[vitali87]

Problem

materialize_group_files used unidiff's is_removed_file to decide whether to delete a path. That property is a heuristic: it is also true for any single hunk whose target length is 0 — i.e. a file that dev truncated to empty (@@ -1,20 +0,0 @@, target still b/a.txt). The sub-PR therefore deleted the file: its tree differed from dev (a.txt | 0) and merging the sub-PR back conflicted (modify/delete). Found by the end-to-end hunt.

Fix

Treat a path as deleted only when patch_file.target_file == "/dev/null" (the first clause of unidiff's own check). Verified in a real repo with a file emptied, a file deleted, an empty file filled, and an ordinary edit: the sub-PR git ls-tree -r now equals dev; on base the emptied file is missing.

Tests

Emptied file materialises as "" (fails on base); real /dev/null deletion still yields None.

Local review: 5/5. 452 tests pass, ruff clean. Stacked on #68 (same function). Follow-up: #90's target_file_modes uses the same heuristic and will get the same one-line change.

[coderabbitai]

Warning

Review limit reached

Next included review available in 51 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 5115b61e-2197-478b-bfc9-4b8205aed3f5

📥 Commits

Reviewing files that changed from the base of the PR and between 671a081 and 8a7f17e.

📒 Files selected for processing (2)

• pr_split/diff_ops/reconstructor.py
• tests/test_reconstructor.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}