13:45 <Dejan> Any plays for 3.7 ? :)

13:45 <Dejan> plans*

13:58 <tos9> Dejan: There's a py3.7 branch.

13:59 <Dejan> Yea, but for some reason it fails to build for last few weeks. cfbolz explained to me once what is going on but I forgot...

13:59 <Dejan> :)

15:52 <raphaelm> Hello! I'm interested in the semantics of Python in general. As such I was wondering: how do you test that Pypy matches CPython's behavior? Which tests do you use? (I opened an issue on gitlab but mattip recommended discussing this on irc)

15:58 <cfbolz> raphaelm: a many pronged strategy ;-)

15:59 <cfbolz> we run the CPython test suite, we write our own tests and run those on CPython

16:03 <mattip> raphaelm: maybe describe a bit more the motivation of your question:

16:03 <mattip> are you considering creating your own interpreter?

16:03 <mattip> did you encounter a mismatch that seems concerning?

16:04 <mattip> would you like to get involved and are wondering how we know what to work on?

16:09 <raphaelm> Thank you! I'm writing a static analyzer (similar to Google's pytype), which can be seen as an interpreter (although it tries to explore all reachable program states, in some sense). As such I have to figure out Python's semantics. My current approach has been to read the doc and often read CPython implementation. I'd now like to test that my tool

16:09 <raphaelm> is correct, ie that it respects Python semantics. Currently, I'm especially interested in tests about the "core" semantics of Python (such as for loops using the iter and next methods ; attribute accesses ; variable scope). I guessed you guys probably have some more tests than the one written in cpython.

16:10 <cfbolz> raphaelm: yes, we definitely have some more tests

16:11 <cfbolz> the easy to understand ones are in file names apptest_*.py, throughout the repository

16:12 <raphaelm> So, no mismatch found between Pypy and CPython. Sadly, I don't have enough time to contribute to pypy's core for now, but I'm trying to clear up

16:12 <cfbolz> raphaelm: no, there are definitely some mismatches

16:12 <raphaelm> (sorry), but I'd be glad to clarify Python semantics one way or another, as well as maybe extend the pyperformance benchmarks

16:13 <raphaelm> cfbolz: thank you! I'll take a look at these tests. I'm aware there are some mismatches, I just wanted to clarify I haven't found new cases

16:13 <cfbolz> raphaelm: right

16:14 <cfbolz> raphaelm: we also have this document: https://doc.pypy.org/en/latest/cpython_differences.html

16:18 <raphaelm> those apptest_* are really nice, thank you!

16:20 <raphaelm> yes, I learned quite a few things from this page, thanks

16:29 <njs> raphaelm: you might find it interesting to get in touch with Mark Shannon: https://pyfound.blogspot.com/2020/04/a-formal-specification-for-cpython.html

16:37 <raphaelm> njs: oh, my next question was about standardizing Python language. That's very interesting, thanks a lot!

16:58 <raphaelm> I'm also wondering: have you ever discussed about a standardization of Python with CPython developers? I think that most languages are defined through a standard rather than an interpreter

17:09 <ronan> marmoute: I can't push due to a server error. Help?

17:09 <ronan> (I'm getting a long remote traceback ending with remote: LookupError: Can't find public descendants of '53b86a37acef955d060e59535d8bde676933c1d1' in topic 'BO-fix-doc-source-links')

17:11 <mattip> ronan: maybe try https://mattermost.heptapod.net/heptapod/channels

17:12 <marmoute> ronan: mattermost is the right place to report this. Share the full traceback there

17:13 <mattip> marmoute: which channel?

17:24 <cfbolz> raphaelm: i don't think there were serious discussions about standardization ever

17:33 <raphaelm> thank you!

17:46 <cfbolz> raphaelm: there are some formalization attempts though, in addition to what nathaniel linked there's also the work by Joe Gibbs Politz etal: http://cs.brown.edu/research/plt/dl/lambda-py/

18:13 <raphaelm> I'm a bit more up-to-date with the academic works (I'm a PhD student working on static analyses of python), but I don't think these attempts provide an understandable semantics for most of the community

18:14 <cfbolz> raphaelm: ah, nice, which uni/group?

18:17 <raphaelm> It's called Sorbonne Université in Paris. We're mainly doing static analysis by abstract interpretation

18:17 <cfbolz> raphaelm: cool (our type inference of rpython is abstract-interpretation-based)

18:23 <raphaelm> Yes, one reviewer pointed us to RPython recently ^^ We're trying to push the inference to take into account object mutation through some heap abstractions, and to value inference too

18:51 <bbot2> Started: http://buildbot.pypy.org/builders/own-linux-x86-64/builds/8148 [ronan: force build, py3.7]

18:51 <bbot2> Started: http://buildbot.pypy.org/builders/pypy-c-jit-linux-x86-64/builds/6963 [ronan: force build, py3.7]

19:09 <cfbolz> raphaelm: RPython has a much simpler job, of course. We simply don't support the hard things

19:49 <bbot2> Failure: http://buildbot.pypy.org/builders/own-linux-x86-64/builds/8148 [ronan: force build, py3.7]

19:52 <bbot2> Failure: http://buildbot.pypy.org/builders/pypy-c-jit-linux-x86-64/builds/6963 [ronan: force build, py3.7]

20:23 <raphaelm> cfbolz: you still support most of Python's semantics ;)

20:24 <cfbolz> Only the easy parts