6 @@
         return new_d
 
     def replace_box(self, oldbox, newbox):
-        oldbox = self.get_repr(oldbox)
         for descr, d in self.heap_cache.iteritems():
             self.heap_cache[descr] = self._replace_box(d, oldbox, newbox)
         for descr, d in self.heap_array_cache.iteritems():
diff --git a/pypy/jit/metainterp/history.py b/pypy/jit/metainterp/history.py
--- a/pypy/jit/metainterp/history.py
+++ b/pypy/jit/metainterp/history.py
@@ -9,12 +9,14 @@
 
 from pypy.jit.metainterp.resoperation import ResOperation, rop
 from pypy.jit.codewriter import heaptracker, longlong
+from pypy.rlib.objectmodel import compute_identity_hash
 
 # ____________________________________________________________
 
 INT   = 'i'
 REF   = 'r'
 FLOAT = 'f'
+STRUCT = 's'
 HOLE  = '_'
 VOID  = 'v'
 
@@ -104,7 +106,7 @@
     getref._annspecialcase_ = 'specialize:arg(1)'
 
     def _get_hash_(self):
-        raise NotImplementedError
+        return compute_identity_hash(self)
 
     def clonebox(self):
         raise NotImplementedError
@@ -133,6 +135,9 @@
     def _get_str(self):
         raise NotImplementedError
 
+    def same_box(self, other):
+        return self is other
+
 class AbstractDescr(AbstractValue):
     __slots__ = ()
 
@@ -168,6 +173,11 @@
         """
         raise NotImplementedError
 
+    def is_array_of_structs(self):
+        """ Implement for array descr
+        """
+        raise NotImplementedError
+
     def is_pointer_field(self):
         """ Implement for field descr
         """
@@ -241,32 +251,15 @@
     def constbox(self):
         return self
 
+    def same_box(self, other):
+        return self.same_constant(other)
+
     def same_constant(self, other):
         raise NotImplementedError
 
     def __repr__(self):
         return 'Const(%s)' % self._getrepr_()
 
-    def __eq__(self, other):
-        "NOT_RPYTHON"
-        # Remember that you should not compare Consts with '==' in RPython.
-        # Consts have no special __hash__, in order to force different Consts
-        # from being considered as different keys when stored in dicts
-        # (as they always are after translation).  Use a dict_equal_consts()
-        # to get the other behavior (i.e. using this __eq__).
-        if self.__class__ is not other.__class__:
-            return False
-        try:
-            return self.value == other.value
-        except TypeError:
-            if (isinstance(self.value, Symbolic) and
-                isinstance(other.value, Symbolic)):
-                return self.value is other.value
-            raise
-
-    def __ne__(self, other):
-        return not (self == other)
-
 
 class ConstInt(Const):
     type = INT
@@ -688,33 +681,6 @@
 
 # ____________________________________________________________
 
-def dict_equal_consts():
-    "NOT_RPYTHON"
-    # Returns a dict in which Consts that compare as equal
-    # are identified when used as keys.
-    return r_dict(dc_eq, dc_hash)
-
-def dc_eq(c1, c2):
-    return c1 == c2
-
-def dc_hash(c):
-    "NOT_RPYTHON"
-    # This is called during translation only.  Avoid using identityhash(),
-    # to avoid forcing a hash, at least on lltype objects.
-    if not isinstance(c, Const):
-        return hash(c)
-    if isinstance(c.value, Symbolic):
-        return id(c.value)
-    try:
-        if isinstance(c, ConstPtr):
-            p = lltype.normalizeptr(c.value)
-            if p is not None:
-                return hash(p._obj)
-            else:
-                return 0
-        return c._get_hash_()
-    except lltype.DelayedPointer:
-        return -2      # xxx risk of changing hash...
 
 def make_hashable_int(i):
     from pypy.rpython.lltypesystem.ll2ctypes import NotCtypesAllocatedStructure
@@ -772,6 +738,7 @@
     failed_states = None
     retraced_count = 0
     terminating = False # see TerminatingLoopToken in compile.py
+    invalidated = False
     outermost_jitdriver_sd = None
     # and more data specified by the backend when the loop is compiled
     number = -1
@@ -962,6 +929,9 @@
     def view(self, **kwds):
         pass
 
+    def clear(self):
+        pass
+
 class Stats(object):
     """For tests."""
 
@@ -974,6 +944,16 @@
         self.loops = []
         self.locations = []
         self.aborted_keys = []
+        self.invalidated_token_numbers = set()
+
+    def clear(self):
+        del self.loops[:]
+        del self.locations[:]
+        del self.aborted_keys[:]
+        self.invalidated_token_numbers.clear()
+        self.compiled_count = 0
+        self.enter_count = 0
+        self.aborted_count = 0
 
     def set_history(self, history):
         self.operations = history.operations
@@ -1052,7 +1032,12 @@
             if loop in loops:
                 loops.remove(loop)
             loops.append(loop)
-        display_loops(loops, errmsg, extraloops)
+        highlight_loops = dict.fromkeys(extraloops, 1)
+        for loop in loops:
+            if hasattr(loop, '_looptoken_number') and (
+                    loop._looptoken_number in self.invalidated_token_numbers):
+                highlight_loops.setdefault(loop, 2)
+        display_loops(loops, errmsg, highlight_loops)
 
 # ----------------------------------------------------------------
 
diff --git a/pypy/jit/metainterp/memmgr.py b/pypy/jit/metainterp/memmgr.py
--- a/pypy/jit/metainterp/memmgr.py
+++ b/pypy/jit/metainterp/memmgr.py
@@ -68,7 +68,8 @@
         debug_print("Loop tokens before:", oldtotal)
         max_generation = self.current_generation - (self.max_age-1)
         for looptoken in self.alive_loops.keys():
-            if 0 <= looptoken.generation < max_generation:
+            if (0 <= looptoken.generation < max_generation or
+                looptoken.invalidated):
                 del self.alive_loops[looptoken]
         newtotal = len(self.alive_loops)
         debug_print("Loop tokens freed: ", oldtotal - newtotal)
diff --git a/pypy/jit/metainterp/optimizeopt/intbounds.py b/pypy/jit/metainterp/optimizeopt/intbounds.py
--- a/pypy/jit/metainterp/optimizeopt/intbounds.py
+++ b/pypy/jit/metainterp/optimizeopt/intbounds.py
@@ -1,3 +1,4 @@
+import sys
 from pypy.jit.metainterp.optimizeopt.optimizer import Optimization, CONST_1, CONST_0, \
                                                   MODE_ARRAY, MODE_STR, MODE_UNICODE
 from pypy.jit.metainterp.history import ConstInt
@@ -5,6 +6,7 @@
     IntUpperBound)
 from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
 from pypy.jit.metainterp.resoperation import rop
+from pypy.rlib.rarithmetic import LONG_BIT
 
 
 class OptIntBounds(Optimization):
@@ -126,14 +128,29 @@
         r.intbound.intersect(v1.intbound.div_bound(v2.intbound))
 
     def optimize_INT_MOD(self, op):
+        v1 = self.getvalue(op.getarg(0))
+        v2 = self.getvalue(op.getarg(1))
+        known_nonneg = (v1.intbound.known_ge(IntBound(0, 0)) and 
+                        v2.intbound.known_ge(IntBound(0, 0)))
+        if known_nonneg and v2.is_constant():
+            val = v2.box.getint()
+            if (val & (val-1)) == 0:
+                # nonneg % power-of-two ==> nonneg & (power-of-two - 1)


More information about the pypy-commit mailing list