Conversation
|
Good idea, lemme merge and try to run a pipeline real quick... |
|
Don't think it works unfortunately (it runs a pipeline on the repo in my profile but not here https://github.com/da-woods/line_profiler/actions/runs/17033461100/job/48280656365) |
|
... oof, didn't see your edit on the other PR. Back to the drawing board then. Thanks for the proposal though! (It's kinda difficult for me to work on Windows stuff since I don't have a Windows machine; so an extra pair of eyes are always extra welcome.) |
|
I have a Windows machine but unfortunately not the right kind of Windows machine for this. If I think of anything else to try I'll suggest it though. I wouldn't spend too much time on this though - it was something I spotted while trying to test Cython on Windows+arm. It's fixed by just turning off a few tests. So it's not really blocking me. |
* Narrow scopes where 3.12 hacks apply
* Update CI to target and test ARM64 Windows
* Fixed expressions and unset ${VSCMD_ARG_TGT_ARCH}
* pre-include `pycore_atomic.h` to circumvent bad funcdefs
* temporarily truncate job matrix to focus on Windows-ARM64
* sidestep inclusion of `pycore_atomic.h`, just mock what we need
* temporarily stop building Windows-AMD64 and ABI3 wheels
* Added missing #define
* Try moving Python internal includes into the C file
* simplify macros
* roll back #392
* vendor in `_Py_atomic_int` from `pycore_atomic.h`
* Skip unavailable platforms in CI
* Allow Cython tests to XFAIL on Windows-ARM64
* Fixed attempted import on the failing path
* Undo eda8d0b and f407d00, restoring the rest of the pipeline
* Temp: truncate pipeline to debug test on Win-ARM64
This reverts commit 3d47c90.
* Set MSVC arch for ARM runners
* Revert soft-XFAIL of Cython test
This partially reverts commit 0a16f5c
but keeps the typing fixes introduced therein.
* Un-truncate pipeline
This reverts commit 072106d.
* Streamline pipeline, added comments
* changelog
* Make suggested changes
Co-authored-by: Jon Crall <erotemic@gmail.com>
---------
Co-authored-by: da-woods <dw-git@d-woods.co.uk>
Co-authored-by: Jon Crall <erotemic@gmail.com>
2 participants
This is my proposed (but untested) fix.
I think the issue is that the C++ code can see the atomic definitions. If they're kept hidden entirely within the C file (and so not visible to C++) then it might work.