Mailman 3 February 2015 - pypy-issue

Issue #1983: JIT crash on PyPy bridge. (pypy/pypy)
by Edd Barrett Feb. 13, 2015

Feb. 13, 2015

New issue 1983: JIT crash on PyPy bridge. https://bitbucket.org/pypy/pypy/issue/1983/jit-crash-on-pypy-bridge Edd Barrett: Discussed with Armin on IRC, this looks like an RPython bug. We are using the 'hippy_bridge' branch of our PyPy fork (in sync with changes with PyPy-2.5.0) and the HippyVM 'pypy_bridge' branch as of today. Our PyPy fork is here: https://bitbucket.org/softdevteam/pypy-hippy-bridge Sorry the test case is not very good. ``` $ /home/vext01/research/pyhyp_experiments/work/pyhyp/hippyvm/pyhyp /tmp/a.php RPython traceback: File "rpython_jit_metainterp_pyjitpl_1.c", line 64176, in MIFrame_run_one_step File "rpython_jit_metainterp_pyjitpl_2.c", line 64796, in handler_getarrayitem_vable_r_1 File "rpython_jit_metainterp_pyjitpl_3.c", line 55472, in MIFrame__opimpl_getarrayitem_vable File "rpython_jit_metainterp_pyjitpl_4.c", line 32219, in MIFrame__nonstandard_virtualizable File "rpython_jit_metainterp_pyjitpl_4.c", line 27757, in MetaInterp_replace_box Fatal RPython error: AssertionError Aborted /tmp/a.php: <?php{ function f_outer($n) { $inc = 2; $inner = embed_py_func("def inner(x): return x + inc");; $tot = 0; for ($i = 0; $i < $n; $i++) { $tot = $inner($tot); } $expect = 2 * $n; assert($tot == $expect); } function run_iter($n) { f_outer($n); } run_iter(1200); }?> $ ~/pyve/testing/bin/python hippy/main.py /tmp/a.php [platform:execute] gcc -c -O3 -pthread -fomit-frame-pointer -Wall -Wno-unused /tmp/usession-hippy_bridge-202/platcheck_0.c - o /tmp/usession-hippy_bridge-202/platcheck_0.o [platform:execute] gcc /tmp/usession-hippy_bridge-202/platcheck_0.o -pthread -lintl -lrt -o /tmp/usession-hippy_bridge-202/p latcheck_0 /home/vext01/research/pyhyp_experiments/work/pyhyp/hippyvm/hippy/sourceparser.py:2121: ParserGeneratorWarning: Token 'H_NEW_ LINE' is unused parser = pg.build() /home/vext01/research/pyhyp_experiments/work/pyhyp/hippyvm/hippy/sourceparser.py:2121: ParserGeneratorWarning: Token 'H_TABU LATURE' is unused ... [platform:execute] gcc -shared /tmp/usession-hippy_bridge-202/rpython/translator/c/src/signals.o -pthread -lrt -o /tmp/usession-hippy_bridge-202/shared_cache/externmod.so (process finishes ok) rpython/jit/metainterp/pyjitpl.py: def replace_box(self, oldbox, newbox): assert isinstance(oldbox, Box) <---? for frame in self.framestack: frame.replace_active_box_in_frame(oldbox, newbox) boxes = self.virtualref_boxes for i in range(len(boxes)): if boxes[i] is oldbox: boxes[i] = newbox if (self.jitdriver_sd.virtualizable_info is not None or self.jitdriver_sd.greenfield_info is not None): boxes = self.virtualizable_boxes for i in range(len(boxes)): if boxes[i] is oldbox: boxes[i] = newbox self.heapcache.replace_box(oldbox, newbox) ```

1 0

Issue #1982: ctypes.Structure subclasses munge `_names` attribute (pypy/pypy)
by Ben Longbons Feb. 13, 2015

Feb. 13, 2015

New issue 1982: ctypes.Structure subclasses munge `_names` attribute https://bitbucket.org/pypy/pypy/issue/1982/ctypesstructure-subclasses-munge… Ben Longbons: The below code will print `{}` in python 2.7, 3.2, and 3.4, but `['foo', 'bar']` in pypy 2.4.0 (in python2 or python3 mode). If there is a reason to introduce new names, they should follow the ctypes `_sunder_` method. (yeah, yeah, I should use `cffi`) ``` #!python import ctypes class Stuff(ctypes.Structure): __slots__ = () _fields_ = [('foo', ctypes.c_int), ('bar', ctypes.c_int)] _names = {} print(Stuff._names) ```

1 0

Issue #1981: Socket timeouts with the bitcoin-python library (pypy/pypy)
by Ivan Voras Feb. 11, 2015

Feb. 11, 2015

New issue 1981: Socket timeouts with the bitcoin-python library https://bitbucket.org/pypy/pypy/issue/1981/socket-timeouts-with-the-bitcoin… Ivan Voras: The bitcoin-python library is a relatively thin wrapper over the bitcoind JSON RPC API. However, after performing some operations, it simply hangs / timeouts on random operations when used with pypy (of course, it works on CPython). The library is pure Python and available here: http://laanwj.github.io/bitcoin-python/ (it's normally installable via pip, etc.). The following simple script demonstrates the problem: ``` from time import time import bitcoinrpc conn = bitcoinrpc.connect_to_local() t1 = time() for x in range(1000): # print x tx = conn.getrawtransaction('a79ffa13308ffc5ff5d157f2f86187d48bdff44e09071b808e5ce6eda7468571', 1) t2 = time() print tx print t2 - t1 ``` When run on CPython, it will complete without problems. When run under PyPy, it will hang after about 30 iterations. Unfortunately, running the above code will require that a bitcoin client (qt or cli) is running on the local machine and has RPC enabled, as well as a full database index...

1 0

Issue #1980: None in a rffi.llexternal call become /* nothing */ after translation (pypy/pypy)
by mattip Feb. 11, 2015

Feb. 11, 2015

New issue 1980: None in a rffi.llexternal call become /* nothing */ after translation https://bitbucket.org/pypy/pypy/issue/1980/none-in-a-rffillexternal-call-be… mattip: Using a None in a RPython rffi.llexternal function call results in ```/*nothing*/``` when translated. Can we get either an error earlier in the translation process or a NULL emitted in code, rather than a crash of c-compilation? ```ztranslation``` passed in this case [Here](https://gist.github.com/mattip/93f22744ad9384f48596) is an example of a complete translated function, from stdlib-2.7.9 pypy/module/_ssl/interp_win32.py

1 0

Issue #1979: PyPy --no-shared cannot load any C extension module (pypy/pypy)
by Armin Rigo Feb. 11, 2015

Feb. 11, 2015

New issue 1979: PyPy --no-shared cannot load any C extension module https://bitbucket.org/pypy/pypy/issue/1979/pypy-no-shared-cannot-load-any-c… Armin Rigo: In a modern PyPy translated with --no-shared, C extension modules cannot be loaded, because the "pypy" executable doesn't export the PyPyXxx symbols.

1 0

Issue #1978: readline: tab should insert 4 spaces when at start of line (pypy/pypy)
by Armin Rigo Feb. 11, 2015

Feb. 11, 2015

New issue 1978: readline: tab should insert 4 spaces when at start of line https://bitbucket.org/pypy/pypy/issue/1978/readline-tab-should-insert-4-spa… Armin Rigo: In the interactive prompt: ``` #!python >>>> if 1: .... <press tab here> ``` Pressing tab here should just insert 4 spaces. It makes no sense to display all 500 possible "completions" of the empty word, and using tab to indent is a very common feature of any editor with a Python mode.

1 0

Issue #1977: lib_pypy/gdbm.py: support FreeBSD (pypy/pypy)
by David Naylor Feb. 10, 2015

Feb. 10, 2015

New issue 1977: lib_pypy/gdbm.py: support FreeBSD https://bitbucket.org/pypy/pypy/issue/1977/lib_pypy-gdbmpy-support-freebsd David Naylor: On FreeBSD 3rd party libraries are installed under /usr/local. This patch teaches gdbm.py where to find gdbm.h and libgdbm.so on FreeBSD

1 0

Issue #1976: lib_pypy/_sqlite3.py: better support for FreeBSD (pypy/pypy)
by David Naylor Feb. 10, 2015

Feb. 10, 2015

New issue 1976: lib_pypy/_sqlite3.py: better support for FreeBSD https://bitbucket.org/pypy/pypy/issue/1976/lib_pypy-_sqlite3py-better-suppo… David Naylor: On FreeBSD the user may redefine the location of /usr/local (aka $LOCALBASE). This patch adds support for this.

1 0

Issue #1975: Pydub module very slow compared to CPython implementation (pypy/pypy)
by cryzed Feb. 10, 2015

Feb. 10, 2015

New issue 1975: Pydub module very slow compared to CPython implementation https://bitbucket.org/pypy/pypy/issue/1975/pydub-module-very-slow-compared-… cryzed: The Pydub module is 5~ times (PyPy 2) and up to 17~ times (PyPy 3) slower compared to the CPython 3 implementation when using the Pydub module. Absolute values were 3.5~ seconds for CPython, 18~ seconds for PyPy 2 and 58~ seconds for PyPy 3. Here is a sample test case: ``` #!python import sys import time import pydub PATHS = [ 'path_to_audio1.wav', 'path_to_audio2.wav', 'path_to_audio3.wav' ] def main(duration): segments = tuple(pydub.AudioSegment.from_file(path) for path in PATHS) # Create a silent audio segment to easily overlay the other segments composition = pydub.AudioSegment.silent(duration) for segment in segments: position = 0 length = len(segment) while position < duration: composition = composition.overlay(segment, position) position += length composition.export('output.wav', format='wav') if __name__ == '__main__': start = time.time() main(int(sys.argv[1])) print(time.time() - start) ``` I use 20 short (~20 seconds) *.ogg audio files for the PATHS and pass in a duration of 60000 milliseconds. Using this script and CPython 3 and PyPy 3 the values were 6~ seconds for CPython 3 and 166 seconds for PyPy 3.

1 0

Issue #1974: Different behaviour for collections of nans from CPython (pypy/pypy)
by David MacIver Feb. 9, 2015

Feb. 9, 2015

New issue 1974: Different behaviour for collections of nans from CPython https://bitbucket.org/pypy/pypy/issue/1974/different-behaviour-for-collecti… David MacIver: All of the following asserts pass in cpython (I've tried on 2.7, 3.3 and 3.4) but some of them fail on pypy (I've tried on pypy-2.5.0 and pypy3-2.4.0). ``` #!python n1 = float('nan') n2 = float('nan') x = [n1] assert n1 in x assert n2 not in x assert x.index(n1) == 0 try: x.index(n2) assert False except ValueError: pass ``` Basically, it seems like in cpython collection methods have a shortcut that treats x and y as equal if x is y even if not x == y. The behaviour for sets is also different. For set like collections CPython treats reference equal nans as equal but other nans as distinct. ``` #!python n1 = float('nan') n2 = float('nan') n3 = float('nan') x = {n1, n2} assert len(x) == 2 assert n1 in x assert n2 in x assert n3 not in x t = {n1} assert t == t assert t == {n1} assert t != {n2} ```

1 0