Mailman 3 August 2017 - pypy-issue

Issue #2642: Deleting attributes in another module is significantly slower than CPython (pypy/pypy)
by afteryu 31 Aug '17

31 Aug '17

New issue 2642: Deleting attributes in another module is significantly slower than CPython https://bitbucket.org/pypy/pypy/issues/2642/deleting-attributes-in-another-… afteryu: I came across a performance regression on PyPy3 (v5.8.0-beta0) when I was doing some nasty namespace hacks. I tried to narrow the problem down, and found that PyPy3, when deleting attributes of another module, is significantly slower than CPython. Below is the test case I managed to come up with. In `test.py`: ```python import foo def benchmark(func): from time import perf_counter def inner(): start = perf_counter() for i in range(1000000): func() end = perf_counter() print(end - start) return inner @benchmark def add_del(): foo.bar = 'baz' del foo.bar if __name__ == '__main__': add_del() ``` Module `foo` is an empty file. Running `test.py` yields the following results: ``` $ pypy3 -V Python 3.5.3 (a37ecfe5f142bc971a86d17305cc5d1d70abec64, Jun 10 2017, 18:23:14) [PyPy 5.8.0-beta0 with GCC 4.2.1 Compatible Apple LLVM 8.1.0 (clang-802.0.42)] $ python3 -V Python 3.6.2 $ pypy3 test.py 1.1875620842911303 $ python3 test.py 0.3927199188619852 ``` Changing from `del foo.bar` to `delattr(foo, 'bar')` gives similar results. The problem disappears, however, if I were to delete an attribute of a class, or an item in a dictionary. Thanks a lot.

1 0

Issue #2641: XML library incorrectly parsing files on Windows (pypy/pypy)
by SimonVM 29 Aug '17

29 Aug '17

New issue 2641: XML library incorrectly parsing files on Windows https://bitbucket.org/pypy/pypy/issues/2641/xml-library-incorrectly-parsing… SimonVM: pypy incorrectly parses XML files using the default xml.etree.ElementTree library. A minimal working example is given below. XML file "test.xml": ``` #!xml <root></root> ``` Python file "test.py": ``` #!python import xml.etree.ElementTree as ET tree = ET.parse("test.xml") root = tree.getroot() print root.tag ``` Expected output of running `pypy test.py` is "root", actual output is "r". Other functions (such as findall) are also affected. This is an issue with latest version of PyPy, v5.8.0-win32 on 64-bit Windows 10. It runs correctly on Linux, as well as for Python v2.7.13.

1 0

Issue #2640: sandbox translation in ubnutu 16.04 LTS 64bit is failing (pypy/pypy)
by kumarsantosh3295 25 Aug '17

25 Aug '17

New issue 2640: sandbox translation in ubnutu 16.04 LTS 64bit is failing https://bitbucket.org/pypy/pypy/issues/2640/sandbox-translation-in-ubnutu-1… kumarsantosh3295: 1.First i have installed all the dependencies for pypy 2.Downloaded pypy from the latest pull from official pypy website 3.Then when i want to translate from pypy/pypy/goal directory to generate "pypy-c-sandbox" executable file,this is end with this following error [translation:info] 2.7.12 (default, Nov 19 2016, 06:48:10) [GCC 5.4.0 20160609] [platform:msg] Set platform with 'host' cc=None, using cc='gcc', version='Unknown' [translation:info] Translating target as defined by targetpypystandalone [translation:info] Error: File "/home/santo/pypy/rpython/translator/goal/translate.py", line 284, in main default_goal='compile') File "/home/santo/pypy/rpython/translator/driver.py", line 566, in from_targetspec spec = target(driver, args) File "targetpypystandalone.py", line 313, in target assert 0, ("--sandbox is not tested nor maintained. If you " [translation:ERROR] AssertionError: --sandbox is not tested nor maintained. If you really want to try it anyway, remove this line in pypy/goal/targetpypystandalone.py. [translation] start debugger... > /home/santo/pypy/pypy/goal/targetpypystandalone.py(313)target() -> assert 0, ("--sandbox is not tested nor maintained. If you " (Pdb+)

1 0

Issue #2639: PyPy3.5 should allow assigning to module object __class__ (pypy/pypy)
by Nathaniel Smith 23 Aug '17

23 Aug '17

New issue 2639: PyPy3.5 should allow assigning to module object __class__ https://bitbucket.org/pypy/pypy/issues/2639/pypy35-should-allow-assigning-t… Nathaniel Smith: I was trying to use [`metamodule`](https://github.com/njsmith/metamodule) on pypy3 5.8.0, but it doesn't work: unlike CPython, you can't assign to module object's `__class__` attribute. This is a narrow exception to the normal "`__class__` assignment is only for heap types" rule that was added in CPython 3.5 – now the rule is "only heap types or modules". [CPython source.](https://github.com/python/cpython/blob/772d809a63f40fd35679da3fb11… PyPy should implement this too so I can deprecate module attributes :-)

1 0

Issue #2638: pypy3 UnicodeDecodeError on start interactive shell (pypy/pypy)
by chrys 21 Aug '17

21 Aug '17

New issue 2638: pypy3 UnicodeDecodeError on start interactive shell https://bitbucket.org/pypy/pypy/issues/2638/pypy3-unicodedecodeerror-on-sta… chrys: Howdy, I pypy3 on my Archbox to use it. Its amazing! But for me it creates an trace at running pypy3 from bash. (but it still opens, and runs, so no crash at all). I mark is as minor since the application is still runnable. here is the trace trance (first line is my bashprompt mabye it has problems with the "✓" somehow?): 16:38 [user@host~] ✓$ pypy3 Python 3.5.3 (a37ecfe5f142bc971a86d17305cc5d1d70abec64, Jun 12 2017, 16:41:39) [PyPy 5.8.0-beta0 with GCC 7.1.1 20170528] on linux Type "help", "copyright", "credits" or "license" for more information. Traceback (most recent call last): File "/opt/pypy3/lib-python/3/site.py", line 430, in register_readline readline.read_history_file(history) File "/opt/pypy3/lib_pypy/pyrepl/readline.py", line 322, in read_history_file for line in f: File "/opt/pypy3/lib-python/3/codecs.py", line 321, in decode (result, consumed) = self._buffer_decode(data, self.errors, final) UnicodeDecodeError: 'utf8' codec can't decode byte 0xe2 in position 80: invalid continuation byte And now for something completely different: ``happy PyPy's year 2010!'' >>>>

1 0

Issue #2637: Cannot create a gRPC server with PyPy (pypy/pypy)
by Omer Katz 20 Aug '17

20 Aug '17

New issue 2637: Cannot create a gRPC server with PyPy https://bitbucket.org/pypy/pypy/issues/2637/cannot-create-a-grpc-server-wit… Omer Katz: This issue is present in PyPy2/3 5.8.0 and possibly previous versions which I haven't checked. When creating a gRPC server we get an RPython fatal error because CPyListStrategy_getstorage_copy is not implemented. You can reproduce this problem by installing grpc: ``` #!bash pip install grpcio ``` and running the following script: ``` #!python import grpc from concurrent import futures server = grpc.server(futures.ThreadPoolExecutor(max_workers=1)) server.start() ``` The code snippet above returns the following error: ``` #!text RPython traceback: File "pypy_interpreter.c", line 38711, in BuiltinCode4_fastcall_4 File "implement_2.c", line 12927, in fastfunc_descr_setslice_4 File "pypy_objspace_std_1.c", line 22711, in EmptyListStrategy_setslice File "pypy_module_cpyext_2.c", line 16832, in CPyListStrategy_getstorage_copy Fatal RPython error: NotImplementedError Aborted (core dumped) ```

1 0

Issue #2636: sys.stderr cannot write big string on Windows (pypy/pypy)
by Armin Rigo 18 Aug '17

18 Aug '17

New issue 2636: sys.stderr cannot write big string on Windows https://bitbucket.org/pypy/pypy/issues/2636/sysstderr-cannot-write-big-stri… Armin Rigo: On Windows, the following script fails with ``IOError: Not enough space: '<stderr>'``. The same script works on CPython 2.7.10. I suspect it's Windows that gives nonsense but CPython works around it. ``` #!python import sys sys.stderr.write('x' * 0xffffff) ```

1 0

Issue #2635: datetime.replace always returns datetime.datetime in pypy3 (pypy/pypy)
by Paul G 16 Aug '17

16 Aug '17

New issue 2635: datetime.replace always returns datetime.datetime in pypy3 https://bitbucket.org/pypy/pypy/issues/2635/datetimereplace-always-returns Paul G: MWE: `demo_datetime_replace.py`: ``` #!python from datetime import datetime class SuperDatetime(datetime): pass if __name__ == "__main__": sdt0 = SuperDatetime(2017, 8, 1) sdt1 = sdt0.replace(2016, 8, 1) print(repr(sdt0)) print(repr(sdt1)) ``` ``` #! bash $ python2 demo_datetime_replace.py SuperDatetime(2017, 8, 1, 0, 0) SuperDatetime(2016, 8, 1, 0, 0) $ python3 demo_datetime_replace.py SuperDatetime(2017, 8, 1, 0, 0) SuperDatetime(2016, 8, 1, 0, 0) $ pypy demo_datetime_replace.py SuperDatetime(2017, 8, 1, 0, 0) SuperDatetime(2016, 8, 1, 0, 0) $ pypy3 demo_datetime_replace.py __main__.SuperDatetime(2017, 8, 1, 0, 0) datetime.datetime(2016, 8, 1, 0, 0) ``` It seems that `pypy3` is *also* the odd man out with respect to how `repr` is treated, but that's a separate and less important issue. It's not clear to me how much "`datetime.replace` returns the same class of object" is part of the Python spec, but it seems reasonable and it's definitely a shift in behavior. There's [something about subclasses of builtin types here](http://doc.pypy.org/en/latest/cpython_differences.html#subclasses-of-…, but it doesn't seem relevant to me, if I'm understanding it right.

1 0

Issue #2634: datetime example where pypy3 is 30x slower (and pypy2 is 3x slower) (pypy/pypy)
by Nathaniel Smith 16 Aug '17

16 Aug '17

New issue 2634: datetime example where pypy3 is 30x slower (and pypy2 is 3x slower) https://bitbucket.org/pypy/pypy/issues/2634/datetime-example-where-pypy3-is… Nathaniel Smith: Test script adapted from gh16ito on IRC: ```python # -*- encoding: utf-8 -*- import time from datetime import datetime, timedelta def time_batch(batch_size, name, fn): start = time.time() for _ in range(batch_size): fn() end = time.time() print("{:.2f} µs/{}".format((end - start) / batch_size * 1e6, name)) def test_func(): delta = timedelta(days=365+31+1, seconds=1+60+3600) dt = datetime(1900, 1, 1) for i in range(200): dt += delta def test_func_ctime(): delta = timedelta(days=365+31+1, seconds=1+60+3600) dt = datetime(1900, 1, 1) for i in range(200): dt += delta dt.ctime() if __name__ == "__main__": while True: time_batch(10000, "test_func", test_func) time_batch(10000, "test_func_ctime", test_func) ``` CPython 2.7: ~22 µs/call CPython 3.6: ~16 µs/call PyPy2 5.8: ~60 µs/call PyPy3 5.8: ~480 µs/call (after a long warm up period)

1 0

Issue #2633: Cannot import BeautifulSoup from bs4 on pypy3 5.8.0 (pypy/pypy)
by Nick Malaguti 15 Aug '17

15 Aug '17

New issue 2633: Cannot import BeautifulSoup from bs4 on pypy3 5.8.0 https://bitbucket.org/pypy/pypy/issues/2633/cannot-import-beautifulsoup-fro… Nick Malaguti: ``` $ docker run -it pypy:3-5.8.0 bash root@11a1fb75b382:/# pip install beautifulsoup4 Collecting beautifulsoup4 Downloading beautifulsoup4-4.6.0-py3-none-any.whl (86kB) 100% |████████████████████████████████| 92kB 2.2MB/s Installing collected packages: beautifulsoup4 Successfully installed beautifulsoup4-4.6.0 root@11a1fb75b382:/# pip install lxml Collecting lxml Downloading lxml-3.8.0.tar.gz (3.8MB) 100% |████████████████████████████████| 3.8MB 254kB/s Building wheels for collected packages: lxml Running setup.py bdist_wheel for lxml ... done Stored in directory: /root/.cache/pip/wheels/e2/52/c3/a7b7aa7dc1294d715fa6eb49d5c63e0d9713652fcc6a609857 Successfully built lxml Installing collected packages: lxml Successfully installed lxml-3.8.0 root@11a1fb75b382:/# pypy3 Python 3.5.3 (a39af0be3a22, Jun 05 2017, 20:18:00) [PyPy 5.8.0-beta0 with GCC 6.2.0 20160901] on linux Type "help", "copyright", "credits" or "license" for more information. And now for something completely different: ``"ah, just in time documentation" (__ap__)'' >>>> from bs4 import BeautifulSoup Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/usr/local/site-packages/bs4/__init__.py", line 35, in <module> from .builder import builder_registry, ParserRejectedMarkup File "/usr/local/site-packages/bs4/builder/__init__.py", line 329, in <module> from . import _lxml File "/usr/local/site-packages/bs4/builder/_lxml.py", line 11, in <module> from lxml import etree File "src/lxml/xmlschema.pxi", line 22, in init lxml.etree (src/lxml/lxml.etree.c:231895) File "src/lxml/xpath.pxi", line 420, in lxml.etree.XPath.__init__ (src/lxml/lxml.etree.c:173999) File "src/lxml/xpath.pxi", line 150, in lxml.etree._XPathEvaluatorBase.set_context (src/lxml/lxml.etree.c:170398) File "src/lxml/xpath.pxi", line 66, in lxml.etree._XPathContext.set_context (src/lxml/lxml.etree.c:169176) File "src/lxml/extensions.pxi", line 255, in lxml.etree._BaseContext.registerLocalFunctions (src/lxml/lxml.etree.c:160400) AttributeError: 'dict' object has no attribute 'iteritems' >>>> ``` There is a similar issue #2498 that involved Cython.

1 0