bug(indexer): Python annotated parameters without a default (x: str) are silently dropped — 41.8% of parameters lost #1520

tools/languages/python.py:304-307

Parameter extraction looks up a name field on every parameter node:

child.child_by_field_name('name')

In tree-sitter-python, typed_default_parameter (x: int = 5) has a name field, but typed_parameter (x: int) does not — it exposes only type. So the lookup returns None and the parameter is silently discarded.

Reproduction

$ .venv/bin/python -c "
from codegraphcontext.tools.tree_sitter_parser import TreeSitterParser
from codegraphcontext.tools.languages.python import PythonTreeSitterParser
import tempfile, pathlib
p = pathlib.Path(tempfile.mkdtemp())/'t.py'
p.write_text('def m(self, x: str, y, z: int = 5): pass\n')
print(PythonTreeSitterParser(TreeSitterParser('python')).parse(str(p))['functions'][0]['args'])
"
['self', 'y', 'z']

x: str is gone. z: int = 5 survives only because it is a different node type.

Blast radius

Measured against the stdlib ast module over CGC's own src/:

real parameters (ast):        3154
extracted by the parser:      1835
lost:                         1319   (41.8%)

Every annotated-but-undefaulted parameter — the dominant style in any typed Python codebase, and in this repo — never becomes a Parameter node and never gets a HAS_PARAMETER edge. Since CGC's own source is heavily annotated, self-indexing loses over 40% of parameters.

Suggested fix

Handle typed_parameter explicitly: its identifier is the first identifier child rather than a name field.

_{Reproduced with the snippet above; the 41.8% figure measured by diffing parser output against ast over src/.}