06:02 <mgorny> mattip: yw, glad to help

06:03 <mgorny> i really do need pypy3.8 ASAP to left py3.7 cruft from Gentoo ;-)

06:03 <mgorny> clean leftover*

06:23 <mattip> mgorny: I will be making a rc2 soon,but we really need the virtualenv fix to go out before the final release

06:25 <mgorny> i agree

09:40 <ctismer> mattip: you said: `that would be os.path.join(sysconfig.get_config_var('LIBDIR'), sysconfig.get_config_var('LDLIBRARY'))`

09:41 <mattip> yes, didn't it work?

09:41 <ctismer> mattip: On macOS, this is pretty wrong. The correct setting is:

09:41 <ctismer> mattip: /Users/tismer/src/QtC/pypy3.7-v7.3.5-osx64/bin/libpypy3-c.dylib

09:42 <ctismer> mattip: but LDLIBRARY uses ".so"

09:44 <ctismer> mattip: Also, on macOS, the python3 equivalent would not concatenate correctly.

09:44 <ctismer> >>> os.path.join(sysconfig.get_config_var('LIBDIR'), sysconfig.get_config_var('LDLIBRARY'))

09:44 <ctismer> '/usr/local/opt/python@3.9/Frameworks/Python.framework/Versions/3.9/lib/Python.framework/Versions/3.9/Python'

09:52 <ctismer> mattip: here is Python itself wrong. The `.dylib` suffix appears in none of the sysconfig variables.

10:00 <mattip> oy. Does it work correctly on CPython?

10:02 <mattip> ahh, nvrmind, I see that _is_ CPython

10:03 <mattip> sysconfig is such a mess. Someone should write a PEP about minimum values that must exist there

10:08 <ctismer> mattip: Yes, and tests should be added to ensure that these minimum things are actually right. As always, that does not help right now, and then I'm going to forget it 😉

11:16 <mattip> it turns out windows does not have those values at all. I don't know how this is supposed to work

11:21 <ctismer> mattip: On windows, our setup worked out-of-the-box, funnily.

13:46 <mattip> cfbolz: now the traceback is showing "Did you mean hash" in the lib-python/3/test_code_module test for

13:46 <mattip> "try: ham\nexcept: eggs\n"

13:47 <mattip> where on CPython it does not have any suggestion

13:48 <mattip> is that because we always import "hash" somehow where CPython does not, or is it an edge case in the distance measure?

20:24 <mattip> arigo_: in CPython3.8, c-extension modules on windows are loaded with LoadLibraryEx() with flags to ignore PATH and use AddDllDirectory instead

20:24 <mattip> this is the PR https://github.com/python/cpython/pull/12302

20:24 <mattip> and this is the issue https://bugs.python.org/issue36085

20:25 <mattip> so I need to modify the cpyext/api.py call to follow suit

20:25 <mattip> does this affect CFFI?

21:12 <mattip> it affect ctypes, which uses rdynload.dlopen() and now must use rdynload.dlopenex(),

21:13 <mattip> the _ctypes.CDLL seems to use both rlib/libffi.py and rlib/clibffi.py

21:13 <mattip> so I think I will leave that for a bugfix if anyone complains