[Python-checkins] CVS: python/dist/src/Lib Cookie.py,1.8,1.9
Tim Peters
tim_one@users.sourceforge.net
Sat, 12 May 2001 17:19:33 -0700
Update of /cvsroot/python/python/dist/src/Lib
In directory usw-pr-cvs1:/tmp/cvs-serv1323/python/dist/src/Lib
Modified Files:
Cookie.py
Log Message:
Get rid of the superstitious "~" in dict hashing's "i = (~hash) & mask".
The comment following used to say:
/* We use ~hash instead of hash, as degenerate hash functions, such
as for ints <sigh>, can have lots of leading zeros. It's not
really a performance risk, but better safe than sorry.
12-Dec-00 tim: so ~hash produces lots of leading ones instead --
what's the gain? */
That is, there was never a good reason for doing it. And to the contrary,
as explained on Python-Dev last December, it tended to make the *sum*
(i + incr) & mask (which is the first table index examined in case of
collison) the same "too often" across distinct hashes.
Changing to the simpler "i = hash & mask" reduced the number of string-dict
collisions (== # number of times we go around the lookup for-loop) from about
6 million to 5 million during a full run of the test suite (these are
approximate because the test suite does some random stuff from run to run).
The number of collisions in non-string dicts also decreased, but not as
dramatically.
Note that this may, for a given dict, change the order (wrt previous
releases) of entries exposed by .keys(), .values() and .items(). A number
of std tests suffered bogus failures as a result. For dicts keyed by
small ints, or (less so) by characters, the order is much more likely to be
in increasing order of key now; e.g.,
>>> d = {}
>>> for i in range(10):
... d[i] = i
...
>>> d
{0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9}
>>>
Unfortunately. people may latch on to that in small examples and draw a
bogus conclusion.
test_support.py
Moved test_extcall's sortdict() into test_support, made it stronger,
and imported sortdict into other std tests that needed it.
test_unicode.py
Excluced cp875 from the "roundtrip over range(128)" test, because
cp875 doesn't have a well-defined inverse for unicode("?", "cp875").
See Python-Dev for excruciating details.
Cookie.py
Chaged various output functions to sort dicts before building
strings from them.
test_extcall
Fiddled the expected-result file. This remains sensitive to native
dict ordering, because, e.g., if there are multiple errors in a
keyword-arg dict (and test_extcall sets up many cases like that), the
specific error Python complains about first depends on native dict
ordering.
Index: Cookie.py
===================================================================
RCS file: /cvsroot/python/python/dist/src/Lib/Cookie.py,v
retrieving revision 1.8
retrieving revision 1.9
diff -C2 -r1.8 -r1.9
*** Cookie.py 2001/04/06 19:39:11 1.8
--- Cookie.py 2001/05/13 00:19:31 1.9
***************
*** 71,76 ****
>>> C["sugar"] = "wafer"
>>> print C
- Set-Cookie: sugar=wafer;
Set-Cookie: fig=newton;
Notice that the printable representation of a Cookie is the
--- 71,76 ----
>>> C["sugar"] = "wafer"
>>> print C
Set-Cookie: fig=newton;
+ Set-Cookie: sugar=wafer;
Notice that the printable representation of a Cookie is the
***************
*** 94,99 ****
>>> C.load("chips=ahoy; vienna=finger")
>>> print C
- Set-Cookie: vienna=finger;
Set-Cookie: chips=ahoy;
The load() method is darn-tootin smart about identifying cookies
--- 94,99 ----
>>> C.load("chips=ahoy; vienna=finger")
>>> print C
Set-Cookie: chips=ahoy;
+ Set-Cookie: vienna=finger;
The load() method is darn-tootin smart about identifying cookies
***************
*** 494,498 ****
if attrs is None:
attrs = self._reserved_keys
! for K,V in self.items():
if V == "": continue
if K not in attrs: continue
--- 494,500 ----
if attrs is None:
attrs = self._reserved_keys
! items = self.items()
! items.sort()
! for K,V in items:
if V == "": continue
if K not in attrs: continue
***************
*** 587,591 ****
"""Return a string suitable for HTTP."""
result = []
! for K,V in self.items():
result.append( V.output(attrs, header) )
return string.join(result, sep)
--- 589,595 ----
"""Return a string suitable for HTTP."""
result = []
! items = self.items()
! items.sort()
! for K,V in items:
result.append( V.output(attrs, header) )
return string.join(result, sep)
***************
*** 596,600 ****
def __repr__(self):
L = []
! for K,V in self.items():
L.append( '%s=%s' % (K,repr(V.value) ) )
return '<%s: %s>' % (self.__class__.__name__, string.join(L))
--- 600,606 ----
def __repr__(self):
L = []
! items = self.items()
! items.sort()
! for K,V in items:
L.append( '%s=%s' % (K,repr(V.value) ) )
return '<%s: %s>' % (self.__class__.__name__, string.join(L))
***************
*** 603,607 ****
"""Return a string suitable for JavaScript."""
result = []
! for K,V in self.items():
result.append( V.js_output(attrs) )
return string.join(result, "")
--- 609,615 ----
"""Return a string suitable for JavaScript."""
result = []
! items = self.items()
! items.sort()
! for K,V in items:
result.append( V.js_output(attrs) )
return string.join(result, "")