Reloading a lazy module that has not yet been materailized should be no-op

[QEDady]

Bug report

Bug description:

Reloading a not-yet materialized lazy module is not supported well. It makes sense for the reloading in this case to be no-op since when the actual loading happens e.g. by accessing an attribute, the new code will be loaded anyway. Currently, the reload tries to reload the module while its loaded is lazy in the first place.

CPython versions tested on:

3.13, 3.12, 3.14, CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-139686: Make reloading a lazy module no-op #139857
• GH-139686: Update versionchanged directive to 'next' in importlib #142649
• GH-139686: Revert "gh-139686: Make reloading a lazy module no-op (GH-139857)" #143584

[ericvsmith]

Please show some example code. Show how it works, and what you expect the behavior to be.

[QEDady]

There is no way to surface the behaviour with the public API. Nevertheless, the private API for _LazyModule should be compatible with importlib.reload. I sent a fix in this PR

[encukou]

GH-139857 broke* refleak buildbots -- tests fail when executed twice:

$ ./python -m test -v test_importlib test_importlib -m 'LazyLoaderTests'
[...]
======================================================================
ERROR: test_attr_unchanged (test.test_importlib.test_lazy.LazyLoaderTests.test_attr_unchanged)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/encukou/dev/cpython/Lib/test/test_importlib/test_lazy.py", line 105, in test_attr_unchanged
    self.assertEqual(TestingImporter.mutated_name, module.__name__)
                                                   ^^^^^^^^^^^^^^^
  File "/home/encukou/dev/cpython/Lib/importlib/util.py", line 213, in __getattribute__
    raise ValueError(f"module object for {original_name!r} "
                      "substituted in sys.modules during a lazy "
                      "load")
ValueError: module object for 'lazy_loader_test' substituted in sys.modules during a lazy load

• similar errors for test_delete_eventual_attr and test_delete_preexisting_attr.

* [edit] The tests pass again when re-run, so buildbot status is “warn” rather that “fail”.

[brettcannon]

I'll revert the change today to fix the buildbots.

[brettcannon]

I have a revert PR out and set to auto-merge.

And now that PEP 810 has been accepted, I don't know how important it is to try again to get the functionality in or if it would be better on PyPI, maybe with #127036 .