Skip to content
PR Triage Apply

PR Triage Apply #1215

Sign in to view logs

PR Triage Apply

PR Triage Apply #1215

Workflow file for this run

.github/workflows/pr-triage-apply.yml at d254604
# 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.
name: PR Triage Apply
# Comment-driven PR triage and lifecycle labels. Inspired by
# rust-lang/triagebot UX.
#
# Two-stage design to defeat the fork-PR token restriction:
# pr-triage-collect.yml runs on issue_comment / pull_request_review with
# a read-only GITHUB_TOKEN (forced for cross-fork events) and uploads
# the raw event payload as artifact `triage-event`. This workflow then
# fires via workflow_run on the base ref, where the GITHUB_TOKEN has
# write perms, downloads the artifact, and applies the labels.
# pull_request_target lifecycle is handled directly here without going
# through the collector — it already gets a write token by design.
#
# Comment commands (parsed line-by-line in PR comments and review bodies):
# /request-review @u [@u2 ...] — request review from one or more
# @users or @org/teams; the command may
# also repeat across lines
# /ready — flip state to S-waiting-on-review
# /author — flip state to S-waiting-on-author
#
# Auth gate:
# /request-review, /ready -> org MEMBER / repo COLLABORATOR/OWNER, or PR author
# /author -> CONTRIBUTOR / MEMBER / COLLABORATOR / OWNER
# (anyone with author_association=CONTRIBUTOR has
# at least one merged commit and is trusted to
# flip any PR back to the author queue, matching
# the implicit /author from a changes_requested
# review)
# changes_requested review -> CONTRIBUTOR / MEMBER / COLLABORATOR / OWNER
#
# Review state (pull_request_review.submitted):
# A changes_requested review by a prior contributor (CONTRIBUTOR or
# above, never a bot) acts as an implicit /author. An explicit command
# in the review body overrides it.
#
# Feedback:
# On issue_comment commands the workflow reacts on the comment: +1 when
# something was applied, "confused" when a recognized command was
# rejected for lack of permission. A failed /request-review posts a
# one-line reply naming the rejected handles. Review-triggered actions
# have no comment to react to and stay log-only.
#
# Lifecycle (pull_request_target):
# opened (non-draft) -> add S-waiting-on-review (only if no S-* present);
# unless the author has repo write access, post
# a one-time welcome comment listing the triage
# commands (marker-guarded, best-effort)
# ready_for_review -> add S-waiting-on-review (only if no S-* present)
# converted_to_draft -> remove both S-* labels
# closed (merged or not) -> remove both S-* labels
#
# SECURITY:
# - workflow_run, pull_request_target are the ONLY triggers here.
# - No actions/checkout of any ref. PR-controlled code is never executed.
# - The default GITHUB_TOKEN is never written to outputs, env files, or
# passed to user-supplied programs. The workflow only calls the GitHub
# REST API via actions/github-script.
# - pull_request_target runs the base-repo workflow with a write token so
# fork-PR lifecycle labels can be applied. workflow_run also runs against
# base ref. Both are safe because we never run fork code: no checkout,
# no exec of PR contents, no token export. The review/comment body
# reaches us via the artifact uploaded by pr-triage-collect.yml; it is
# treated as untrusted text, parsed only by the command regex. See
# https://securitylab.github.com/resources/github-actions-preventing-pwn-requests/
on:
pull_request_target:
types: [opened, ready_for_review, converted_to_draft, closed]
workflow_run:
workflows: ["PR Triage Collect"]
types: [completed]
permissions:
pull-requests: write
issues: write
contents: read
# Required for actions/download-artifact@v8 to fetch the triage-event
# artifact uploaded by the upstream PR Triage Collect run via run-id.
actions: read
concurrency:
# cancel-in-progress: false keeps the active run, but GH still replaces
# any pending run with the latest arrival on burst commands. Net effect
# under N rapid commands: the active run plus the latest arrival land,
# everything queued in between is dropped. Acceptable because /ready
# and /author are idempotent (state is reflected in the final label)
# and /request-review is the only non-idempotent command; a 3-command
# /request-review burst in <1s lands at most 2 reviewers, larger bursts
# lose proportionally more. Pending-queue retention key (queue:) is not
# yet in the GH Actions schema; revisit when available.
#
# workflow_run leg cannot key on PR number until after artifact parse,
# so it falls back to the upstream run id. Each collect run is unique,
# so this widens the group to one-run-per-upstream-collect and same-PR
# bursts on this leg do NOT serialize. Correctness does not depend on
# serialization: the setLabels helper performs an atomic replace-all
# (list current labels, drop the S-* sibling, push target, single
# setLabels call), so two interleaved label flips converge on the
# last-finishing run's target instead of fabricating a both-labels
# state. Per-PR serialization on this leg is a noise/cost
# optimization, blocked on GH adding job-level concurrency so the
# parsed PR number can re-key.
group: pr-triage-apply-${{ github.event.pull_request.number || github.event.workflow_run.id }}
cancel-in-progress: false
jobs:
apply:
if: |
github.event_name == 'pull_request_target'
|| (github.event_name == 'workflow_run'
&& github.event.workflow_run.conclusion == 'success')
runs-on: ubuntu-24.04-arm
timeout-minutes: 5
steps:
- name: Download triage event artifact
if: github.event_name == 'workflow_run'
uses: actions/download-artifact@v8
with:
name: triage-event
run-id: ${{ github.event.workflow_run.id }}
github-token: ${{ secrets.GITHUB_TOKEN }}
path: payload/
- name: Parse triage event payload
if: github.event_name == 'workflow_run'
env:
UPSTREAM_EVENT: ${{ github.event.workflow_run.event }}
run: |
set -euo pipefail
shopt -s nullglob
# The artifact contains exactly one file: the JSON event payload
# uploaded as `${{ github.event_path }}`. Its on-disk name is
# whatever the runner used (e.g. event.json); resolve it dynamically.
payload=( payload/*.json payload/event.json )
f=""
for candidate in "${payload[@]}"; do
[[ -f "$candidate" ]] && f="$candidate" && break
done
if [[ -z "$f" ]]; then
echo "triage payload not found under payload/" >&2
ls -la payload/ >&2 || true
exit 1
fi
# Body is attacker-controlled. Route it via a file rather than
# $GITHUB_ENV: heredoc terminators can be forged inside the body
# to close the block early and inject arbitrary env vars (e.g.
# NODE_OPTIONS, GITHUB_PATH) into the subsequent github-script
# step, which holds a write GITHUB_TOKEN. The other fields below
# are constrained single-line values (numeric ids, enum states,
# GH login charset) and can stay in $GITHUB_ENV.
jq -r '.comment.body // .review.body // ""' "$f" > payload/body.txt
{
echo "COMMENT_AUTHOR=$(jq -r '.comment.user.login // .review.user.login // ""' "$f")"
echo "COMMENT_ASSOC=$(jq -r '.comment.author_association // .review.author_association // ""' "$f")"
echo "COMMENT_ID=$(jq -r '.comment.id // ""' "$f")"
echo "PR_NUMBER=$(jq -r '.pull_request.number // .issue.number // ""' "$f")"
echo "REVIEW_STATE=$(jq -r '.review.state // ""' "$f")"
echo "TRIAGE_EVENT_NAME=${UPSTREAM_EVENT}"
} >> "$GITHUB_ENV"
- name: Dispatch
uses: actions/github-script@v9
env:
COMMENT_AUTHOR: ${{ env.COMMENT_AUTHOR }}
COMMENT_ASSOC: ${{ env.COMMENT_ASSOC }}
COMMENT_ID: ${{ env.COMMENT_ID }}
PR_NUMBER: ${{ env.PR_NUMBER || github.event.pull_request.number }}
REVIEW_STATE: ${{ env.REVIEW_STATE }}
TRIAGE_EVENT_NAME: ${{ env.TRIAGE_EVENT_NAME || github.event_name }}
with:
script: |
const LABEL_REVIEW = 'S-waiting-on-review';
const LABEL_AUTHOR = 'S-waiting-on-author';
const COMMITTER_ASSOCS = new Set(['MEMBER', 'COLLABORATOR', 'OWNER']);
// Wider gate for the implicit changes_requested -> author
// flip: a formal "Request changes" review is a deliberate,
// attributable action, so any prior contributor can drive
// it, not just committers. Derived from COMMITTER_ASSOCS so
// the two cannot drift. Bots are excluded separately.
const REVIEW_AUTHOR_ASSOCS = new Set(['CONTRIBUTOR', ...COMMITTER_ASSOCS]);
const prNumber = Number(process.env.PR_NUMBER);
// Number('') === 0 and an unknown jq path also yields ''.
// Without this gate the run would silently issue API calls
// against issue_number: 0, get a 404 swallowed by the
// pulls.get catch below, and complete green with no labels
// applied. Fail loudly instead so the misconfiguration shows
// up as a red Actions run.
if (!Number.isInteger(prNumber) || prNumber <= 0) {
core.setFailed(`invalid PR_NUMBER: ${process.env.PR_NUMBER}`);
return;
}
// Effective event name. On the workflow_run leg the artifact
// payload was produced by issue_comment / pull_request_review,
// and the upstream event name was preserved into the env in
// the parse step. On the pull_request_target leg we read the
// GH-provided context.eventName directly.
const eventName = process.env.TRIAGE_EVENT_NAME || context.eventName;
// Welcome comment posted once when a non-draft PR is opened.
// The leading HTML marker is invisible in rendered Markdown
// and is what postWelcomeOnce scans for to stay idempotent.
const WELCOME_MARKER = '<!-- iggy-pr-triage-welcome -->';
const WELCOME_BODY = [
WELCOME_MARKER,
'Thanks for the PR. It is labeled `S-waiting-on-review` and queued for review.',
'',
'Slash commands (own line, regular comment) move it around the queue:',
'',
'- `/ready` - back to `S-waiting-on-review` after addressing feedback',
'- `/author` - flip to `S-waiting-on-author` while you finish changes',
'- `/request-review @user-or-team` - request a reviewer',
'',
'See [CONTRIBUTING.md](https://github.com/apache/iggy/blob/master/CONTRIBUTING.md#pr-triage-commands) for details.',
].join('\n');
// Retry transient GitHub API failures so a single 502 cannot
// abort a command. Transient = 5xx, 429, or a network error
// with no HTTP status. Anything else (404, 403, 422) is not
// retried and surfaces to the caller unchanged.
const withRetry = async (fn, label) => {
const delays = [500, 1500, 4000];
for (let attempt = 0; ; attempt++) {
try {
return await fn();
} catch (e) {
const s = e.status;
const transient = s === undefined || s === 429
|| (s >= 500 && s < 600);
if (!transient || attempt >= delays.length) throw e;
core.warning(`${label}: transient failure ` +
`(${s ?? e.code ?? 'network'}), retry ` +
`${attempt + 1}/${delays.length} in ${delays[attempt]}ms`);
await new Promise(r => setTimeout(r, delays[attempt]));
}
}
};
// Atomic replace-all on the S-* slot: list current labels,
// drop both S-* siblings, optionally push the target, single
// setLabels PUT. Used by both the command leg (add = the
// target state) and the lifecycle close/converted_to_draft
// path (add = null -> just clear both S-*). One primitive
// for both legs so the lifecycle DELETE pair cannot race the
// command-leg PUT and resurrect a cleared S-* on a closed PR.
//
// Per-PR concurrency is NOT guaranteed on the workflow_run
// leg (see concurrency: comment above), so two command runs
// can land in parallel. The PUT-style setLabels makes
// interleaved flips converge on the last-finishing run's
// target instead of fabricating a both-labels state (which
// the lifecycle hasState gate would then treat as terminal
// and never clean up).
//
// The list+setLabels pair is not atomic against unrelated
// mutators, so a racing /label add by another workflow
// between our list and our setLabels would be overwritten.
// Acceptable: iggy's other workflows do not touch labels.
//
// Pagination is mandatory: an unpaginated per_page: 100 fetch
// on a PR carrying more than 100 labels would silently drop
// every label beyond page 1 on the setLabels PUT.
const replaceStateLabel = async (add) => {
const live = await withRetry(() => github.paginate(
github.rest.issues.listLabelsOnIssue,
{
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
per_page: 100,
},
), 'listLabelsOnIssue (replaceStateLabel)');
const next = live
.map(l => l.name)
.filter(n => n !== LABEL_REVIEW && n !== LABEL_AUTHOR);
if (add) next.push(add);
await withRetry(() => github.rest.issues.setLabels({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
labels: next,
}), `setLabels ${add ?? '(clear S-*)'}`);
};
// Best-effort commenter feedback. A failed reaction or reply
// must never fail the run, so each swallows its own errors.
// Reactions exist only for issue comments — review-triggered
// commands have no comment to react to, so commentId is null
// and react() is a no-op.
const commentId = Number(process.env.COMMENT_ID) || null;
const react = async (content) => {
if (!commentId) return;
try {
await withRetry(() => github.rest.reactions.createForIssueComment({
owner: context.repo.owner,
repo: context.repo.repo,
comment_id: commentId,
content,
}), `react ${content}`);
} catch (e) {
core.warning(`reaction ${content} failed: ${e.message}`);
}
};
const reply = async (text) => {
try {
await withRetry(() => github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
body: text,
}), 'createComment');
} catch (e) {
core.warning(`reply failed: ${e.message}`);
}
};
// True when `login` has push (write) access to this repo --
// the core team, by repo/team/org grant combined. On an org
// repo, author_association cannot answer this: it reports
// MEMBER for every org member, not just repo writers, so the
// collaborator-permission endpoint is the only accurate
// signal. On failure return false -- a stray welcome on a
// maintainer PR is harmless, a missing one on a contributor
// PR defeats the feature.
const hasWriteAccess = async (login) => {
try {
const { data } = await withRetry(() =>
github.rest.repos.getCollaboratorPermissionLevel({
owner: context.repo.owner,
repo: context.repo.repo,
username: login,
}), `getCollaboratorPermissionLevel ${login}`);
return data.permission === 'write' || data.permission === 'admin';
} catch (e) {
core.warning(`write-access check failed for ${login}: ${e.message}`);
return false;
}
};
// Post WELCOME_BODY once on PR open. Best-effort: a failed
// post must never fail the run or block labeling, the same
// contract as react()/reply(). Idempotent via a listComments
// scan for WELCOME_MARKER, so a manual workflow re-run does
// not double-post.
const postWelcomeOnce = async () => {
try {
// Paginated so the marker is still detected if an Actions
// re-run fires after the PR has accumulated more than 100
// comments. Without it the welcome would double-post.
const existing = await withRetry(() => github.paginate(
github.rest.issues.listComments,
{
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
per_page: 100,
},
), 'listComments');
if (existing.some(c => c.body && c.body.includes(WELCOME_MARKER))) {
core.info('welcome: already posted, skipping');
return;
}
await withRetry(() => github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
body: WELCOME_BODY,
}), 'welcome createComment');
core.info('welcome: posted');
} catch (e) {
core.warning(`welcome comment failed: ${e.message}`);
}
};
// ----- pull_request_target lifecycle -----
if (eventName === 'pull_request_target') {
const action = context.payload.action;
const pr = context.payload.pull_request;
core.info(`lifecycle event=${action} draft=${pr.draft}`);
if (action === 'closed' || action === 'converted_to_draft') {
// Atomic clear via replaceStateLabel(null): a single
// setLabels PUT that drops both S-* in one call. Using
// the same primitive as the command leg closes the
// cross-leg TOCTOU where two independent removeLabel
// calls here could be interleaved with a command-leg
// list+PUT and let S-waiting-on-review resurrect on a
// closed PR.
await replaceStateLabel(null);
core.info(`lifecycle: cleared S-* labels (${action})`);
return;
}
if (action === 'opened' || action === 'ready_for_review') {
if (pr.draft) {
core.info('lifecycle: draft PR, no label');
return;
}
// Welcome comment only on the initial non-draft open,
// and only for authors without repo write access -- the
// core team already knows the triage commands, the
// comment is contributor onboarding. ready_for_review is
// excluded: a PR opened as a draft and later marked ready
// gets no welcome, keeping scope at "non-draft open".
// Bots (dependabot, renovate, ...) are skipped: the
// welcome targets human contributor onboarding and bots
// cannot drive triage commands anyway (gated by isBot).
if (action === 'opened') {
const author = pr.user && pr.user.login;
const authorIsBot = (pr.user && pr.user.type === 'Bot')
|| (author && /\[bot\]$/.test(author));
if (!author) {
core.info('welcome: skipped, PR has no author');
} else if (authorIsBot) {
core.info(`welcome: skipped, ${author} is a bot`);
} else if (await hasWriteAccess(author)) {
core.info(`welcome: skipped, ${author} has write access`);
} else {
await postWelcomeOnce();
}
}
// Read live labels rather than the webhook-frozen
// pr.labels — a comment-driven /author or /ready may have
// landed between the lifecycle event and this run. Paginated
// so a PR with more than 100 labels still reports hasState
// correctly.
const live = await withRetry(() => github.paginate(
github.rest.issues.listLabelsOnIssue,
{
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
per_page: 100,
},
), 'listLabelsOnIssue');
const liveNames = live.map(l => l.name);
const hasState = liveNames.includes(LABEL_REVIEW)
|| liveNames.includes(LABEL_AUTHOR);
core.info(`lifecycle: has_state=${hasState}`);
if (hasState) {
core.info('lifecycle: S-* label already present, no change');
return;
}
await withRetry(() => github.rest.issues.addLabels({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: prNumber,
labels: [LABEL_REVIEW],
}), `lifecycle addLabels ${LABEL_REVIEW}`);
core.info(`lifecycle: added ${LABEL_REVIEW}`);
return;
}
core.info(`lifecycle: unhandled action ${action}`);
return;
}
// ----- issue_comment / pull_request_review commands -----
// Both event types feed the same path: COMMENT_AUTHOR/_ASSOC/
// _ID, PR_NUMBER and REVIEW_STATE resolve from the artifact
// uploaded by pr-triage-collect.yml via the parse step above.
// body is routed through a file (payload/body.txt) rather
// than $GITHUB_ENV -- see the parse step's comment for the
// injection rationale. The file is absent on the
// pull_request_target leg (no parse step), and lifecycle
// returned above, so falling back to '' is safe.
const fs = require('fs');
const body = fs.existsSync('payload/body.txt')
? fs.readFileSync('payload/body.txt', 'utf8')
: '';
// A changes_requested review is an implicit /author: the ball
// is back with the author. Gated by REVIEW_AUTHOR_ASSOCS
// (checked below) -- wider than the /author command. An
// explicit command in the review body still takes precedence.
// REVIEW_STATE is empty for issue_comment.
const reviewState = process.env.REVIEW_STATE || '';
const reviewWantsAuthor = reviewState === 'changes_requested';
// Skip everything if there is neither a command line nor an
// actionable review state. Avoids a pulls.get on every
// drive-by comment and every approve/comment-only review.
// Leading whitespace is tolerated so that " /ready" matches
// the same way the line-by-line dispatch (which trims) does.
// The trailing `(?:\s|$)` rejects suffixed prose like
// `/ready-to-merge` — `\b` fires at hyphen/slash and would
// silently flip state.
const COMMAND_RE = /^[ \t]*\/(request-review|ready|author)(?:\s|$)/m;
if (!COMMAND_RE.test(body) && !reviewWantsAuthor) {
core.info('no command in body and no actionable review state, skipping');
return;
}
const commentAuthor = process.env.COMMENT_AUTHOR;
const commentAssoc = process.env.COMMENT_ASSOC;
// Bot-suffixed accounts (e.g. dependabot[bot]) cannot drive
// commands as committer or PR author — would let bots
// self-flip review state.
const isBot = /\[bot\]$/.test(commentAuthor);
const isCommitter = COMMITTER_ASSOCS.has(commentAssoc) && !isBot;
// Permitted to drive the implicit changes_requested flip;
// see REVIEW_AUTHOR_ASSOCS.
const isReviewFlipper = REVIEW_AUTHOR_ASSOCS.has(commentAssoc) && !isBot;
// Live state read. The artifact carries the webhook-frozen
// payload at delivery time, plus we now have ~30s of
// artifact-handoff latency on top; comment-while-open +
// close-arriving-later would otherwise re-label a now-closed
// PR. Run on both committer and non-committer paths --
// correctness over the single API call saved on the
// committer fast-path.
let prData;
try {
prData = await withRetry(() => github.rest.pulls.get({
owner: context.repo.owner,
repo: context.repo.repo,
pull_number: prNumber,
}), 'pulls.get');
} catch (e) {
if (e.status === 404) {
core.info('PR not found (deleted or transferred), skipping');
return;
}
throw e;
}
if (prData.data.state === 'closed') {
core.info('PR is closed, skipping');
return;
}
// Reject command runs against a PR that turned draft after
// the comment was posted. The pull_request_target lifecycle
// clears S-* on converted_to_draft; without this gate the
// ~30-90s artifact-handoff window lets a stale command
// re-add S-waiting-on-review onto a now-draft PR.
if (prData.data.draft) {
core.info('PR is draft, skipping');
return;
}
// pr.user can be null for deleted GitHub accounts; treat the
// PR as having no recognized author rather than crashing.
const prAuthor = (prData.data.user && prData.data.user.login) || null;
const isPrAuthor = !!prAuthor && commentAuthor === prAuthor && !isBot;
core.info(`comment_author=${commentAuthor} assoc=${commentAssoc} ` +
`committer=${isCommitter} pr_author=${isPrAuthor} bot=${isBot}`);
const lines = body.split(/\r?\n/).map(l => l.trim());
let sawReassign = false;
let sawReady = false;
let sawAuthor = false;
// Outcome tracking for commenter feedback (see react() above).
// applied: at least one command took effect. denied: at least
// one recognized command was rejected for lack of permission.
let applied = false;
let denied = false;
// /request-review handles accumulate across every matching line
// (and every handle on a line); the requestReviewers call is
// deferred until after the scan so N handles cost one API call,
// not N. Sets dedupe handles repeated within the same comment.
const reviewers = new Set();
const teamReviewers = new Set();
// GH username: alnum + hyphen, no leading/trailing hyphen.
// Optional team slug: '/' then alnum + underscore + hyphen.
// Fully anchored so a malformed handle (empty slug @foo/, extra
// segment @foo/bar/baz) is rejected per-token, not silently
// truncated.
const HANDLE_RE = /^@([A-Za-z0-9](?:[A-Za-z0-9-]*[A-Za-z0-9])?(?:\/[A-Za-z0-9_-]+)?)$/;
for (const line of lines) {
// Unlike /ready and /author, /request-review may legitimately
// repeat across lines, so no first-match-wins guard and no
// early loop break.
const rr = line.match(/^\/request-review(?:\s+(.*))?$/);
if (rr) {
sawReassign = true;
if (!(isCommitter || isPrAuthor)) {
core.info(`/request-review: ignored, ${commentAuthor} lacks permission`);
denied = true;
continue;
}
const rest = (rr[1] || '').trim();
if (!rest) {
core.info('/request-review: no reviewer specified');
continue;
}
for (const token of rest.split(/\s+/)) {
const hm = token.match(HANDLE_RE);
if (!hm) {
core.warning(`/request-review: ignoring malformed handle "${token}"`);
continue;
}
const handle = hm[1];
// Team slugs go through team_reviewers as bare slug, no org.
if (handle.includes('/')) {
teamReviewers.add(handle.split('/')[1]);
} else {
reviewers.add(handle);
}
}
continue;
}
if (!sawReady && /^\/ready(?:\s|$)/.test(line)) {
sawReady = true;
if (!(isCommitter || isPrAuthor)) {
core.info(`/ready: ignored, ${commentAuthor} lacks permission`);
denied = true;
} else {
await replaceStateLabel(LABEL_REVIEW);
core.info(`/ready: ${LABEL_REVIEW} <- ${LABEL_AUTHOR}`);
applied = true;
}
continue;
}
if (!sawAuthor && /^\/author(?:\s|$)/.test(line)) {
sawAuthor = true;
// Wider than /ready: anyone GitHub recognizes as a returning
// contributor (>=1 merged commit) can flip any PR to the
// author queue. Same gate as the implicit /author from a
// changes_requested review.
if (!isReviewFlipper) {
core.info(`/author: ignored, ${commentAuthor} not a committer or returning contributor`);
denied = true;
} else {
await replaceStateLabel(LABEL_AUTHOR);
core.info(`/author: ${LABEL_AUTHOR} <- ${LABEL_REVIEW}`);
applied = true;
}
continue;
}
}
if (reviewers.size > 0 || teamReviewers.size > 0) {
const params = {
owner: context.repo.owner,
repo: context.repo.repo,
pull_number: prNumber,
};
if (reviewers.size > 0) params.reviewers = [...reviewers];
if (teamReviewers.size > 0) params.team_reviewers = [...teamReviewers];
const requested = [...reviewers, ...teamReviewers].join(', ');
try {
await withRetry(
() => github.rest.pulls.requestReviewers(params),
'requestReviewers',
);
core.info(`/request-review: requested ${requested}`);
applied = true;
} catch (e) {
// Transient 5xx is retried by withRetry before reaching
// here. What lands in this catch is a non-transient
// failure — typically the 422 GitHub returns for the
// whole batch on a single bad handle (non-collaborator,
// unknown team). Surface it to the commenter; the run log
// also names the rejected set.
core.warning(`/request-review: failed to request [${requested}]: ${e.message}`);
const why = e.status === 422
? 'a handle is not a repo collaborator, or the team is unknown'
: `GitHub API error (${e.status ?? 'network'})`;
await reply(`\`/request-review\`: could not request ${requested} - ${why}.`);
}
}
// Implicit /author from a changes_requested review. Skipped
// when the review body already carried an explicit /ready or
// /author -- the command wins. Wider gate than the /author
// command: any prior contributor, not just committers (see
// REVIEW_AUTHOR_ASSOCS).
if (reviewWantsAuthor && !sawReady && !sawAuthor) {
if (!isReviewFlipper) {
core.info(`review changes_requested: ignored, ` +
`${commentAuthor} (${commentAssoc}) not permitted`);
} else {
await replaceStateLabel(LABEL_AUTHOR);
core.info(`review changes_requested: ${LABEL_AUTHOR} <- ${LABEL_REVIEW}`);
applied = true;
}
}
if (!sawReassign && !sawReady && !sawAuthor && !reviewWantsAuthor) {
core.info('no command matched');
}
// Commenter feedback. denied wins over applied: if any part of
// the comment was rejected the commenter should see it, even
// when another command in the same comment worked. react() is
// a no-op for review-triggered runs (no comment to react to).
if (denied) {
await react('confused');
} else if (applied) {
await react('+1');
}