You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[bug] nightly: canvas_selection_history fails resolving edited text asset after undo #300
Job and actual failure log. The other 41 extended scripts passed. This is the remaining extended-lane failure after #249; production-browser and perf-outputs now pass.
Scenario / reproduction path
Use the pinned revision with locked dependencies and the isolated fixture/Vite servers as configured in nightly.yml:
python tests/browser/creation_server.py --port 8019
OPENDIRECTOR_API_URL=http://127.0.0.1:8019 npm run dev -- --port 5179 --strictPort
# With Playwright 1.58.2/Chromium installed as in the workflow:
CREATION_WEB_URL=http://127.0.0.1:5179 node tests/browser/canvas_selection_history.mjs
The script creates a text node, performs copy/paste/duplicate/delete/undo, enters explicit text editing, writes ABC then Backspace to AB, presses Escape, waits for the visible saved indicator, then reads the original item's binding from the server. Exact source.
Expected: the saved text binding resolves an asset and its content is AB. Actual: neither binding.materialOverride?.assetId nor binding.objectRef?.assetId is present at this read. Execution aborts before the later media kinds and save/retry assertions, so this script's remaining checks are not validated.
Environment and uncertainty
Hosted Ubuntu 22.04.5, Node 24.22.1, Python 3.12.16, Playwright 1.58.2, Chromium headless 145; isolated fixture providers, no live paid inference.
This report is based on the executed hosted log and source inspection; I did not independently rerun the browser. Root cause is unknown: the available log does not distinguish a test/save synchronization race, a changed binding contract, or product persistence failure. Do not infer user data loss from this assertion alone.
Next diagnostic / acceptance
Capture the post-Escape command response, saved-state transition and original item's server binding/content; verify whether the text is actually persisted before changing the test. If it is a race, synchronize on the actual save/version or binding, not merely a pre-existing saved label. If persistence is wrong, fix the product and add a regression. Repeated isolated runs should exercise all four node kinds and the subsequent coalescing/storage-busy assertions, retaining the AB content check.
Applicability and deduplication
Current main af428f9c8ddbbd616106f1bdd4d189370175395c retains the identical script. All 297 issues/PRs were read across 3 pages plus related discussions and current updates. #110 addressed an old model-picker incompatibility and #249 advisory hardening; neither tracks this exact undefined-asset assertion. #153 is a wider performance/history roadmap. No exact duplicate found. Refs #200, #273.
Activity
[test] canvas: wait for persisted selection edits (ThinkFlowLab#307)