forkserver.set_forkserver_preload() silent preload import failures when sys.path is required

[doublex]

Bug report

Bug description:

Bug in def main(): https://github.com/python/cpython/blob/main/Lib/multiprocessing/forkserver.py#L167

The param sys_path is ignored. Result: ModuleNotFoundError for preloaded modules.

a) Using sys_path fixes this issue
b) Maybe better remove "pass" and report and error to simplify problem solving

    if preload:
        if '__main__' in preload and main_path is not None:
            process.current_process()._inheriting = True
            try:
                spawn.import_main_path(main_path)
            finally:
                del process.current_process()._inheriting
+        if sys_path:
+            sys.path = sys_path
        for modname in preload:
            try:
                __import__(modname)
            except ImportError:
-                pass
+                warnings.warn('forkserver: preloading module failed %s' % modname)

CPython versions tested on:

3.12

Operating systems tested on:

No response

Linked PRs

• gh-117378: Fix multiprocessing forkserver preload sys.path inheritance. #126538
• [3.13] gh-117378: Fix multiprocessing forkserver preload sys.path inheritance. (GH-126538) #126632
• [3.12] gh-117378: Fix multiprocessing forkserver preload sys.path inheritance. (GH-126538) #126633
• gh-117378: Clear up the NEWS entry wording #126634
• gh-117378: Only run the new multiprocessing SysPath test when appropriate #126635
• [3.13] gh-117378: Only run the new multiprocessing SysPath test when appropriate (GH-126635) #126652
• [3.12] gh-117378: Only run the new multiprocessing SysPath test when appropriate (GH-126635) #126653
• [3.13] gh-117378: Clear up the NEWS entry wording (GH-126634) #126668
• [3.12] gh-117378: Clear up the NEWS entry wording (GH-126634) #126669

[gpshead]

Thanks! This is an ironic bug. It has been there from the start for the forkserver. I suspect in a lot of environments it just didn't matter as the default path or PYTHONPATH from the environment was correct. But I can easily imagine situations where this would defeat the purpose of preloading or even potentially cause subtle bugs based on "" vs the absolute path determined at parent multiprocessing import time being in that preload's sys.path.

[doublex]

@gpshead
Thank you very much for your efforts!