fix(parsers/python): resolve self/cls-aliased method calls in the call graph

[gadievron]

Root cause

CallGraphBuilder._resolve_call_node (libs/openant-core/parsers/python/call_graph_builder.py) matched a method-call receiver only when it was the literal name self or cls. A receiver that is a local alias of self/cls — obj = self; obj.method(), or alias = cls; alias.method() inside a classmethod — fell through to _resolve_module_call, which finds no import/class for a plain local variable and returns None. The self/class-method call produced no call-graph edge, so the callee can look unreachable (a false negative in the reachable set).

Reproduction

class C:
    def target(self):
        return 1
    def caller(self):
        obj = self
        return obj.target()

Before: call_graph["m.py:C.caller"] is [] (the C.caller -> C.target edge is missing).

How

• New _collect_self_aliases(tree): scans the function body for local names single-bound to self/cls (a name assigned more than once, or ever rebound to anything else, is not treated as an alias — so no spurious edge).
• _extract_calls_from_code builds self_aliases = {'self','cls'} | _collect_self_aliases(tree) and passes it to _resolve_call_node.
• _resolve_call_node takes self_aliases (defaulting to {'self','cls'} so existing callers are unchanged) and resolves any <alias>.method() through the existing _resolve_self_call. The two prior self/cls branches collapse into one obj.id in self_aliases check.

This only adds edges that were previously dropped; it never removes one — so it raises reachability, never lowers it. The change is confined to the receiver-resolution branch (no new subsystem / data model).

Regression test

libs/openant-core/tests/parsers/python/test_call_graph_self_calls.py (+62 lines):

• test_self_alias_method_call_edge — obj = self; obj.target() edge present.
• test_cls_alias_method_call_edge — alias = cls; alias.target() edge present.
• test_reassigned_alias_does_not_force_self_edge — soundness guard: a name reassigned away from self is NOT aliased.

RED (before fix):

FAILED tests/parsers/python/test_call_graph_self_calls.py::test_self_alias_method_call_edge
FAILED tests/parsers/python/test_call_graph_self_calls.py::test_cls_alias_method_call_edge
2 failed, 5 passed in 157.48s

GREEN (after fix), full tests/parsers/python/:

7 passed in 50.40s

Compatibility

None. _resolve_call_node's new parameter is optional and defaults to the previous behavior. No public API change.

Notes

The canonical per-parser tests/parsers/python/test_callgraph_symmetry.py / test_python_schema_completeness.py do not yet exist in this repo (a pre-existing gap across all parsers, not introduced here); this fix extends the existing test_call_graph_self_calls.py.

Author notes

• Implementing lines: _collect_self_aliases (new), the obj.id in self_aliases receiver branch, and the self_aliases thread-through in _extract_calls_from_code.
• Input fixed: obj = self; obj.target() — previously no C.caller -> C.target edge; now resolved.
• Likely pushback: "wrong edge if the alias is reassigned?" — no; only single-unconditional = self/= cls bindings count, pinned by test_reassigned_alias_does_not_force_self_edge.

Coordination with open PRs

#112 adds a _build_alias_map to the same resolution path, but for function-value aliases (fn = helper; fn()) — its docstring notes class/method targets are out of scope. This PR handles the distinct self/cls alias case (obj = self; obj.method()), which #112 leaves unresolved. The two are complementary but touch the same code; they should be unified rather than stacked. Happy to rebase onto #112 and fold the self/cls aliasing into its alias map.

[gadievron]

Merge-order note (not a defect — flagging for landing order)

This is an explicit follow-up to #112 (same parsers/python/call_graph_builder.py) — consider adding Depends-on: #112. Union: _resolve_call_node must thread all three — local_types + aliases (#112) and self_aliases (this PR). Whichever lands second will conflict; keep all three params.