ci: auto-close PRs whose code is harvested into main

Solves the long-standing hygiene problem where contributor PRs whose
code lands via maintainer cherry-pick stay open + CONFLICTING forever,
even though their fix is credited in the CHANGELOG. v0.8.29 alone left
~5 such PRs open (#1421, #1429, #1442, #1465 — verified separately).

New workflow `.github/workflows/auto-close-harvested.yml`:

  * Triggers on push to main.
  * For each commit in the push, scans the commit body for lines
    matching `harvested from (PR )?#N` (case-insensitive).
  * For each matched PR number, closes the PR with a templated
    thank-you that links back to the merged commit, thanks the
    contributor by handle, and points them at CONTRIBUTING for
    landing future PRs via the faster direct-merge path.
  * Idempotent — already-closed PRs are skipped with a log line,
    not an error.
  * Concurrency-guarded so two near-simultaneous main pushes can't
    both try to close the same PR.

Two commit-message patterns are recognised:

  * `Harvested from PR #1234 by @username`  (preferred form, used
    in the templates the maintainer paste-uses for harvests)
  * `harvested from #1234`                  (case-insensitive
    fallback for older / shorter forms)

The convention is documented in CONTRIBUTING.md, which also adds a
new "How Your Contribution Lands" section explaining the harvest
model upfront so contributors know what to expect — closing their
PR isn't rejection, and the credit lives in the commit message and
CHANGELOG.

Permissions on the workflow: `pull-requests: write` to close + comment,
`issues: write` for the comment (PR comments are issue comments under
the hood), `contents: read` for the checkout.
This commit is contained in:
Hunter Bown
2026-05-11 22:49:28 -05:00
parent 2e8322bb84
commit 80fc0046e1
2 changed files with 197 additions and 0 deletions
+145
View File
@@ -0,0 +1,145 @@
name: Auto-close harvested PRs
# When a commit on main contains a "Harvested from PR #N" line in its
# message, close PR #N with a templated thank-you that links back to
# the merged commit. Solves the long-standing problem where contributor
# PRs whose code lands via maintainer cherry-pick stay open and
# `CONFLICTING` forever, even though their fix is credited in the
# CHANGELOG.
#
# The expected commit-message convention is documented in
# CONTRIBUTING.md. Two patterns are recognised:
#
# * `Harvested from PR #1234 by @username` (preferred)
# * `harvested from #1234` (case-insensitive fallback)
#
# The first match's PR number is closed; multiple PRs can be closed
# per commit by repeating the line. The match runs on the commit
# body only, not on the subject line, so the subject can describe
# the change naturally without baking a number into it.
on:
push:
branches: [main]
permissions:
contents: read
pull-requests: write
issues: write
# Only one auto-close run at a time so two near-simultaneous main
# pushes can't both try to close the same PR (the second would just
# fail with "Pull request is already closed", harmless but noisy).
concurrency:
group: auto-close-harvested
cancel-in-progress: false
jobs:
close:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
with:
# We need at least the commits that this push introduced.
# fetch-depth: 0 is the simplest correct option; the
# alternative (fetching just `before..after`) is fragile
# when force-pushes happen.
fetch-depth: 0
- name: Close PRs referenced by harvested-from lines
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
BEFORE_SHA: ${{ github.event.before }}
AFTER_SHA: ${{ github.event.after }}
shell: bash
run: |
set -euo pipefail
# The first push to a fresh branch has BEFORE_SHA = 0000…0000.
# In that case fall back to the latest commit only — we don't
# want to scan the entire history.
if [[ "${BEFORE_SHA}" == "0000000000000000000000000000000000000000" || -z "${BEFORE_SHA:-}" ]]; then
RANGE="${AFTER_SHA}"
RANGE_ARGS=("-1" "${AFTER_SHA}")
else
RANGE="${BEFORE_SHA}..${AFTER_SHA}"
RANGE_ARGS=("${RANGE}")
fi
echo "Scanning commit range: ${RANGE}"
# `git log --format=%H%n%B%n--END--` separates commits with a
# sentinel so multi-line bodies don't get mangled.
mapfile -t commits < <(git log "${RANGE_ARGS[@]}" --format="%H")
if [[ ${#commits[@]} -eq 0 ]]; then
echo "No commits in range; nothing to do."
exit 0
fi
declare -A processed_prs=()
for sha in "${commits[@]}"; do
body="$(git log -1 --format=%B "${sha}")"
# Two patterns, both case-insensitive on the keyword:
# "Harvested from PR #1234 by @username" (preferred form)
# "harvested from #1234" (short form)
mapfile -t pr_numbers < <(
printf '%s\n' "${body}" \
| grep -oiE 'harvested from (pr )?#[0-9]+' \
| grep -oE '#[0-9]+' \
| tr -d '#' \
| sort -u || true
)
if [[ ${#pr_numbers[@]} -eq 0 ]]; then
continue
fi
short_sha="${sha:0:12}"
subject="$(git log -1 --format=%s "${sha}")"
for pr in "${pr_numbers[@]}"; do
key="${pr}-${sha}"
if [[ -n "${processed_prs[${key}]:-}" ]]; then
continue
fi
processed_prs[${key}]=1
# Idempotency: skip if the PR is already closed.
state="$(gh pr view "${pr}" --json state --jq .state 2>/dev/null || echo "MISSING")"
if [[ "${state}" == "CLOSED" || "${state}" == "MERGED" ]]; then
echo "PR #${pr} is already ${state}; skipping."
continue
fi
if [[ "${state}" == "MISSING" ]]; then
echo "::warning::PR #${pr} not found or inaccessible; skipping."
continue
fi
author="$(gh pr view "${pr}" --json author --jq '.author.login' 2>/dev/null || echo "")"
greeting="Hi"
if [[ -n "${author}" ]]; then
greeting="Thanks @${author}"
fi
body_text=$(cat <<EOF
${greeting} — your contribution landed in [\`${short_sha}\`](https://github.com/${GITHUB_REPOSITORY}/commit/${sha}) on \`main\`:
> ${subject}
Closing this PR now that the code is on \`main\`. Credit lives in the commit message and (where applicable) the \`CHANGELOG.md\` entry for the next release. Apologies for not closing this at the time of the merge — the auto-close workflow is new in v0.8.31.
If you want to land more work and would prefer your future PRs merge cleanly without a harvest step, the [\`CONTRIBUTING.md\`](https://github.com/${GITHUB_REPOSITORY}/blob/main/CONTRIBUTING.md) doc has a short note on what makes a contribution mergeable as-is.
EOF
)
echo "Closing PR #${pr} (harvested in ${short_sha})"
if ! gh pr close "${pr}" \
--repo "${GITHUB_REPOSITORY}" \
--comment "${body_text}"; then
echo "::warning::Failed to close PR #${pr}; continuing"
fi
done
done
echo "Auto-close pass complete."