Description

StanFromIreland

opened on Jul 23, 2026

Issue body actions

Bug report

Bug description:

See for example, this run:

======================================================================
FAIL: test_concurrent_lazy_import_reification (test.test_lazy_import.ThreadSafetyTests.test_concurrent_lazy_import_reification)
Multiple threads racing to reify the same lazy import should succeed.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/var/lib/buildbot/workers/rise-riscv64-3/3.x.rise-riscv64-3.nogil/build/Lib/test/test_lazy_import/__init__.py", line 1820, in test_concurrent_lazy_import_reification
    self.assertEqual(errors, [], f"Errors occurred: {errors}")
    ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Lists differ: [(9, AttributeError("module 'test.test_laz[39 chars]'"))] != []

First list contains 1 additional elements.
First extra element 0:
(9, AttributeError("module 'test.test_lazy_import.data' has no attribute 'basic2'"))

+ []
- [(9,
-   AttributeError("module 'test.test_lazy_import.data' has no attribute 'basic2'"))] : Errors occurred: [(9, AttributeError("module 'test.test_lazy_import.data' has no attribute 'basic2'"))]

----------------------------------------------------------------------
Ran 128 tests in 78.334s

FAILED (failures=1)
test test_lazy_import failed

CPython versions tested on:

CPython main branch

Operating systems tested on:

No response

Linked PRs

• gh-154561: Check module dict after trying to load submodule #154742

Activity

1. StanFromIreland

   added

   type-bugAn unexpected behavior, bug, or errorAn unexpected behavior, bug, or error

   testsTests in the Lib/test dirTests in the Lib/test dir

   topic-free-threading

   topic-lazy-imports

   interpreter-core(Objects, Python, Grammar, and Parser dirs)(Objects, Python, Grammar, and Parser dirs)

   and removed

   testsTests in the Lib/test dirTests in the Lib/test dir

   on Jul 23, 2026
2. johnslavik commented on Jul 23, 2026

   

   johnslavik

   on Jul 23, 2026

   Member

   More actions

   @pablogsal
3. johnslavik commented on Jul 23, 2026

   

   johnslavik

   on Jul 23, 2026

   Member

   More actions

   Oh, I thought this was new but it's an old one. I'll try to work on this. I reproduced it locally.
4. johnslavik

   self-assigned this

   on Jul 23, 2026
5. BHUVANSH855 commented on Jul 24, 2026

   

   BHUVANSH855

   on Jul 24, 2026

   Contributor

   More actions

   Hi! I've been investigating the intermittent failure in test_concurrent_lazy_import_reification and have been able to reproduce it locally.

   So far I've instrumented both the C and Python sides of the lazy import path. Here's what I've observed:

   • setattr(parent_module, child, module) succeeds, and an immediate assert getattr(parent_module, child) is module always passes.
   • _find_and_load_lazy_submodule() is just a wrapper around _find_and_load(..., lazy_submodule=True).
   • _find_and_load_unlocked() publishes the child module on the parent and doesn't appear to remove it afterwards.
   • During a failing run, I see the following sequence:
     • the attribute is initially missing (has_key=0, contains=1),
     • it is successfully published,
     • the pending entry is discarded,
     • later another thread still gets AttributeError, and by then the pending-set entry is already gone (contains=0).

   At this point I'm not seeing any Python-level code that removes the published attribute after setattr(), so I'm wondering if I'm looking in the wrong place.

   Would you recommend continuing to investigate the free-threaded module attribute implementation / synchronization, or is there another part of the lazy import implementation that would be a better place to look?
6. brittanyrey commented on Sep 18, 2026

   

   brittanyrey

   on Sep 18, 2026

   Contributor

   More actions

   This no longer repros.
   It looks like it was fixed when the fix for #155194 went in.
   cc: @johnslavik to close
7. brittanyrey

   added a commit that references this issue on Sep 18, 2026

   Fix free-threaded race reifying a lazy submodule attribute

   83b692b
8. johnslavik commented on Sep 18, 2026

   

   johnslavik

   on Sep 18, 2026

   Member

   More actions

   I've stress-tested the test under a higher number of threads and confirm this no longer reproduces. Thank you @brittanyrey!

   Reacted by Brittany Reynoso
9. johnslavik

   closed this as completedon Sep 18, 2026
10. StanFromIreland

    closed this as a duplicateof lazy imports allow lazy import math.pi #155194 on Sep 18, 2026

Issue actions