Conversation
This reverts commit 6098e9b.
…interpreter was built with no-GIL support
This reverts commit b4a48c1.
mhammond
left a comment
mhammond
left a comment
There was a problem hiding this comment.
I didn't get far here - I check the first dozen or so files and none of the changes I saw were related to the gil at all.
Co-authored-by: Avasam <samuel.06@hotmail.com>
This comment was marked as resolved.
This comment was marked as resolved.
This comment was marked as resolved.
This comment was marked as resolved.
| @@ -1,3 +1,5 @@ | |||
| Original patch message is as follows: | |||
There was a problem hiding this comment.
I'm all for a swig update (it's oooooooold, and I'd like to look into type-stubs autogeneration solutions, such as swig/swig#735 (comment)), but I have 1 concern, and 1 technical debt issue I'd like to raise up.
First, I would highly prefer for @mhammond (or another maintainer, could be me, but Mark himself would be best) to regenerate the binary files. Out of principle for open source trust, since binary files cannot meaningfully be inspected or reviewed. (the XZ Utils backdoor is still fresh in my memory).
Second, I haven't looked too deeply into the patching of the SWIG source code we do, but I wonder if it can be reduced. Are there new features/configurations released in the past 26-28 years that can help us? Is there anything we can request upstream ? (just like how I've actively been fixing and updating setuptools/distutils to remove our MSVCCompiler hack). If we could eventually replace the vendored SWIG by https://pypi.org/project/swig/ it would be the dream.
There was a problem hiding this comment.
Please do, as I don't have access to MSVC
If it's only PythonWin failing, I personally wouldn't mind partial free-threaded support that omits the editor. As Mark once said: #2450 (review)
|
| #line 640 "configure" | ||
| #include "confdefs.h" | ||
| -main(){return(0);} | ||
| +int main(){return(0);} |
There was a problem hiding this comment.
Are any of these new patch addition necessary?
As for removing redundant patches, as well as fixing bad patch file formatting, I open #2694
There was a problem hiding this comment.
Apparently so, as I was testing compilation with a modern clang version
There was a problem hiding this comment.
I mean, were any SWIG changes necessary, not to compile SWIG itself on modern tooling. But did you need a new binary for any of your other code changes here.
The only actual differences I can find in this patch are add int before main and removing a register keyword
There was a problem hiding this comment.
None for the SWIG interface files, afaik
There was a problem hiding this comment.
SWIG updated in #2694 to drop obsolete Python 2 support.
Looking at all the changes under SWIG/* it seems mostly like adding the source code and not files that were needed for free-threaded/no-GIL support ?
Co-authored-by: Avasam <samuel.06@hotmail.com>
3 participants
Should resolve #2303
Also interesting observations about test failures:
On no-GIL interpreters running on x86, only one test failure occurs: https://github.com/clin1234/pywin32/actions/runs/19752747214/job/56598607801#step:10:381
On x64, multiple test failures occur: https://github.com/clin1234/pywin32/actions/runs/19752747214/job/56598607814#step:10:463