Fix infinite semantic analysis loop when using from_queryset

[lee3jjang]

I have made things!

Related issues

• Closes Maximum semantic analysis iteration count reached error with django-stubs 5.0.3 & 5.0.4 #2373

Summary

This PR fixes the "maximum semantic analysis iteration count reached" error that was introduced in django-stubs 5.0.3.

Problem

When create_manager_info_from_from_queryset_call() returns None, the code unconditionally calls add_symbol_table_node() with a PlaceholderNode on every iteration. This sets mypy's progress = True, preventing the semantic analysis loop from terminating.

The issue was introduced in PR #2228 as a hack for python/mypy#17402, but the implementation causes infinite iterations when the from_queryset call cannot be resolved immediately.

Solution

Check if a PlaceholderNode already exists before adding a new one. This prevents redundant symbol table modifications that trigger infinite iterations while still maintaining the original hack for python/mypy#17402.

Changes

• Added a check for existing PlaceholderNode before calling add_symbol_table_node() in create_new_manager_class_from_from_queryset_method()

[UnknownPlatypus]

This would be nice to have a test showing the max iteration issue so that we can be sure this resolves it. See #2228 for an exemple

[lee3jjang]

I tried extensively to create a test that triggers the exact "maximum semantic analysis iteration count reached" error, but it was difficult because create_manager_info_from_from_queryset_call() returning None is transient - mypy resolves forward references after a few iterations.

The test test_from_queryset_repeated_deferral_with_forward_metaclass does exercise the None path (verified with debug output), but the forward metaclass resolves quickly before hitting max iterations.

Real-world validation: This fix resolved the error in my production codebase that was failing with django-stubs 5.0.3+.

[sobolevn]

Looks correct to me, thank you!