07:03 <njs> anyone have any guesses about why calling gc.collect() repeatedly could fail to collect an unreachable object, BUT ONLY if running with coverage enabled?

07:20 <njs> oh gods, is this another interaction between tracing and fast locals? I'm doing 'del e; print(locals())' and with coverage enabled, the print shows 'e' as still present

10:29 <cfbolz> njs: hey :-)

10:29 <cfbolz> in pypy or in cpython?

16:25 <mattip> we should really upgrade our coverage support since the first place many library developers meet PyPy is when running CI, and many of them turn on coverage by default

16:26 <mattip> so they get a bad impression about pypy from the very first CI run

16:26 <mattip> slow, buggy, ...

16:34 <mattip> cfbolz: there are a couple of at_exit tests failing in test_threading

16:34 <mattip> http://buildbot.pypy.org/summary/longrepr?testname=unmodified&builder=pypy-c-jit-linux-x86-64&build=8233&mod=lib-python.3.test.test_threading

16:35 <mattip> maybe connected to both njs's report and the compileall problem?

16:36 <mattip> on py3.9, which is probably not what njs is testing, but maybe the test fails similarly on older versions...

16:36 <cfbolz> mattip: right

16:36 <cfbolz> (I'll probably get to it on Monday only)

17:00 <mattip> no problem

17:15 <njs> cfbolz: on pypy, specifically the latest 3.8 release, but based on CI results I suspect the same thing happens on {3.7, 3.8} {release, nightly}

17:30 <cfbolz> njs: ok, link?

17:48 <Cheery> I'm implementing constraint handling rules into my makeshift ukanren implementation.

17:48 <Cheery> I don't know if I get it because it's difficult.

17:49 <Cheery> incredibly simple idea, but complicated to implement. There's operational semantics paper that seem like it helps though. :)

18:17 <mattip> there is also a test_subprocess test failing on py3.7 and up

18:17 <mattip> about signals

18:50 <njs> cfbolz: I don't have a minimized test case ATM, but this is the test: https://github.com/python-trio/trio/blob/5ea2e898ed3c78e4d4793eee1496575697b8d9b7/trio/tests/test_dtls.py#L699-L727

18:51 <njs> after the 'del e' on line 720, if you add a 'print(locals())' then 'e' is still present, iff running on pypy under coverage

18:52 <njs> I suspect it has to do somehow with the call to `await force_receive_loop_to_start()` capturing 'e' in a cell, even though that frame is dead and gone before the 'del e'

19:20 <mattip> we never implemented posix_spawn for py3.8

19:20 <mattip> https://docs.python.org/3/library/os.html#os.posix_spawn

19:34 <njs> minimized test case: https://foss.heptapod.net/pypy/pypy/-/issues/3656

19:46 <Cheery> all right.. I need to read the paper carefully so that I didn't miss anything, but otherwise I have my first constraint handling rule engine.

19:48 <Cheery> code or it didn't happen

19:49 <Cheery> https://gist.github.com/cheery/285e81d55ca4d9ec44d628978545a866

19:53 <Corbin> CHR! Good times.

19:56 <cfbolz> njs: thank you

20:21 <Cheery> ok. immediately few bugs and issues I didn't notice.

20:22 <Cheery> I'll add the rest of the rules tmr. and see how it works in practice.

20:22 <Cheery> Also going to finish reading the paper to see if I missed anything important.

20:22 <Cheery> I don't have propagation history for instance, but not sure what that is exactly.