search

conda-forge / lief-feedstock

A conda-smithy repository for lief.
BSD 3-Clause "New" or "Revised" License
1 stars 15 forks source link

Windows builds fail on master #11

Closed mbargull closed 4 years ago

mbargull commented 4 years ago 
2020-07-29T19:34:40.0295905Z import: 'lief'
2020-07-29T19:34:40.0525917Z Fatal Python error: PyThreadState_Get: no current thread
2020-07-29T19:34:40.0527722Z 
2020-07-29T19:34:42.7685337Z Tests failed for py-lief-0.10.1-py37h1834ac0_0.tar.bz2 - moving package to D:\bld\broken

Successful run from gh-10: https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=185904 https------dev.azure.com--conda-forge--84710dde-1620-425b-80d0-4cf5baca359d--_apis--build--builds--185904--logs--31.log https://dev.azure.com/conda-forge/84710dde-1620-425b-80d0-4cf5baca359d/_apis/build/builds/185904/logs/31

Failed run on master: https://dev.azure.com/conda-forge/feedstock-builds/_build/results?buildId=185926 https://dev.azure.com/conda-forge/84710dde-1620-425b-80d0-4cf5baca359d/_apis/build/builds/185926/logs/51 https------dev.azure.com--conda-forge--84710dde-1620-425b-80d0-4cf5baca359d--_apis--build--builds--185926--logs--51.log

The build from gh-10 failed after merging to master. I restarted it today, but still the same error. It's a bit strange that we get this failure now with 3.7 and not 3.8 as we did before.

isuruf commented 4 years ago

Maybe the patch that we reverted in 3.8 got backported to 3.7?

mbargull commented 4 years ago

Ah, good call. Will check.

mbargull commented 4 years ago

Hm, doesn't look like it did: https://github.com/python/cpython/blob/v3.7.8/Python/dynload_win.c#L221

(Assuming we are talking about https://github.com/conda-forge/python-feedstock/blob/cf7ba76a05cbf67883306ae2a83d3c626b94f172/recipe/patches/0026-Revert-part-of-https-bugs.python.org-issue33895-http.patch#L50 )

mingwandroid commented 4 years ago

3.7 doesn't need this patch. Arguably 3.8 doesn't either now. Still, any other lib that calls into the Python DSO from any static initialization will result in the exact same problem and I kind of feel that we should keep our Python patched until this issue has been shaken out of the Windows/Python 3.8 ecosystem. Alternatively we could change my default to go along with upstream and keep the env var around as a way to test for this particular problem if we suspect we have run into it again.

Thoughts welcome!