AttributeError: 'NoneType' object has no attribute 'file' — module-level async def with nested def / async for / async with #31

Describe the bug

codegraph (1.2.0, current PyPI release) crashes with

AttributeError: 'NoneType' object has no attribute 'file'
  File ".../codegraph/parser.py", line 74, in _nest_function
    newfunc = Function(func_name, ob.file, lineno, ob)

whenever a module-level async def contains any of:

• a nested def / async def
• an async for statement or comprehension
• an async with statement

Minimal repro

Three lines, no third-party code involved:

async def foo():
    return [x async for x in bar()]

$ codegraph repro.py
AttributeError: 'NoneType' object has no attribute 'file'

Control cases that do not crash (only the module-level async def variant does):

Impact

The whole tree is parsed in one pass, so a single offending file makes codegraph <dir> fail for the entire directory. On a real-world Python project this made 16 of 326 modules unparseable, i.e. any directory containing one of them could not be analysed at all.

Root cause

create_objects_array() uses tokenize plus an indentation stack to resolve parent/child relations. Two defects in the async branch (codegraph/parser.py:199-218):

1. The token after async is never validated. It is discarded via _, _, _ = next(g)[0:3] (comment: "next will be def") and the next NAME is used as the function name, so async for / async with produce a bogus function — in async for x in ... the "name" becomes x.
2. None is pushed on the stack for a module-level async def. The else branch only does tree.append(AsyncFunction(...)) without assigning cur_func, then stack.append((cur_func, thisindent)) pushes None unconditionally. The synchronous def branch assigns cur_func and guards the push with if cur_func:.

The None stays at the bottom of the stack for the entire function body (indent 0 is never popped by anything at a deeper indent), so the next nested def / async token calls _nest_function(None, ...) → ob.file → AttributeError.

Neither defect crashes on its own — only the combination does, which is why the failure is partial rather than total.

Environment

• codegraph 1.2.0 (also reproduced on 1.1.0)
• Python 3.13 / 3.14, macOS

Additional context

The in-flight Release/1.3.0 (#26) replaces the tokenizer parser with an AST-based one; I verified that branch parses all 326 modules of the affected project without error, so 1.3.0 should resolve this on the default path. Two follow-ups:

• users on the released 1.2.x still hit this today — I opened a PR with a minimal patch for that line;
• the legacy codegraph/parser.py on the release/1.3.0 branch still carries both defects, so it is worth patching there as well if that path stays reachable.