Python: AST extractor misses ~53% of Python class inheritance edges (cross-file base classes)

[oskbjo]

When a Python class inherits from a base class imported from another file, the AST extractor fails to emit an inherits edge. Instead it produces a weaker uses [INFERRED] edge pointing at the import line rather than the class definition.

Repro:

  # module_a.py
  class BaseDriver:
      ...

  # module_b.py
  from module_a import BaseDriver

  class ChildDriver(BaseDriver):  # <-- should produce inherits [EXTRACTED]
      ...

Actual: ChildDriver --uses [INFERRED]--> BaseDriver (at import line)
Expected: ChildDriver --inherits [EXTRACTED]--> BaseDriver (at class def line)

Same-file inheritance is handled correctly. The issue is that the extractor doesn't resolve imported names back to their class node when the base class comes from another module.

Measurement on a ~900-file Python codebase:

• 820 inheritance relationships found via ast.ClassDef.bases
• 389 (47%) correctly emitted as inherits edges (same-file or simple cases)
• 431 (53%) missing, represented only as uses [INFERRED]

Impact:

• graphify path between parent and child classes routes through unrelated intermediate nodes instead of showing the direct inheritance chain
• Community detection separates parent/child classes that should cluster together
• Driver/plugin hierarchies with deep cross-file inheritance chains are especially affected

Environment: graphify 0.9.42, Python 3.12, Windows

[NithishKumar04]

I reproduced this on current v8 and traced it through final graph construction. The extracted inherits [EXTRACTED] edge is overwritten by a later uses [INFERRED] edge. Aliased imports and duplicate class names also need exact import-aware resolution.

I am preparing a focused fix with raw-extraction, built-graph, ambiguity, alias, and incremental regression coverage. I checked the overlapping #1231 and #2684 work and will keep this change compatible rather than duplicating their scope.