Repository navigation
bug(cli/mcp): find_importers reads module.alias but the alias is stored on the IMPORTS relationship, so import_alias is always null #1515
Description
tools/code_finder.py:657
who_imports_module collects the alias from the Module node:
MATCH (file:File)-[imp:IMPORTS]->(module:Module)
...
WITH file, repo, COLLECT({
imported_module: module.name,
import_alias: module.alias, // <- always null
full_import_name: module.full_import_name
}) AS importsBut the writer stores the alias on the relationship, not the node (tools/indexing/persistence/writer.py:581 and :597):
r.alias = row.alias
...
SET r.alias = coalesce(row.alias, ""),Module nodes are never written with an alias property, so module.alias evaluates to null on every row. Note the relationship is already bound as imp in the same pattern — the data is right there.
Impact
The find_importers MCP tool reports import_alias: null for every result, including import numpy as np where the alias is the whole point of the query. An agent asking "what name is this module imported under here?" gets nothing back and has no way to tell that the field is broken rather than genuinely absent.
Suggested fix
- import_alias: module.alias,
+ import_alias: imp.alias,Environment
| Commit | c0e0bed (main) |
| Python | 3.12.3, Linux |
Verified by cross-checking the query's property reads against every alias write in writer.py.
Activity
github-actions commentedon Aug 1, 2026
Hi! 👋 Join our CodeGraphContext Discord channel to collaborate: https://discord.gg/dR4QY32uYQ
Shashankss1205 commentedon Aug 1, 2026
Cross-linking: this shares a root cause with item 3 of #1533 (the C++ toolkit reads
m.aliasinquery_tool_languages/cpp_toolkit.py:46).In both places the alias is read off the
Modulenode, butwriter.py:580-598only ever writes it to theIMPORTSrelationship:Whoever picks either one should fix both call sites together — the change is
m.alias→i.aliasin each. A repo-wide grep for.aliasagainstModulepatterns would be worth doing at the same time in case there is a third.chore(release): bump version to 0.5.5 (#1569)