Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 21 additions & 0 deletions .github/scripts/fork-head-gate-workflow.test.py
Original file line number Diff line number Diff line change
Expand Up @@ -424,6 +424,27 @@ def caller_group(event, run_id, sha="", workflow_ref="o/p/.github/workflows/gate
"caller's run (the gate job's own group is per head: a newer job there reads the newest state)",
lost == [("review", 2)], str(lost))

# #379: the review trigger's live check is a written setup step, and the release notes tell a pin move what to copy.
# The fields the lookup relies on are tested above only against fakes, so whoever sets the trigger up checks them once.
readme = (root / "fork-ci/actions/head-check/README.md").read_text()
setup = readme[readme.find("\nSetup ("):]
setup = setup[:setup.find("\n\n", setup.find("\n8. "))] if "\n8. " in setup else ""
check("README setup: step 8 is the live check (post a review, the gate runs; the warning that says the fields differ)",
"post a review on one" in setup and "review run: #N has the head" in setup
and "does not name a head" in setup and "head=owner:branch" in setup and "fallback" in setup, setup[-300:])
gate_src = GATE_PY.read_text()
check("README setup: the warning it names is head_gate's own text",
"the review trigger's run does not name a head" in setup.replace("\n ", " ")
and "the review trigger's run does not name a head" in gate_src)
notes = (root / "fork-ci/RELEASE-NOTES.md").read_text()
entry = next((e for e in notes.split("\n### ")[1:] if "review-trigger.example.yml" in e and "#375" in e.split("\n")[0]),
"")
upgrade = entry[entry.find("**Upgrade:**"):] if "**Upgrade:**" in entry else ""
check("release notes: the review trigger has an entry whose Upgrade copies the trigger and the caller's workflow_run, "
"if and concurrency, pin first, then the live check (setup step 8)",
all(w in upgrade for w in ("review-trigger.example.yml", "gate-caller.example.yml", "`workflow_run`", "`if`",
"`concurrency`", "pinned to an older release", "setup step 8")), upgrade[:300])

if failures:
print(f"\n{len(failures)} failed")
sys.exit(1)
Expand Down
19 changes: 19 additions & 0 deletions fork-ci/RELEASE-NOTES.md
Original file line number Diff line number Diff line change
Expand Up @@ -242,6 +242,25 @@ then behind, and the update that brings it up to date is a new head, judged on t
refuses the fix too ("not JSON"), and an owner has to push the fix by hand, as for any base config that does not
parse.

### head-check gate: reviews and inline review comments run the gate again (#375, #392, #379)

A new example workflow, `review-trigger.example.yml` ("Head gate review trigger"), runs on reviews and inline review
comments and does nothing but finish. The gate caller runs on `workflow_run` of it and judges the head of the open pull
request whose head branch, head repository and head sha are the run's (the number comes from the API, never from the
run). Two open pull requests on one head (one branch into two bases) are judged once, in the job for the lowest number.
Before this, a review left the check showing the older answer until the next push or issue comment. First in `ci/v15`.
After `ci/v19` (#379), a review run whose head fields are missing or odd judges nothing with a warning (before: a plain
log line).

**Upgrade:** nothing breaks if you only move the pin: a caller without the new trigger never runs on `workflow_run`.
To use it, in one pull request on the default branch, after the pin is at a release that has this entry: copy
`review-trigger.example.yml` to `.github/workflows/` unchanged, and add the caller's `workflow_run` trigger, its `if`
(with the event clause) and its `concurrency` from `gate-caller.example.yml`. The pin moves first or in that pull
request: a caller with `workflow_run` pinned to an older release is refused on that event (its guard job fails, nothing
is posted). Then do the
live check once ([head-check README](actions/head-check/README.md), setup step 8): post a review on one open pull
request and check that the gate ran for it. The GitHub fields the lookup uses are tested only against fakes.

### pr-queue and head-check: stricter rule order in `review` (#378)

A rule of the same weight above another rule that lacks one of its lenses is now a rule-order problem, and only the
Expand Down
16 changes: 15 additions & 1 deletion fork-ci/actions/head-check/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -320,6 +320,17 @@ Setup (the repository owner's steps; nothing here creates them):
([pr-queue](../pr-queue/README.md)), so the queue counts only that App's
check runs. Release the code first, then bump the pins, then add the key:
an older queue refuses a key it does not know.
8. The live check, once, after the review trigger is in: post a review on one
open pull request and check that the gate ran for it. The caller's run on
`workflow_run` logs `review run: #N has the head …`, and the check run on
that head is new. The GitHub fields the lookup uses (the review run's head
branch, head repository and head sha, and the list filter
`head=owner:branch`) are tested only against fakes. If the run warns
"the review trigger's run does not name a head", or logs "no open pull
request has the head" for a review on an open pull request whose head did
not move, they differ: reviews then never run the gate again. Report it
on the fork and use the fallback (an issue comment or a dispatch, see
"Run again after a review") until it is fixed.

After a change to the caller, a new pin, or a change to the queue config or
the lens rules on the default branch, dispatch the backfill again: it judges
Expand Down Expand Up @@ -431,7 +442,10 @@ head "not passed" unless it judged the head and the head passed.
pull request read and git's `refs/pull/N/head` both fail every try.
- A review trigger's run whose pull request cannot be found: the list of
open pull requests with its head fails every try (the guard job fails), or
no open pull request has its head (nothing to judge, the log says why).
no open pull request has its head (nothing to judge, the log says why). A
run that names no head (its head fields missing or odd) judges nothing
with a warning: GitHub's fields are not what the lookup expects (setup
step 8).
More than one is no such case: the head is judged once, in the job for
the lowest number.
- The backfill cannot list the open pull requests after its tries, or finds
Expand Down
7 changes: 5 additions & 2 deletions fork-ci/actions/head-check/head_gate.py
Original file line number Diff line number Diff line change
Expand Up @@ -326,8 +326,11 @@ def review_run_pr(repo, run, gh, out):
return [], None
if not (pr_queue.SHA.fullmatch(sha) and pr_queue.REPO_RE.fullmatch(head_repo) and 0 < len(branch) <= BRANCH_MAX
and branch.isprintable()):
out("nothing to judge: the run does not name a head (head sha, head repository, head branch); dispatch the "
'gate with the pull request number')
# A warning, not a log line: a review run always has these fields (GitHub's, of a pull request event), so a
# run without them means GitHub's fields are not what this lookup expects, and no review would ever re-run the
# gate while every run stays green. The annotation makes that seen (the live check after setup looks for it).
out("::warning::head gate: nothing to judge: the review trigger's run does not name a head (head sha, head "
'repository, head branch); dispatch the gate with the pull request number')
return [], None
owner = urllib.parse.quote(head_repo.split('/')[0], safe='')
query = (f'repos/{repo}/pulls?state=open&per_page=100&head={owner}:'
Expand Down
7 changes: 7 additions & 0 deletions fork-ci/actions/head-check/head_gate.test.py
Original file line number Diff line number Diff line change
Expand Up @@ -556,6 +556,13 @@ def test_a_run_of_another_event_or_with_odd_fields_judges_nothing_and_reads_noth
code, out, gh = self.review(run=run)
self.assertEqual((code, self.found(), gh.calls), (0, [], []), out)
self.assertIn('nothing to judge', out)
# #379: a review run always has its head fields, so odd ones mean GitHub's fields are not what the
# lookup expects (no review would ever re-run the gate): a warning the live check can see, one line.
# Another event is normal (an older caller without the event clause): a plain log line.
odd = 'event' not in run
self.assertEqual(out.startswith('::warning::head gate: nothing to judge: '), odd, out)
self.assertEqual(out.count('::'), 2 if odd else 0, out)
self.assertNotIn('::error::', out)

def test_the_listing_reads_only_the_fields_it_uses(self): # #379: no draft (preflight fails a draft's head)
self.assertEqual(G.HEAD_JQ, '.[] | [.number, .head.sha, .head.repo.full_name] | @json')
Expand Down
Loading