Repository navigation
[Bug]: PyPy Win64 wheels use incorrect version #20929
Description
Bug summary
I didn't upload wheels for PyPy on Windows for 3.5.0b1, because the version was incorrect: matplotlib-3.6.0.dev0+g83c1a5104b.d20210824-pp37-pypy37_pp73-win_amd64.whl
Code for reproduction
n/aActual outcome
This appears to be the case for master as well:
7191328 08-27-2021 23:04 matplotlib-3.5.0.dev1817+g944f06953e-cp37-cp37m-macosx_10_9_x86_64.whl
10661571 08-27-2021 23:12 matplotlib-3.5.0.dev1817+g944f06953e-cp37-cp37m-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
10001027 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-cp37-cp37m-manylinux_2_5_i686.manylinux1_i686.whl
10273115 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-cp37-cp37m-manylinux_2_5_x86_64.manylinux1_x86_64.whl
7183741 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e-cp37-cp37m-win_amd64.whl
7038001 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e-cp37-cp37m-win32.whl
7193266 08-27-2021 23:04 matplotlib-3.5.0.dev1817+g944f06953e-cp38-cp38-macosx_10_9_x86_64.whl
10705272 08-27-2021 23:12 matplotlib-3.5.0.dev1817+g944f06953e-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
10018929 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.whl
10302299 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-cp38-cp38-manylinux_2_5_x86_64.manylinux1_x86_64.whl
7181568 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e-cp38-cp38-win_amd64.whl
7037218 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e-cp38-cp38-win32.whl
7193297 08-27-2021 23:04 matplotlib-3.5.0.dev1817+g944f06953e-cp39-cp39-macosx_10_9_x86_64.whl
10679999 08-27-2021 23:12 matplotlib-3.5.0.dev1817+g944f06953e-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl
10000769 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.whl
10282650 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-cp39-cp39-manylinux_2_5_x86_64.manylinux1_x86_64.whl
7180141 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e-cp39-cp39-win_amd64.whl
7035920 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e-cp39-cp39-win32.whl
7143689 08-27-2021 23:04 matplotlib-3.5.0.dev1817+g944f06953e-pp37-pypy37_pp73-macosx_10_9_x86_64.whl
7346827 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-pp37-pypy37_pp73-manylinux_2_12_i686.manylinux2010_i686.whl
7329514 08-27-2021 22:53 matplotlib-3.5.0.dev1817+g944f06953e-pp37-pypy37_pp73-manylinux_2_12_x86_64.manylinux2010_x86_64.whl
7176417 08-27-2021 22:50 matplotlib-3.5.0.dev1817+g944f06953e.d20210827-pp37-pypy37_pp73-win_amd64.whl
Expected outcome
All wheels use the same version
Operating system
No response
Matplotlib Version
master
Matplotlib Backend
No response
Python version
No response
Jupyter version
No response
Other libraries
No response
Installation
No response
Conda channel
No response
Activity
QuLogic commentedon Sep 29, 2021
After a bunch of debugging, the problem appears to be the symlinks:
but I don't understand why the tree is only dirty for building PyPy on Windows only, and not any of the other Windows wheels.
tacaswell commentedon Sep 29, 2021
My knee-jerk guess is that distutils is not synced between cpython and pypy ? Are we using the same version of git on all of the windows machines that are building the wheels? If I am understanding this correctly, it looks like something has resolved the symlinks by relpacing them with a copy of the real file?
I am a bit confused about why we have symlinks (they came in via #17539 which I do not understand).
QuLogic commentedon Sep 29, 2021
With that name, it triggers symbolic icon support in GTK3, which recolours based on the theme. This doesn't work in GTK4 though, so we might need to restructure everything into an icon theme, and drop the symlinks entirely.
tacaswell commentedon Sep 30, 2021
If they do not have to be actual symlinks maybe just accept 2 copies of those files?
Is it worth raising this with the cibuildwheel folks? It might be something with how the repo is checked out?
QuLogic commentedon Oct 7, 2021
See also analysis on pypa/cibuildwheel#817 (comment)