Yicong-Huang commented on code in PR #8096:
URL: https://github.com/apache/texera/pull/8096#discussion_r3911023372
##########
.asf.yaml:
##########
@@ -111,6 +111,11 @@ github:
- context: Required Checks
- context: Check License Headers
- context: Validate PR title
+ # Every release/* label must be approved by that branch's release
+ # manager before the merge, so a merged PR's labels are exactly
+ # the branches it was backported to
+ # (.github/workflows/backport-approval-check.yml).
+ - context: Backport Approvals
Review Comment:
This registers `Backport Approvals` in a ruleset whose `ref_name.include`
also covers `refs/heads/release/v1.1` and `refs/heads/release/v1.2` — and
nothing on those branches can produce it.
Verified rather than inferred: the workflow, `release_branches.py` and
`release-branches.yml` all 404 on both release refs; `gh api
.../rules/branches/release%2Fv1.2` returns exactly the three contexts required
today; merged PR #7905 reports those same three. And `pull_request` /
`merge_group` runs use the workflow from the merge ref, which for a backport PR
is the release-branch base — confirmed against a run on #8124.
Fourteen PRs into `release/v1.2` are open right now, nearly all draft
backports this pipeline opened. They become unmergeable the moment this lands.
Scoping the context to the default branch closes it.
##########
.github/workflows/backport-approval-check.yml:
##########
@@ -0,0 +1,296 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Blocks the merge while a `release/*` label on the PR has not been approved by
+# that branch's release manager (.github/release-branches.yml).
+#
+# The point is that a merged PR's labels should be a truthful record. A label
+# only nominates a branch, so before this check a manager declined a backport
by
+# staying silent — and the label stayed on, which read afterwards as "this
+# shipped in 1.2" when it had not. Here declining is an action: the manager
+# either approves, or removes their label. Since the merge waits for every
+# remaining label to be approved, the labels on a merged PR are exactly the
+# branches Direct Backport Push then sends the fix to.
+#
+# The author's job is to ask the manager for that call, not to make it for
them.
+# Editing labels needs triage access, so outside contributors and Renovate or
+# Dependabot PRs cannot remove a label themselves — the failure report says so
+# and names who to ask.
+#
+# `Backport Approvals` is listed in .asf.yaml's required_status_checks, so this
+# job must produce a result on EVERY pull request, or a PR that never runs it
+# waits forever. That is why it is not conditioned on the title, the labels, or
+# the base branch: it always runs and reports success when there is nothing to
+# approve. Merge groups are re-evaluated rather than waved through, because a
+# queued PR's own checks are no longer consulted once it is queued; a group
+# whose PRs cannot be resolved fails closed rather than passing unchecked. Do
+# not rename the job — its display name is the required context.
+
+name: Backport Approval Check
+
+on:
+ pull_request:
+ types:
+ - opened
+ - reopened
+ - synchronize
+ - labeled
+ - unlabeled
+ # Retargeting arrives as `edited`, and the verdict depends on the base
+ # branch: a backport PR into release/* has nothing to approve, the same
+ # PR moved onto main does.
+ - edited
+ pull_request_review:
+ types:
+ - submitted
+ - dismissed
+ merge_group:
+
+# Read-only: this job never writes to the PR, and `pull_request_review` runs
+# with the base repository's token even for fork PRs.
+permissions:
+ contents: read
+ pull-requests: read
+
+concurrency:
+ group: backport-approvals-${{ github.event.pull_request.number || github.ref
}}
+ cancel-in-progress: true
+
+jobs:
+ backport-approvals:
+ # Do not rename — this display name is the required status check context
+ # referenced in .asf.yaml.
+ name: Backport Approvals
+ runs-on: ubuntu-latest
+ steps:
+ # The PR's own head (or the queued group's), so the config is the one
that
+ # will govern this change once it merges. Checking out the base would not
+ # be a security boundary anyway: on `pull_request` the workflow
definition
+ # itself comes from the PR. What bounds this is the event — a fork gets a
+ # read-only token and no secrets, and nothing here writes to the PR.
+ - name: Checkout
+ uses: actions/checkout@v7
+ with:
+ ref: ${{ github.event.pull_request.head.sha ||
github.event.merge_group.head_sha }}
+ persist-credentials: false
+
+ - name: Read release branches
+ id: targets
+ run: |
+ entries=$(python3 .github/scripts/release_branches.py
.github/release-branches.yml)
+ echo "entries=${entries}" >> "$GITHUB_OUTPUT"
+ echo "Release branch targets: ${entries}"
+
+ - name: Check release manager approvals
+ uses: actions/github-script@v9
+ env:
+ ENTRIES: ${{ steps.targets.outputs.entries }}
+ with:
+ script: |
+ const { owner, repo } = context.repo;
+ const managers = new Map(
+ JSON.parse(process.env.ENTRIES || "[]").map((e) => [
+ e.branch,
+ e.manager || "",
+ ])
+ );
+
+ // A merge group carries several PRs and its commits are the squash
+ // merges, so the number is parsed out of each subject — the same
way
+ // precheck.yml resolves them.
+ async function pullRequestNumbers() {
+ if (context.eventName !== "merge_group") {
+ return [context.payload.pull_request.number];
+ }
+ const mg = context.payload.merge_group;
+ const { data: comparison } = await
github.rest.repos.compareCommits({
+ owner, repo, base: mg.base_sha, head: mg.head_sha,
+ });
+ return comparison.commits.map((c) => {
+ const m = c.commit.message.split("\n",
1)[0].match(/\(#(\d+)\)\s*$/);
+ if (!m) throw new Error(`no PR number in the message of
${c.sha}`);
+ return Number(m[1]);
+ });
+ }
+
+ // Latest review state per login. COMMENTED never changes an
approval;
+ // a later CHANGES_REQUESTED or DISMISSED revokes one. listReviews
is
+ // documented to return reviews chronologically, so the last state
+ // recorded for a login is the one that counts. `null` means the
list
+ // could not be read at all — kept distinct from an empty list,
since
+ // "we did not look" must never be reported as "nobody approved".
+ async function reviewStates(number) {
+ try {
+ const reviews = await
github.paginate(github.rest.pulls.listReviews, {
+ owner, repo, pull_number: number, per_page: 100,
+ });
+ const states = new Map();
+ for (const review of reviews) {
+ const login = review.user?.login;
+ if (!login || review.state === "COMMENTED") continue;
+ states.set(login.toLowerCase(), review.state);
+ }
+ return states;
+ } catch (e) {
+ core.warning(
+ `listReviews for #${number} failed (${e.status ?? "?"}):
${e.message}`
+ );
+ return null;
+ }
+ }
+
+ // Managers gate their own branch and nothing else, so approvals
+ // compose: an approval from the v1.2 manager alone clears v1.2 and
+ // leaves v1.3 blocked.
+ function decide(targets, author, states) {
+ const authorLogin = (author || "").toLowerCase();
+ const cleared = [];
+ const blocked = [];
+ for (const target of targets) {
+ // release-branches.yml is the authorization source, so a label
+ // naming a branch it does not list has nobody who could
approve
+ // it. That differs from an entry that deliberately omits
+ // `manager`, and the difference is reachable: retiring a
branch
+ // means dropping its entry while its label lives on.
+ if (!managers.has(target)) {
+ blocked.push({ target, manager: "", state: "UNCONFIGURED" });
+ continue;
+ }
+ const manager = managers.get(target);
+ if (!manager) {
+ cleared.push({ target, note: "no release manager configured
for this branch" });
+ continue;
+ }
+ // GitHub does not let anyone approve their own PR, so a
manager
+ // who wrote the fix counts as having signed off. Logins are
+ // case-insensitive while `manager` is typed by hand, so fold
both.
+ if (manager.toLowerCase() === authorLogin) {
+ cleared.push({ target, note: `@${manager} is the author of
this PR` });
+ continue;
+ }
+ const state = states ? states.get(manager.toLowerCase()) :
"UNKNOWN";
+ if (state === "APPROVED") {
+ cleared.push({ target, note: `approved by @${manager}` });
+ } else {
+ blocked.push({ target, manager, state: state || "NONE" });
+ }
+ }
+ return { cleared, blocked };
+ }
+
+ // Say only what the state proves. A dismissal lands on DISMISSED
+ // whatever the review was — GitHub dismisses a blocking
+ // CHANGES_REQUESTED as readily as it clears a stale approval — so
+ // naming it an approval would put a guess in the manager's mouth.
+ function reason(e) {
+ if (e.state === "UNCONFIGURED") {
+ return "this branch is not listed in
.github/release-branches.yml, " +
+ "so no release manager governs it — drop the label, or add
the " +
+ "branch back to that file";
+ }
+ if (e.state === "DISMISSED") {
+ return `@${e.manager}'s review was dismissed, so no approval
stands`;
+ }
+ if (e.state === "CHANGES_REQUESTED") {
+ return `@${e.manager} requested changes and has not approved
since`;
+ }
+ if (e.state === "UNKNOWN") {
+ return `this PR's reviews could not be read, so
@${e.manager}'s ` +
+ `approval could not be confirmed`;
+ }
+ return `needs an approving review from @${e.manager}`;
+ }
+
+ let numbers;
+ try {
+ numbers = await pullRequestNumbers();
+ } catch (e) {
+ // Fail closed. Only reachable for a merge group whose commits
+ // cannot be mapped back to PRs, and a group whose approvals
+ // cannot be read is not a group this can vouch for — passing it
+ // would leave the queue ungated in exactly the window this job
+ // exists to cover. Ejecting the group is recoverable: its PRs
+ // re-queue, and each still holds its own passing check.
+ core.setFailed(
+ `Could not resolve the pull requests in this merge group ` +
+ `(${e.message}). Refusing to pass a check that could not be ` +
+ `computed.`
+ );
+ return;
+ }
+
+ const lines = [];
+ let failed = false;
+
+ for (const number of numbers) {
+ const { data: pr } = await github.rest.pulls.get({
+ owner, repo, pull_number: number,
+ });
+ // A PR already targeting a release branch is itself a backport
and
+ // is not backported onward; `no-backport-needed` vetoes the lot,
+ // the same way it does in precheck and in the post-merge push.
+ const labels = (pr.labels || []).map((l) => l.name);
+ const targets =
+ pr.base.ref !== "main" || labels.includes("no-backport-needed")
Review Comment:
This empties the target list under the veto, so the check passes — but
nothing anywhere removes the `release/*` labels. `grep -rn removeLabel` over
`.github/` returns nothing.
So the PR merges still carrying a label that promises a backport,
`direct-backport-push.yml:296-297` skips it, and the label stands as a false
record — which is the exact artifact this PR exists to eliminate.
`precheck.yml:330-331` makes it reachable by design: it documents adding
`no-backport-needed` mid-review as the supported way to cancel, and by then the
auto-labeler has already applied the labels. Removing them alongside the veto
would close it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
