Reject release jobs failed in earlier run attempts

docs/releasing.md

@@ -51,7 +51,7 @@ count. So a build reaches people only when you publish it, after testing it.
    The script checks that the tag is on GitHub, that exact-commit main CI
    (including e2e), native amd64/ARM64 Decky runtime checks, and the tagged
    release workflow are green with no failed,
-   cancelled, skipped or running jobs, and that all ten installers
+   cancelled, skipped or running jobs in any run attempt, and that all ten installers
    are attached, each with the SHA-256 digest GitHub records. It also checks
    that the draft does not begin with a `WITHDRAWN` notice, so the preserved
    v0.4.48 draft cannot accidentally be offered again. It checks

scripts/verify-release-gates.py

@@ -77,19 +77,25 @@ def successful_jobs(run, required):
     attempt = run.get("run_attempt")
     if not isinstance(attempt, int) or attempt < 1:
         raise GateError(f"run {run['id']} has no valid attempt number")
-    data = api(f"actions/runs/{run['id']}/attempts/{attempt}/jobs?per_page=100")
-    jobs = data.get("jobs") or []
-    if data.get("total_count") != len(jobs):
-        raise GateError(f"run {run['id']} has more jobs than the gate inspected")
-    names = [job.get("name") for job in jobs]
-    if len(names) != len(set(names)) or not required.issubset(names):
+    latest_jobs = None
+    for number in range(1, attempt + 1):
+        data = api(f"actions/runs/{run['id']}/attempts/{number}/jobs?per_page=100")
+        jobs = data.get("jobs") or []
+        if data.get("total_count") != len(jobs):
+            raise GateError(f"run {run['id']} attempt {number} has more jobs than the gate inspected")
+        names = [job.get("name") for job in jobs]
+        if not jobs or len(names) != len(set(names)):
+            raise GateError(f"run {run['id']} attempt {number} has missing or duplicate jobs")
+        bad = [f"{job.get('name')}: {job.get('status')}/{job.get('conclusion')}"
+               for job in jobs if job.get("status") != "completed" or job.get("conclusion") != "success"]
+        if bad:
+            raise GateError(f"run {run['id']} attempt {number} has unfinished, failed, cancelled, or skipped jobs: {bad}")
+        latest_jobs = jobs
+    names = [job.get("name") for job in latest_jobs]
+    if not required.issubset(names):
         missing = sorted(required.difference(names))
-        raise GateError(f"run {run['id']} has missing or duplicate jobs: {missing}")
-    bad = [f"{job.get('name')}: {job.get('status')}/{job.get('conclusion')}"
-           for job in jobs if job.get("status") != "completed" or job.get("conclusion") != "success"]
-    if bad:
-        raise GateError(f"run {run['id']} has unfinished, failed, cancelled, or skipped jobs: {bad}")
-    print(f"gate passed: {run.get('html_url')} ({len(jobs)} jobs)")
+        raise GateError(f"run {run['id']} has missing jobs in its latest attempt: {missing}")
+    print(f"gate passed: {run.get('html_url')} ({len(latest_jobs)} jobs, {attempt} attempts)")
 
 
 def main(tag):

tests/fakegh/gh

@@ -91,17 +91,22 @@ elif method == "GET" and endpoint.startswith(repo + "actions/workflows/"):
     is_main = workflow != "release.yml"
     run_id = {"checks.yml": 501, "release.yml": 502, "decky-runtime.yml": 503}[workflow]
     run = {"id": run_id, "head_sha": gate.get("main_sha" if is_main else "release_sha", commit),
-           "head_branch": branch, "event": "push", "run_attempt": 1,
+           "head_branch": branch, "event": "push", "run_attempt": gate.get("release_attempts", 1) if run_id == 502 else 1,
            "status": gate.get("decky_status" if run_id == 503 else ("main_status" if is_main else "release_status"), "completed"),
            "conclusion": gate.get("decky_conclusion" if run_id == 503 else ("main_conclusion" if is_main else "release_conclusion"), "success"),
            "html_url": "https://github.com/saphid/frame-control/actions/runs/%s" % run_id}
     print(json.dumps({"workflow_runs": [run]}))
 elif method == "GET" and endpoint.startswith(repo + "actions/runs/"):
-    run_id = endpoint[len(repo + "actions/runs/"):].split("/")[0]
+    parts = endpoint[len(repo + "actions/runs/"):].split("/")
+    run_id = parts[0]
+    attempt = int(parts[2])
     names = {"501": main_jobs, "502": release_jobs, "503": decky_jobs}[run_id]
     names = [name for name in names if name != gate.get("omit_job")]
     jobs = [{"name": name, "status": "completed",
-             "conclusion": gate.get("bad_job_conclusion", "success") if name == gate.get("bad_job") else "success"}
+             "conclusion": (gate.get("prior_bad_job_conclusion", "failure")
+                            if run_id == "502" and attempt == 1 and name == gate.get("prior_bad_job")
+                            else gate.get("bad_job_conclusion", "success") if name == gate.get("bad_job")
+                            else "success")}
             for name in names]
     print(json.dumps({"total_count": len(jobs), "jobs": jobs}))
 elif method == "GET" and endpoint.startswith(repo + "releases?"):

tests/test_publish_release.py

@@ -134,6 +134,17 @@ class PublishRelease(unittest.TestCase):
             self.assertIsNone(upload)
             self.assertFalse(any("-X" in call for call in log))
 
+    def test_refuses_a_prior_cancelled_installer_after_successful_rerun(self):
+        gates = {"release_attempts": 2,
+                 "prior_bad_job": "build (macos-latest, dist, app/dist/*.dmg app/dist/*.zip)",
+                 "prior_bad_job_conclusion": "cancelled"}
+        p, log, upload = self.run_script([draft()], "v9.8.7", gates=gates)
+        self.assertNotEqual(p.returncode, 0)
+        self.assertIn("attempt 1", p.stderr)
+        self.assertIn("cancelled", p.stderr)
+        self.assertIsNone(upload)
+        self.assertFalse(any("-X" in call for call in log))
+
     def test_refuses_missing_or_nonexact_gate_results(self):
         for gates in ({"omit_job": "e2e"}, {"main_sha": "c" * 40},
                       {"decky_conclusion": "failure"}, {"release_conclusion": "failure"},