CI: real-stack Playwright tail is red on main and blocks exact-head lifecycle merge #44

New issue

Closed

opened 2026-08-25 08:17:25 +00:00 by lost-rob0t · 1 comment

lost-rob0t commented

2026-08-25 08:17:25 +00:00

(Migrated from github.com)

Problem

quasar's current main baseline and the Auto-Dig lifecycle PR both fail at the same repository-native CI step: Real-stack desktop, mobile, mutation, and reload paths.

This blocks exact-head-green merge of #43 even though the #41 owning-boundary gates themselves pass.

Evidence

main SHA 560208d0906cf564dd59050bca9a0f9341a3c55d, CI run #718 / run 32216643080: steps 1–13 pass, step 14 (Real-stack desktop, mobile, mutation, and reload paths) fails, packaged production smoke is skipped.
#43 exact head 10f3f223ec066d6250cb0ce238802456d0ce9bf8, CI run #740 / run 32821728914: the same shape — static/frontend/build, complete Common Lisp transaction/persistence/control-plane suite, real quasar-web load, development-stack smoke, and Playwright installation all pass; step 14 fails and the packaged tail is skipped.
#43 changes only:
- control-plane/src/autodig-control.lisp
- control-plane/tests/autodig-control-tests.lisp
- systems/quasar-control.asd
- systems/quasar-tests.asd
  No frontend/Playwright files are changed.

Required diagnosis

Capture the exact failing Playwright spec/assertion from a reproducible run; do not infer from step-level status alone.
Reproduce against current main first. If the failure is baseline, fix it on the owning UI/E2E boundary with a failing regression test before changing production/UI behavior.
If the branch uniquely changes the result despite the same step shape, prove the causal path before modifying #43.
Do not skip, quarantine, weaken, or mark the Playwright gate allowed-to-fail merely to merge #43.

Acceptance

Current main passes the complete repository-native CI including real-stack Playwright and packaged production smoke.
#43 is rebased/refreshed only if needed, and its exact candidate head passes the same complete CI before merge.

Related: #41, PR #43.

## Problem `quasar`'s current `main` baseline and the Auto-Dig lifecycle PR both fail at the same repository-native CI step: **Real-stack desktop, mobile, mutation, and reload paths**. This blocks exact-head-green merge of #43 even though the #41 owning-boundary gates themselves pass. ## Evidence - `main` SHA `560208d0906cf564dd59050bca9a0f9341a3c55d`, CI run #718 / run `32216643080`: steps 1–13 pass, step 14 (`Real-stack desktop, mobile, mutation, and reload paths`) fails, packaged production smoke is skipped. - #43 exact head `10f3f223ec066d6250cb0ce238802456d0ce9bf8`, CI run #740 / run `32821728914`: the same shape — static/frontend/build, complete Common Lisp transaction/persistence/control-plane suite, real `quasar-web` load, development-stack smoke, and Playwright installation all pass; step 14 fails and the packaged tail is skipped. - #43 changes only: - `control-plane/src/autodig-control.lisp` - `control-plane/tests/autodig-control-tests.lisp` - `systems/quasar-control.asd` - `systems/quasar-tests.asd` No frontend/Playwright files are changed. ## Required diagnosis - Capture the exact failing Playwright spec/assertion from a reproducible run; do not infer from step-level status alone. - Reproduce against current `main` first. If the failure is baseline, fix it on the owning UI/E2E boundary with a failing regression test before changing production/UI behavior. - If the branch uniquely changes the result despite the same step shape, prove the causal path before modifying #43. - Do not skip, quarantine, weaken, or mark the Playwright gate allowed-to-fail merely to merge #43. ## Acceptance - Current `main` passes the complete repository-native CI including real-stack Playwright and packaged production smoke. - #43 is rebased/refreshed only if needed, and its exact candidate head passes the same complete CI before merge. Related: #41, PR #43.

lost-rob0t commented

2026-08-25 10:14:10 +00:00

(Migrated from github.com)

RAGE/TDD diagnosis from PR #45 diagnostics is now concrete. CI run 32830522781 retained the Playwright artifact (playwright-failure-32830522781). The baseline failures are a mixed stale-E2E set after unified shell #40, not an Auto-Dig #43 regression:

application.spec.ts still expects visible legacy Graph layout; current accessible control is Maltego graph layout.
mobile-fit-agent-disclosure.spec.ts uses /Layout/, which now matches the exact radial Layout category plus five Layout: ... commands and trips Playwright strict mode.
graph-controls.spec.ts waits for old Focus selection; current desktop control is exact Focus.
graph-fullscreen.spec.ts and graph-viewport.spec.ts resolve the correct full-viewport button, but the unrelated Document update conflict notice physically intercepts pointer events; the viewport tests need to clear that notice before exercising the control.
large-import.spec.ts expects 2001 documents; current shell intentionally renders locale-formatted 2,001 documents (and graph count follows the same formatting).

Opened draft PR #46 at candidate 04b5ae2b9b50fdb4e6330dceb0ca0b936fe3947f with only E2E/diagnostic CI repairs. No production graph behavior is changed, and no test is skipped/softened. Exact-head CI #760 is running; merge remains blocked until the complete suite including packaged production smoke is green.

RAGE/TDD diagnosis from PR #45 diagnostics is now concrete. CI run `32830522781` retained the Playwright artifact (`playwright-failure-32830522781`). The baseline failures are a mixed stale-E2E set after unified shell #40, not an Auto-Dig #43 regression: - `application.spec.ts` still expects visible legacy `Graph layout`; current accessible control is `Maltego graph layout`. - `mobile-fit-agent-disclosure.spec.ts` uses `/Layout/`, which now matches the exact radial `Layout` category plus five `Layout: ...` commands and trips Playwright strict mode. - `graph-controls.spec.ts` waits for old `Focus selection`; current desktop control is exact `Focus`. - `graph-fullscreen.spec.ts` and `graph-viewport.spec.ts` resolve the correct full-viewport button, but the unrelated `Document update conflict` notice physically intercepts pointer events; the viewport tests need to clear that notice before exercising the control. - `large-import.spec.ts` expects `2001 documents`; current shell intentionally renders locale-formatted `2,001 documents` (and graph count follows the same formatting). Opened draft PR #46 at candidate `04b5ae2b9b50fdb4e6330dceb0ca0b936fe3947f` with only E2E/diagnostic CI repairs. No production graph behavior is changed, and no test is skipped/softened. Exact-head CI #760 is running; merge remains blocked until the complete suite including packaged production smoke is green.