Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

[Bug]: changed-files rebuild (hook/watch) evicts placeholder nodes for files that were never scanned

Closed Beginner friendly
  • Is something wrong?
#4,160 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

  • #4161 by @rohit-jsfreaky — closed without merging

Assessment

Difficulty
2/5
Estimated time
1-3 hours
Newbie friendliness
85/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
python
Domain
backend

Research direction

The bug lives in the _reconcile_existing_graph function's corpus sweep logic; start there to modify the eviction check for missing source paths. Use the previous scan's manifest.json (only rewritten after reconcile finishes) as deletion evidence: only evict a missing path if it was present in the last manifest, otherwise preserve placeholder nodes for files that were never scanned. Run the provided repro.py script to confirm placeholder nodes and their edges are retained across changed-files rebuilds, matching full rebuild output.

Written by the indexing model from the issue text.

Description

bug
Pre-flight checks
  • I have checked the Troubleshooting section in the README
What happened?

An extractor can mint a node for a file that is not in the checkout: the target of an unresolved dynamic import (import('./queue.js')), or a project a .sln / .slnx / <ProjectReference> names. The node carries the missing path as its source_file, but it is the referring file's output (label ./queue.js, edges only from the referrer).

On a changed-files rebuild (_rebuild_code(changed_paths=...), used by the post-commit hook and graphify watch), the corpus sweep in _reconcile_existing_graph reads that missing path as a deleted source and evicts the node. The unchanged referrer's preserved edges to it then dangle and are dropped. A full rebuild of the same tree keeps both, so after every hook or watch rebuild the graph silently differs from a full rebuild, and it flips back on the next graphify update.

This goes against CONTRIBUTING ("Cross-file edges: must be preserved during incremental updates") and against the intent of #1795 ("require deletion evidence before evicting a missing source"): a path that was never scanned has no deletion evidence.

On graphify's own repo (975 files, v8 5c7b847), editing one unrelated file and running the changed-files rebuild loses 8 nodes and 15 edges compared with a full rebuild of the same tree: the placeholders for tests/fixtures/dynamic_import.ts's missing ./queue.js, ./mayaEngine.js, ./staticHelper, and the .csproj projects sample.sln, sample.slnx and sample.csproj reference.

Related to #3570 (same symptom class); the commenters there found that cause fixed by 32b209a, and this is a different mechanism that still reproduces on 0.9.77.

Steps to reproduce
cat > repro.py <<'EOF'
import json, tempfile
from pathlib import Path
from graphify.watch import _rebuild_code

def placeholder(c):
    g = json.loads((c / "graphify-out/graph.json").read_text(encoding="utf-8"))
    nodes = [n["id"] for n in g["nodes"] if n.get("source_file", "").endswith("queue.js")]
    edges = [(e["source"], e["relation"]) for e in g["links"] if e["target"] in nodes]
    return nodes, edges

if __name__ == "__main__":
    c = Path(tempfile.mkdtemp()) / "corpus"; c.mkdir()
    (c / "app.ts").write_text("export async function load() {\n  return import('./queue.js');\n}\n")
    (c / "other.ts").write_text("export function other() {\n  return 1;\n}\n")
    _rebuild_code(c, no_cluster=True, acquire_lock=False, force=True)
    print("full rebuild:              ", placeholder(c))
    (c / "other.ts").write_text("export function other() {\n  return 2;\n}\n")
    _rebuild_code(c, changed_paths=[Path("other.ts")], no_cluster=True, acquire_lock=False, force=True)
    print("changed-files (other.ts):  ", placeholder(c))
    _rebuild_code(c, no_cluster=True, acquire_lock=False, force=True)
    print("full rebuild, same tree:   ", placeholder(c))
EOF
PYTHONHASHSEED=0 python repro.py
Error output or graph output
full rebuild:               (['queue'], [('app_load', 'imports_from'), ('app', 'dynamic_import')])
changed-files (other.ts):   ([], [])
full rebuild, same tree:    (['queue'], [('app_load', 'imports_from'), ('app', 'dynamic_import')])

On graphify's own repo, after a one-file edit: changed-files rebuild 18,617 nodes / 37,920 edges vs full rebuild of the same tree 18,625 / 37,935.
Graphify version

0.9.77 (5c7b847)

Operating System

Windows

Python Version

3.12

Installation Method

built from source (git clone)

Additional Environment Details

AST-only, no provider environment variables, PYTHONHASHSEED=0. A clean checkout of v8 at 5c7b847 reproduces it. Found while comparing changed-files and full rebuilds of graphify's own repo.

Additional context

Fix ready: use the previous scan (manifest.json, rewritten only after reconcile) as the deletion evidence. A missing path that was in the last scan is evicted as before; one that was never scanned is kept while a referrer that is not re-extracted this run still points at it. No manifest: old behaviour. PR right after this.

Dominant language
Python
Stars
124k
Forks
11.9k
No merged PRs in 30d

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from Graphify-Labs/graphify

All issues in Graphify-Labs/graphify

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.