feat(review): skip the automatic round when a synchronize only refreshes the base

.github/scripts/review-base-refresh-gate.sh

@@ -0,0 +1,149 @@
+#!/usr/bin/env bash
+# Decide whether an automatic `synchronize` review round may be skipped
+# because the push only refreshed the base branch (#10111): an update-branch
+# merge fires `synchronize` like any push, but a full automatic round costs
+# hours for a head whose PR-side diff is unchanged.
+#
+# Two facts are certified before any review compute is spent: every
+# first-parent commit since the head the last completed automatic round
+# reviewed is a two-parent merge of the base branch, and the PR's own
+# three-dot diff kept the same canonical digest — every changed and
+# context byte is hashed after stripping only unstable diff metadata
+# (index lines, hunk offsets), so an upstream edit touching the PR's own
+# hunks (whitespace included) fails the equality and the full round runs.
+# The reviewed head is the newest SUBMITTED ledger-marked review
+# (APPROVED, CHANGES_REQUESTED, or COMMENTED, with a submitted_at) posted
+# by the AUTHENTICATED account — resolved live via `gh api user`, same
+# norm as the fallback dedup — so a participant posting the marker text
+# in their own review, a PENDING draft, or a DISMISSED review can never
+# certify a head this gate would skip.
+#
+# Fail-open contract: every probe error or unmatched shape records
+# skip=false and the full round runs, and the script never exits non-zero —
+# the caller treats a step failure as a pipeline failure worth a fallback
+# comment, which a skipped gate is not.
+#
+# Outputs (GITHUB_OUTPUT): skip=true|false, reviewed_sha, reason.
+set -uo pipefail
+
+# Required environment, restated so a misconfigured caller dies naming the
+# missing input instead of probing git with empty arguments.
+PR_NUMBER="${PR_NUMBER:?}"
+EVENT_HEAD_SHA="${EVENT_HEAD_SHA:?}"
+BASE_REF="${BASE_REF:?}"
+RUN_URL="${RUN_URL:?}"
+GITHUB_REPOSITORY="${GITHUB_REPOSITORY:?}"
+GITHUB_OUTPUT="${GITHUB_OUTPUT:?}"
+GITHUB_STEP_SUMMARY="${GITHUB_STEP_SUMMARY:?}"
+
+SKIP=false
+REVIEWED_SHA=''
+REASON=''
+
+# Canonical digest of the diff between two commits: every changed and
+# context byte is hashed, stripping only unstable metadata (index lines,
+# hunk offsets). `git patch-id` is whitespace-insensitive and would
+# certify an indentation-only rewrite of PR-owned lines.
+diff_digest() {
+  local diff
+  diff="$(git diff "$1" "$2" 2>/dev/null)" || return 1
+  [[ -n "${diff}" ]] || return 0
+  printf '%s\n' "${diff}" |
+    sed -E '/^index [0-9a-f]+\.\.[0-9a-f]+/d; /^@@ /d' |
+    git hash-object --stdin
+}
+
+decide() {
+  REASON='pr head fetch failed'
+  git fetch --no-tags --quiet origin "refs/pull/${PR_NUMBER}/head" || return
+  REASON='event head not reachable from the PR head ref'
+  git cat-file -e "${EVENT_HEAD_SHA}^{commit}" 2>/dev/null || return
+  REASON='base branch ref unavailable'
+  if ! git rev-parse -q --verify "origin/${BASE_REF}^{commit}" >/dev/null; then
+    git fetch --no-tags --quiet origin "${BASE_REF}" || return
+    git rev-parse -q --verify "origin/${BASE_REF}^{commit}" >/dev/null || return
+  fi
+
+  REASON='reviewed-head lookup failed'
+  local bot_login reviews
+  bot_login="$(gh api user --jq '.login')" && [[ -n "${bot_login}" ]] || return
+  reviews="$(gh api "repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/reviews" --method GET --paginate -F per_page=100)" || return
+  REVIEWED_SHA="$(
+    printf '%s' "${reviews}" | jq -sr --arg login "${bot_login}" '
+      [.[][]
+       | select(.user.login == $login)
+       | select(.state == "APPROVED" or .state == "CHANGES_REQUESTED"
+                or .state == "COMMENTED")
+       | select(.submitted_at != null)
+       | select(.body | contains("<!-- qwen-review-ledger"))]
+      | last | .commit_id // empty'
+  )" || return
+  REASON='no completed automatic round on this PR'
+  [[ -n "${REVIEWED_SHA}" ]] || return
+  REASON='reviewed head not in the fetched history'
+  git cat-file -e "${REVIEWED_SHA}^{commit}" 2>/dev/null || return
+  REASON='reviewed head is not an ancestor of the pushed head'
+  git merge-base --is-ancestor "${REVIEWED_SHA}" "${EVENT_HEAD_SHA}" 2>/dev/null || return
+
+  # Bounded walk: a stale reviewed head must not buy an unbounded ancestry
+  # scan, and past ten first-parent steps this is not the cheap shape the
+  # gate exists for anyway.
+  local count c p2
+  REASON='commit walk failed'
+  count="$(git rev-list --first-parent --count "${REVIEWED_SHA}..${EVENT_HEAD_SHA}" 2>/dev/null)" || return
+  [[ "${count}" =~ ^[0-9]+$ ]] || return
+  REASON='no new commits since the reviewed head'
+  [[ "${count}" -ge 1 ]] || return
+  REASON="too many commits since the reviewed head (${count})"
+  [[ "${count}" -le 10 ]] || return
+  for c in $(git rev-list --first-parent "${REVIEWED_SHA}..${EVENT_HEAD_SHA}" 2>/dev/null); do
+    REASON="non-merge commit since the reviewed head: ${c}"
+    p2="$(git rev-parse -q --verify "${c}^2" 2>/dev/null)" || return
+    REASON="octopus merge since the reviewed head: ${c}"
+    if git rev-parse -q --verify "${c}^3" >/dev/null 2>&1; then return; fi
+    REASON="merge of a non-base branch since the reviewed head: ${c}"
+    git merge-base --is-ancestor "${p2}" "origin/${BASE_REF}" 2>/dev/null || return
+  done
+
+  local mb_r mb_h digest_r digest_h
+  REASON='merge-base resolution failed'
+  mb_r="$(git merge-base "origin/${BASE_REF}" "${REVIEWED_SHA}" 2>/dev/null)" || return
+  mb_h="$(git merge-base "origin/${BASE_REF}" "${EVENT_HEAD_SHA}" 2>/dev/null)" || return
+  REASON='PR-side diff computation failed'
+  digest_r="$(diff_digest "${mb_r}" "${REVIEWED_SHA}")" || return
+  digest_h="$(diff_digest "${mb_h}" "${EVENT_HEAD_SHA}")" || return
+  REASON='empty PR-side diff'
+  { [[ -n "${digest_r}" ]] && [[ -n "${digest_h}" ]]; } || return
+  REASON='the PR-side diff changed'
+  [[ "${digest_r}" = "${digest_h}" ]] || return
+
+  REASON=''
+  SKIP=true
+}
+decide
+
+{
+  echo "skip=${SKIP}"
+  echo "reviewed_sha=${REVIEWED_SHA}"
+  echo "reason=${REASON}"
+} >> "${GITHUB_OUTPUT}"
+if [[ "${SKIP}" != 'true' ]]; then
+  echo "Base-refresh gate: full round proceeds (${REASON})." >> "${GITHUB_STEP_SUMMARY}"
+  exit 0
+fi
+echo "Base-refresh gate: head ${EVENT_HEAD_SHA} only merges ${BASE_REF} into reviewed head ${REVIEWED_SHA}; skipping the automatic round." >> "${GITHUB_STEP_SUMMARY}"
+
+# One marker-deduped note per PR, updated in place via the shared upsert
+# helper (author-scoped lookup, retried). Blank line after the marker, or
+# none of the prose renders — the HTML-block quirk the ack comment documents.
+# Note failures only warn: the skip decision is already recorded above.
+body_file="$(mktemp "${RUNNER_TEMP:-/tmp}/qwen-review-base-refresh.XXXXXX")" || exit 0
+# shellcheck disable=SC2016
+printf '<!-- qwen-review-base-refresh -->\n\n🔁 **Base refresh detected** — head `%s` only merges `%s` into the last reviewed head `%s`; the PR-side diff is unchanged, so that review still applies and no new automatic round was spent ([this run](%s)). Comment `@qwen-code /review` for a fresh full review.\n\n<details>\n<summary>中文说明</summary>\n\n🔁 **检测到 base 刷新** —— head `%s` 仅把 `%s` 合入上次已评审的 head `%s`,PR 侧 diff 未变,原有评审结论仍然适用,本次未消耗新的自动评审轮次([本次运行](%s))。如需全新完整评审,请评论 `@qwen-code /review`。\n\n</details>' \
+  "${EVENT_HEAD_SHA}" "${BASE_REF}" "${REVIEWED_SHA}" "${RUN_URL}" \
+  "${EVENT_HEAD_SHA}" "${BASE_REF}" "${REVIEWED_SHA}" "${RUN_URL}" > "${body_file}"
+"$(dirname "${BASH_SOURCE[0]}")/upsert-bot-comment.sh" \
+  "${GITHUB_REPOSITORY}" "${PR_NUMBER}" '<!-- qwen-review-base-refresh -->' "${body_file}" \
+  || echo "::warning::could not post the base-refresh note"
+rm -f "${body_file}"
+exit 0

.github/workflows/qwen-code-pr-review.yml

@@ -674,14 +674,43 @@ jobs:
             echo "auto_review=$AUTO_REVIEW"
           } >> "$GITHUB_OUTPUT"
 
+      # An update-branch base refresh (merging current main into the PR
+      # branch) fires `synchronize` like any push, but adds nothing on the
+      # PR side — and a full automatic round costs hours (#10111: two base
+      # refreshes on PR #9729 spent one full round plus one cancelled one).
+      # The script certifies that shape deterministically (pure base-merge
+      # chain since the last submitted ledger-marked reviewed head,
+      # PR-side diff unchanged apart from hunk offsets and index lines)
+      # and posts one marker-deduped note when it
+      # skips; its full contract lives in its header. Everything fails OPEN
+      # into the full round — the script never exits non-zero (a step
+      # failure would be misreported as a pipeline failure by the fallback
+      # comment) and continue-on-error is the belt for an unexpected crash.
+      - name: 'Skip base-refresh-only synchronize'
+        id: 'base_refresh'
+        if: |-
+          steps.context.outputs.should_run == 'true' &&
+          steps.context.outputs.auto_review == 'true' &&
+          github.event.action == 'synchronize'
+        continue-on-error: true
+        timeout-minutes: 5
+        env:
+          GH_TOKEN: '${{ secrets.CI_BOT_PAT }}'
+          PR_NUMBER: '${{ steps.context.outputs.pr_number }}'
+          EVENT_HEAD_SHA: '${{ github.event.pull_request.head.sha }}'
+          BASE_REF: '${{ github.event.pull_request.base.ref }}'
+          RUN_URL: '${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}'
+        run: |-
+          .github/scripts/review-base-refresh-gate.sh
+
       - name: 'Setup Node.js for hosted review'
-        if: "steps.context.outputs.should_run == 'true' && (github.repository != 'QwenLM/qwen-code' || vars.MAINTAINER_ECS_RUNNER_DISABLED == 'true')"
+        if: "steps.context.outputs.should_run == 'true' && steps.base_refresh.outputs.skip != 'true' && (github.repository != 'QwenLM/qwen-code' || vars.MAINTAINER_ECS_RUNNER_DISABLED == 'true')"
         uses: 'actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e' # v6.4.0
         with:
           node-version: '22.x'
 
       - name: 'Install Qwen CLI if missing'
-        if: "steps.context.outputs.should_run == 'true'"
+        if: "steps.context.outputs.should_run == 'true' && steps.base_refresh.outputs.skip != 'true'"
         run: |-
           set -euo pipefail
           if command -v qwen >/dev/null 2>&1; then
@@ -702,7 +731,7 @@ jobs:
       # inside (set +e, trailing exit 0) are its suspenders. It only decides
       # which rung this runner can reach.
       - name: 'Install capture tools (tmux + freeze)'
-        if: "steps.context.outputs.should_run == 'true'"
+        if: "steps.context.outputs.should_run == 'true' && steps.base_refresh.outputs.skip != 'true'"
         continue-on-error: true
         # continue-on-error bounds failure, not duration: a hung `sudo
         # apt-get update` against a stalled mirror has no other bound and
@@ -853,7 +882,7 @@ jobs:
 
       - name: 'Run review'
         id: 'review'
-        if: "steps.context.outputs.should_run == 'true'"
+        if: "steps.context.outputs.should_run == 'true' && steps.base_refresh.outputs.skip != 'true'"
         env:
           GH_TOKEN: '${{ secrets.CI_BOT_PAT }}'
           OPENAI_API_KEY: '${{ secrets.REVIEW_OPENAI_API_KEY }}'