Yicong-Huang commented on code in PR #8096: URL: https://github.com/apache/texera/pull/8096#discussion_r3921600865
########## .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: Verified at `77d573c1a` — the contradiction is reported and it blocks: `intoMain && vetoed` maps every `release/*` label to a `VETOED` entry (`:264-269`), those land in `blocked` (`:281`), and `blocked.length > 0` sets `failed` (`:285`). A PR can no longer merge carrying both. On your question — keep it here, don't move it into `backport-auto-label.yml`. Auto-removing the label would make the veto silently rewrite the record this check exists to make truthful, and it would do so from the one workflow that runs `pull_request_target` with a write token, where an unattended label mutation is the most expensive kind of mistake. Reporting leaves the choice with the person who has the context for it, and either resolution clears the check. The residual is documentation, not code: `precheck.yml:327-331` still calls adding the veto mid-review the way to cancel a backport, which is now only half the action — raising that separately. -- 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]
