bump cibuildwheel action to v2.17.0

[arvidn]

No description provided.

[xavier2k6]

@arvidn try using Boost 1.82.0 instead.

[arvidn]

it looks like there's some progress. Now the failure is a rename that happens after the build, here.

rename build\lib.win-amd64-cpython-37\libtorrent.cp37-win_amd64.pyd -> build\lib.win-amd64-cpython-37\libtorrent\__init__.cp37-win_amd64.pyd
error: [WinError 2] The system cannot find the file specified: 'build\\lib.win-amd64-cpython-37\\libtorrent.cp37-win_amd64.pyd' -> 'build\\lib.win-amd64-cpython-37\\libtorrent\\__init__.cp37-win_amd64.pyd'

[xavier2k6]

It should make no difference to error but just as an FYI v2.18.0 of cibuildwheel action has been released.

[qstokkink]

Thanks for picking this up! I'd like to see this fixed, so I'll support to the best of my ability.

[Click me for analysis]

If I'm interpreting the log correctly, this is the "rename" log entry just before failure:

libtorrent/bindings/python/setup.py

Line 361 in 31003b7

log.info("rename %s -> %s", src, dst)

I assume the following os.rename(src, dst) is the cause of the crash.

On the other builds this succeeds. For example, Mac:

common.copy /Users/runner/work/libtorrent/libtorrent/bindings/python/build/lib.macosx-10.9-x86_64-cpython-37/libtorrent.cpython-37m-darwin.so
    ...updated 276 targets...
    rename build/lib.macosx-10.9-x86_64-cpython-37/libtorrent.cpython-37m-darwin.so -> build/lib.macosx-10.9-x86_64-cpython-37/libtorrent/__init__.cpython-37m-darwin.so

However, on Windows, build\lib.win-amd64-cpython-37\libtorrent.pyd is apparently created instead of the expected build\lib.win-amd64-cpython-37\libtorrent.cp37-win_amd64.pyd:

common.copy D:\a\libtorrent\libtorrent\bindings\python\build\lib.win-amd64-cpython-37\libtorrent.pyd
    bin\msvc-14.3\rls\adrs-mdl-64\cxstd-14-iso\pythn-3.7\libtorrent.pyd
            1 file(s) copied.
    ...updated 266 targets...
    rename build\lib.win-amd64-cpython-37\libtorrent.cp37-win_amd64.pyd -> build\lib.win-amd64-cpython-37\libtorrent\__init__.cp37-win_amd64.pyd

As I see it, you could remedy this by either changing the setup.py output-detection logic or the build logic that creates the output in the first place (probably in the jamfile).

[qstokkink]

@arvidn if you want some help and you don't mind me hammering your GitHub Actions builds for a bit, I can try getting the action (back) up and running. Just let me know and I'll pick this up.

[qstokkink]

According to the following line, the EXT_SUFFIX is empty on Windows. I'm not 100% sure but that might be causing the failure.

2024-05-11T16:35:18.8519849Z   using python : 3.7 : "C:\\Users\\runneradmin\\AppData\\Local\\Temp\\cibw-run-ffbtcev5\\cp37-win_amd64\\build\\venv\\Scripts\\python.exe" : "C:\\Users\\runneradmin\\AppData\\Local\\Temp\\cibw-run-ffbtcev5\\cp37-win_amd64\\build\\venv\\include" "C:\\Users\\runneradmin\\AppData\\Local\\pypa\\cibuildwheel\\Cache\\nuget-cpython\\python.3.7.9\\tools\\include" "C:\\Users\\runneradmin\\AppData\\Local\\pypa\\cibuildwheel\\Cache\\nuget-cpython\\python.3.7.9\\tools\\Include" : "C:\\Users\\runneradmin\\AppData\\Local\\Temp\\cibw-run-ffbtcev5\\cp37-win_amd64\\build\\venv\\libs" : <libtorrent-python>on : "" ;

I think this may be due to #7649 switching from distutils.sysconfig.get_config_var("EXT_SUFFIX") to sysconfig.get_config_var("EXT_SUFFIX"). The former contains a workaround specific to Python 3.7 for Windows (see here). If this is the case, then you need to add in a workaround for sys.version_info < (3, 10) and platform.system() == 'Windows' in your setup.py to set the EXT_SUFFIX to _imp.extension_suffixes()[0] for that specific platform.

[arvidn]

the next issue:

+ bash -c "cd 'D:\a\libtorrent\libtorrent/bindings/python' && python test.py"
  Error: Command bash -c "cd 'D:\a\libtorrent\libtorrent/bindings/python' && python test.py" failed with code 2816. None

[arvidn]

exit code 2816, according to this means:

The EventMapping table refers to an invalid control [4] on dialog [2] for the event [3]

Possibly some issue related to the program running in a console, without a GUI.

the other python test fails with exit code 0xC0000005, which is ACCESS VIOLATION

[xavier2k6]

exit code 2813

@arvidn Where did you see this?

#7672 (comment) shows 2816

The EventMapping table refers to an invalid control [4] on dialog [2] for the event [3]

[arvidn]

https://github.com/arvidn/libtorrent/actions/runs/9427920137/job/25972643164?pr=7672#step:5:995

oops, you're right

[qstokkink]

Almost there 🎉 I think 2816 is simply a segfault (11) in disguise due to bash running on Windows (for reasons explained here).

It would be easier to debug if we could check out the .whl that was produced. Could you skip the test command and share the wheel?