recipe-maintainers/cc-ci
0
0
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity

runner: never embed a fossil screenshot from a collided run dir #45

Merged
autonomic-bot merged 2 commits from fix/stale-run-screenshot into main 2026-10-05 17:46:25 +00:00
Conversation 0 Commits 2 Files Changed 5
+169 -5
autonomic-bot commented 2026-10-05 17:41:55 +00:00
Owner
Copy Link
Copy Source

What: !testme on gitea PR #10 (drone build 35, 2026-10-05) rendered a summary card whose embedded app screenshot was custom-html-tiny's — while the run itself was genuinely gitea, level 5, all tests passing. Cosmetics-only bug; verdicts were never affected.

Root cause chain (evidence in /var/lib/cc-ci-runs/35/):

  1. /var/lib/cc-ci-runs/ still holds old-era run dirs numbered up to 1377; drone's build counter restarted when its DB was re-created (2026-09-27), so new build numbers collide with old dirs (35 was one, first hit because that old dir happened to contain a screenshot).
  2. The run reused the collided dir without cleaning it — fresh results.json/summary.* landed next to the May-31 fossil screenshot.png (its mtime was still May 31; size 12950 B = exactly custom-html-tiny's real screenshot per test_blank_threshold_brackets_observed_sizes).
  3. screenshot.capture()'s SCREENSHOT-hook branch skipped the snap when out_path already existed ("the hook may have saved it" — no hook in the repo saves the PNG; every SCREENSHOT hook only navigates). So capture() returned the fossil as "captured" and the card embedded it.

Fix (cosmetics-only, R7 — verdict/level logic untouched):

  • results.fresh_artifact_dir(path): at run start remove ONLY the files/dirs this run owns (results.json, summary.html/png, badge.svg, lint.txt, screenshot.png, junit/); anything else in the dir is left untouched; never raises.
  • run_recipe_ci.py calls it right after computing run_artifact_dir.
  • screenshot.capture(): the hook branch now ALWAYS snaps (same settle + blank-retry path as the default branch) — a pre-existing file is never trusted.

Evidence:

  • Unit: 4 new tests (owned-only removal / dir creation / odd entries; fake-playwright capture regression proving a fossil out_path is overwritten and the hook still fires). Full unit suite: 319 passed, only the 4 pre-existing main failures (identical set on f675941).
  • Live E2E (cold, host, custom-html-tiny full stages with a simulated collision: pre-seeded fossil screenshot.png + ghost results.json in the run dir): fossil wiped, fresh PNG captured (screenshot: captured), results.json recipe=custom-html-tiny level=5, summary.html mentions only custom-html-tiny, teardown clean (no leftover stacks/volumes).
  • scripts/lint.sh: PASS (branch is rebased on post-#44 main, which already fixed the two pre-existing lint failures).

Dependency: stacked on #44 (merged) — the branch starts at the squashed #44 commit.

NOT merged — for operator review.

**What:** `!testme` on gitea PR #10 (drone build 35, 2026-10-05) rendered a summary card whose embedded app screenshot was **custom-html-tiny's** — while the run itself was genuinely gitea, level 5, all tests passing. Cosmetics-only bug; verdicts were never affected. **Root cause chain (evidence in `/var/lib/cc-ci-runs/35/`):** 1. `/var/lib/cc-ci-runs/` still holds **old-era run dirs numbered up to 1377**; drone's build counter restarted when its DB was re-created (2026-09-27), so new build numbers collide with old dirs (35 was one, first hit because that old dir happened to contain a screenshot). 2. The run reused the collided dir **without cleaning it** — fresh `results.json`/`summary.*` landed next to the May-31 fossil `screenshot.png` (its mtime was still May 31; size 12950 B = exactly custom-html-tiny's real screenshot per `test_blank_threshold_brackets_observed_sizes`). 3. `screenshot.capture()`'s SCREENSHOT-hook branch **skipped the snap when `out_path` already existed** ("the hook may have saved it" — no hook in the repo saves the PNG; every `SCREENSHOT` hook only navigates). So `capture()` returned the fossil as "captured" and the card embedded it. **Fix (cosmetics-only, R7 — verdict/level logic untouched):** - `results.fresh_artifact_dir(path)`: at run start remove ONLY the files/dirs this run owns (`results.json`, `summary.html/png`, `badge.svg`, `lint.txt`, `screenshot.png`, `junit/`); anything else in the dir is left untouched; never raises. - `run_recipe_ci.py` calls it right after computing `run_artifact_dir`. - `screenshot.capture()`: the hook branch now ALWAYS snaps (same settle + blank-retry path as the default branch) — a pre-existing file is never trusted. **Evidence:** - Unit: 4 new tests (owned-only removal / dir creation / odd entries; fake-playwright capture regression proving a fossil `out_path` is overwritten and the hook still fires). Full unit suite: **319 passed**, only the 4 pre-existing main failures (identical set on f675941). - Live E2E (cold, host, custom-html-tiny full stages with a **simulated collision**: pre-seeded fossil `screenshot.png` + ghost `results.json` in the run dir): fossil wiped, fresh PNG captured (`screenshot: captured`), `results.json` recipe=custom-html-tiny level=5, `summary.html` mentions only custom-html-tiny, teardown clean (no leftover stacks/volumes). - `scripts/lint.sh`: PASS (branch is rebased on post-#44 main, which already fixed the two pre-existing lint failures). **Dependency:** stacked on #44 (merged) — the branch starts at the squashed #44 commit. NOT merged — for operator review.
autonomic-bot added 2 commits 2026-10-05 17:41:55 +00:00
runner: never embed a fossil screenshot from a collided run dir
continuous-integration/drone/push Build is passing
Details
934ec8b8f0 
!testme gitea PR #10 (drone build 35, 2026-10-05) rendered a summary card whose
app screenshot was custom-html-tiny's — a fossil from a May-31 run dir with the
same number. Root cause chain:
- /var/lib/cc-ci-runs/ keeps old-era run dirs numbered up to 1377; drone's
  build counter restarted when its DB was re-created (2026-09-27), so new
  builds collide with old dirs (35 was one).
- the run reused the collided dir without cleaning it: fresh results.json/
  summary.* landed NEXT TO the old screenshot.png.
- screenshot.capture()'s SCREENSHOT-hook branch skipped the actual snap when
  out_path already existed ('the hook may have saved it' — no hook does), so
  capture() 'succeeded' without writing and the card embedded the fossil.

Fix, cosmetics-only (R7 — verdict logic untouched):
- results.fresh_artifact_dir(): at run start, remove ONLY the files/dirs this
  run owns (results.json, summary.*, badge.svg, lint.txt, screenshot.png,
  junit/) from its artifact dir; anything else is left alone; never raises.
- run_recipe_ci calls it right after computing run_artifact_dir.
- screenshot.capture(): the hook branch ALWAYS snaps (same settle/blank-retry
  path as the default branch) — a pre-existing file is never trusted.

Unit tests: fresh_artifact_dir (owned-only removal, dir creation, odd entries)
+ capture() fake-playwright regression proving a fossil out_path is overwritten.
tests: unit coverage for the fossil-screenshot fix
continuous-integration/drone/push Build is passing
Details
8e7008e422 
(follow-up to previous commit — test appends initially landed in the wrong
clone; content identical)
autonomic-bot requested review from trav 2026-10-05 17:41:55 +00:00
autonomic-bot requested review from notplants 2026-10-05 17:41:55 +00:00
autonomic-bot merged commit 35ba99768c into main 2026-10-05 17:46:25 +00:00
Sign in to join this conversation.
Reviewers
No Reviewers
trav
notplants
Labels
No items
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
autonomic autonomic-bot kawaiipunk (KawaiiPunk) notplants trav (Trav Fryer)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: recipe-maintainers/cc-ci#45