check-merge-drops.yml

Reports content that a merge commit silently dropped. For every merge commit in a range, with parents P1, P2 and merge base B, it collects the lines each side added since the merge base (git diff B P1 and git diff B P2, added lines only) and reports those that appear nowhere in the merge’s tree.

That is what resolving a conflict by keeping one side of a whole file looks like. Every line the other side added to that file vanishes, while the merge result reads as a clean, plausible file, so no check that reads only the current tree can see the loss. A line added on one side since the merge base is new to the history, so the other side cannot have deleted it; a merge that keeps both sides’ hunks keeps it, and a clean merge reports nothing.

Deliberately not reported:

• Moved, rewrapped, or bullet-split content. A line is reported only when its text occurs nowhere in the merge’s tree. Whitespace runs are collapsed and line-leading list item markers (-, *, +, 1.) are stripped when comparing flattened text, so a paragraph the resolution moved to another file, rewrapped at different line breaks, or split into a bulleted list passes.
• Reworded lines. When both sides edited the same line and the resolution wrote a third version combining them, neither side’s line survives verbatim, yet nothing was lost. A missing line is not reported when a line in the merge’s copy of the same file that is new in the merge (not present in any parent) is at least similarity alike, word by word. Unrelated lines from other parents or the merge base cannot mask a dropped line.

The check is warn-only by default, since a merge can legitimately drop a line the other side superseded. The full report, with the dropped lines grouped by merge, parent and file, goes to the job summary, and each affected merge gets one warning annotation. On a pull request the range is the merges the PR adds; on a push it is the merges the push brought in. A squash merge is not a merge commit, so run the check on pull requests to see a branch’s merges before they are squashed.

Inputs

Input	Type	Default	Description
python-version	string	'3.x'	Python version to set up.
globs	string	'*.md *.qmd'	Space-separated git pathspecs whose added lines are checked (recursive by default). The search for moved content always covers the whole tree.
paths-ignore	string	''	Comma- or newline-separated glob patterns (relative paths) to skip; supports *, ?, and recursive **.
min-length	string	'20'	Minimum length, in characters after whitespace normalisation, for an added line to be checked; inclusive. Shorter lines (fences, list markers, closing :::) recur everywhere and are skipped.
similarity	string	'0.6'	Word-level similarity, 0 to 1, at or above which a line in the merge’s copy of the same file that is new in the merge counts as a rewording of a missing line. '1' reports every line whose exact text is gone.
fail	boolean	false	Fail the workflow when a merge dropped content; otherwise warn only.
runs-on	string	empty (CI_RUNNER variable, else ubuntu-latest)	Runner label for the job. Use a self-hosted label only for a private repository; never for a public one, whose pull requests can run arbitrary code on it (#1167).

Permissions

Only contents: read (the default), so no permissions: block is needed.

Example

# Copy to .github/workflows/check-merge-drops.yml in your repo.
# Reports content a merge commit silently dropped: lines either side added
# since the merge base that appear nowhere in the merge's tree -- what
# resolving a conflict by keeping one side of a whole file looks like.
# Warn-only by default; the report goes to the job summary.
# Only contents:read is needed (the default), so no permissions block is required.
name: Check merge commits for dropped content

on:
  push:
    branches:
      - main
  pull_request:
  workflow_dispatch:

jobs:
  check:
    uses: Morrison-Lab/gha/.github/workflows/check-merge-drops.yml@v3
    # with:
    #   globs: '*.md *.qmd *.R'               # pathspecs whose added lines are checked
    #   paths-ignore: 'CHANGELOG.md'          # skip known-noisy paths
    #   min-length: 30                        # skip lines shorter than this
    #   similarity: 1                         # report reworded lines too
    #   fail: true                            # block instead of warn
    #   runs-on: private-runner                # self-hosted runner label (private repositories only)

See examples/check-merge-drops.yml for the full caller stub.

Running it locally

The script takes the same settings as flags, so it can audit a repository’s existing history:

python3 check-merge-drops/check-merge-drops.py --base <older-commit> --head origin/main