test(e2e): bound OfflineWorkMetadata durable-store waits

[Fooftilly]

Summary

Focused fix for the OfflineWorkMetadata APP_READY / body wedge that hung Full E2E (~301s watchdog) on #227 and #212. Not a #212 bug; cascade should stay held until this merges.

Root cause (pinned)

Two unbounded page.evaluate shapes in this module:

1. pending / settled_conflicts / operations — one evaluate awaited listOperations() before checking a JS deadline. A never-settling store left the 300s watchdog as first kill.
2. page.evaluate("… => prksNavigate(…)") — Playwright awaits the returned navigation Promise. When leave/render never settles, evaluate hangs until the watchdog. Reproduced with body-stage diags: hang stage CARD_META_NAV on card_meta's navigate evaluate (after both pending() calls had already completed). Leaving metadata edit open made leave wait on an unsaved-draft confirm the test never clicks (intermittent 30s wait_for_selector after voiding navigate alone).

Changes

• Route pending, settled_conflicts, and operations through wait_for_async (per-await race). Predicates unchanged in strength.
• Shared navigate(): exit metadata edit, void prksNavigate, wait on DOM (same pattern as test_offline).
• Bound cached list/person IndexedDB helper reads via wait_for_async sentinels.
• Blank-page regressions: never-settle helpers fail ~0.4s; void navigate returns immediately when the navigation Promise hangs.

Non-goals

• No watchdog / Playwright timeout inflation, retries, sleeps, Full E2E suppression, or test: refresh E2E timing baseline from CI measurements #212 changes.
• Module-only audit (no repo-wide helper migration).

Validation

• Micro-repro: old pending deadline only observed after a 2s-delayed listOperations; wait_for_async fails near the caller timeout for never-settle.
• Diagnostic pin: hang at CARD_META_NAV before void-navigate fix.
• tests.e2e.test_wait_for_async PASS (helper + void-navigate regressions).
• Stress after full fix (acknowledgement reconcile/keeps/patches + online pending page):
  • serial --jobs 1 × 10 PASS
  • parallel --jobs 2 × 3 PASS
  • test_acknowledgement_patches_every_cached_representation alone × 15 PASS
• Not a full E2E gate.

  

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Repository: Fooftilly/PRKS/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: a0060895-38ce-4656-b454-5b819ae653c8

📥 Commits

Reviewing files that changed from the base of the PR and between 58a44c4 and c585bbe.

📒 Files selected for processing (2)

• tests/e2e/test_wait_for_async.py
• tests/e2e/test_work_metadata_offline.py

 ___________________________________________________________________
< This refactor is like rearranging furniture during an earthquake. >
 -------------------------------------------------------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

_{Comment @coderabbitai help to get the list of available commands.}

[qodo-code-review]

PR Summary by Qodo

Bound OfflineWorkMetadata E2E waits and navigation

🐞 Bug fix 🧪 Tests 🕐 20-40 Minutes

AI Description

• Bounds durable-store polling so stalled IndexedDB promises fail at caller timeouts.
• Makes navigation fire-and-forget after leaving metadata edit, then waits for DOM readiness.
• Adds blank-page regressions for stalled stores, empty queues, and hung navigation promises.

Diagram

graph TD
  T["Offline metadata tests"] --> H["Bound helpers"] --> W["wait_for_async"] --> S[("Durable store")]
  H --> N["Navigation helper"] --> E["Exit metadata edit"] --> P["Void prksNavigate"] --> D["DOM readiness"]

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Inline Promise.race guards

• ➕ Could bound each browser-side operation without changing helper usage.
• ➕ Keeps timeout behavior colocated with each JavaScript expression.
• ➖ Duplicates deadline and diagnostic logic across test helpers.
• ➖ Increases the chance of inconsistent timeout behavior and future unbounded awaits.

2. Await bounded navigation promises

• ➕ Preserves direct navigation rejection and completion reporting.
• ➕ Could distinguish navigation failures from later DOM-readiness failures.
• ➖ Requires reliable cancellation or timeout behavior inside application navigation.
• ➖ Does not address leave flows blocked by an unhandled unsaved-draft confirmation.

Recommendation: Keep the PR's shared wait_for_async and fire-and-forget navigation approach. It reuses established per-await timeout behavior, preserves existing predicate strength, avoids duplicated deadline code, and uses destination DOM state as the authoritative navigation completion signal.

Files changed (2) +227 / -76

Bug fix (1) +139 / -76

test_work_metadata_offline.pyBound durable-store reads and centralize safe navigation +139/-76

Bound durable-store reads and centralize safe navigation

• Routes operation polling and cached list/entity reads through 'wait_for_async', using truthy wrappers where null or empty results are valid. Introduces a navigation helper that exits metadata edit mode and voids 'prksNavigate', replaces direct navigation evaluations throughout the suite, and URL-encodes dynamic route values in Python.

tests/e2e/test_work_metadata_offline.py

Tests (1) +88 / -0

test_wait_for_async.pyAdd regressions for bounded offline metadata helpers +88/-0

Add regressions for bounded offline metadata helpers

• Adds blank-page Chromium tests proving 'pending', 'settled_conflicts', and 'operations' fail near their caller timeout when 'listOperations()' never settles. Also verifies empty operation lists remain valid and voided navigation does not await a hung Promise.

tests/e2e/test_wait_for_async.py

[qodo-code-review]

1. Route assertions can inspect stale cards 🐞 Bug ☼ Reliability

navigate() discards the prksNavigate() Promise and returns before the workspace’s asynchronous
leave-and-render chain completes, while callers such as progress_ids() wait on generic selectors
that may still match the departing route. When switching between Progress status routes, the
existing non-loading .prks-tile--main can satisfy the predicate before the requested status
renders, causing progress_ids() to read the previous group’s work identifiers and exposing other
converted callers with shared card selectors to the same stale-render race.

Agent Prompt

## Issue description
`navigate()` starts `prksNavigate()` with `void`, allowing callers to continue before the requested route has committed and rendered. Generic card and tile selectors may already exist on the departing route, so same-page changes such as switching Progress status query parameters can inspect stale content and produce timing-dependent results.

## Fix Focus Areas
- tests/e2e/test_work_metadata_offline.py[430-442]
- tests/e2e/test_work_metadata_offline.py[884-894]

## Recommended Fix
Keep evaluation non-blocking so a hung navigation Promise cannot block indefinitely, but make `navigate()` or each route-specific caller wait for a bounded, unambiguous condition proving that the requested route has been applied and its new render has completed. Use a condition such as the expected canonical `location.hash` together with completion of the destination route’s loading state; in `progress_ids()`, specifically verify that the requested encoded status is active instead of accepting any already-rendered `.prks-tile--main` before reading identifiers.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools