bug(graph): graph_name is silently discarded for dead_code and find_complexity, querying the default graph instead of the requested one #1531

tools/code_finder.py — self._active_graph = graph_name at :1204, clobbered by :792 (find_dead_code) and :1379 (find_most_complex_functions)

analyze_code_relationships sets the active graph on the shared instance field, then dispatches to inner methods at :1275 and :1284 without forwarding graph_name. Both inner methods default the parameter to None and re-assign the same instance field, so self.driver resolves to db_manager.get_driver(None) — the default graph.

Reproduction

Instrumenting get_driver to record the graph name it receives:

dead_code       -> [None]      # graph_name="mygraph" was passed in
find_complexity -> [None]
find_callers    -> ['mygraph'] # control: forwards correctly

Impact

FalkorDB is the default backend and is multi-graph, with one graph per indexed repository. So:

analyze_code_relationships(query_type="dead_code", graph_name="myrepo")

silently queries the default graph and returns dead-code results for a different repository — or an empty list if the default graph is empty. Same for find_complexity. The caller gets a plausible-looking answer about the wrong codebase, with no error.

Suggested fix

Forward graph_name explicitly at both call sites:

-return self.find_dead_code(...)
+return self.find_dead_code(..., graph_name=graph_name)

More robustly, stop routing graph selection through mutable instance state — passing it as a parameter throughout removes this whole class of clobbering.

_{Reproduced by instrumenting get_driver and invoking each query type.}