[roger-datocms]

(Claude here [Opus 5.5])

Toast fixes from #152. It's independent of #171.

Problem

The plugin SDK has two toasts, and they behave differently. I measured both in a sandbox with a probe plugin:

	Colour	Closes by itself	Promise resolves
ctx.notice()	green (success)	after ~6 s	when it closes (6.0 s)
ctx.alert()	red (error)	never	only when the user clicks the X

Record Bin awaited toasts in several flows, so those flows stalled:

• Blocked deletions: onBeforeItemsDestroy awaited a green notice. That covers "could not verify a complete archive" and the over-200-records guard. The hook stayed open about 6 s after deciding, and the CMS reported Record Bin as slow ("taking longer then usual… onBeforeItemsDestroy").
• Failed restores: "Restore record" stayed disabled until the user closed the red alert. The finally that resets loading and the in-progress lock ran only after the awaited alert.
• Restore with failed archive cleanup: the plugin waited for the user to close the alert before opening the restored record.
• Settings: "Save" stayed busy after a failed save until the alert closed, and so did "Disconnect" after a failed disconnect. A webhook cleanup failure showed a green notice and held a successful save for about 6 s.

Fix

• Blocked deletions and the webhook cleanup failure now use ctx.alert(), the error toast. Elsewhere, the plugin already uses it for failures.
• None of these flows await a toast anymore. alertFn in BinOutlet is now fire-and-forget, like noticeFn.
• The delete hook still awaits all its CMA work. A genuinely slow capture still triggers the CMS warning.
• Backup notice (new): deleting now shows "Record Bin is saving a copy of this record before it is deleted. This takes a few seconds." right away. For several records it says "…saving copies of N records…". It's fired, not awaited, like the alerts above. It closes on page change, because a successful delete leaves the record, or after 15 s.

Tests

• Delete hook: three tests mock a toast that never settles: capture and the backup notice in lambdaLessCapture.test.ts, and the over-200 guard in index.test.tsx.
• Restore: a new BinOutlet.test.tsx covers the Lambda-less restore flow, with an alert that never settles. A failed restore must finish, and a restore with failed cleanup must still open the record. This exports executeRestoreWithoutLambda for the test.
• All four new tests time out without the fix.
• The ConfigScreen changes have no unit tests. Testing its handlers needs a DOM test setup, which the plugin doesn't have.

npm test (107 tests; 108 combined with #171) and tsc --noEmit pass in record-bin. Biome reports one error, a format error in recordBinModel.test.ts that is already on master.

Sandbox results

Flow	master	This branch
Blocked deletion	Green toast, hook returns at 7.7 s, slow-plugin warning at 5.6 s	Red toast, hook returns at 1.6 s, no warning
Failed restore	"Restore record" disabled until the alert is closed (still disabled at 15 s)	Re-enabled at 1.1 s, with the alert
Any deletion	No feedback during the few-second backup	"Saving a copy" notice at 0.1 s; still no slow-plugin warning

Blocked deletion, about 6 s after confirming. Before: a green toast and the slow-plugin warning.

After: only a red error toast.

Failed restore, 15 s after clicking with the alert still open. Before: "Restore record" is still disabled.

After: the button is ready for a retry.

Narrated before/after walkthrough of #171 + #172 in a legacy-flag sandbox (captions below the video):

record-bin-171-172-walkthrough.mp4

🤖 Generated with Claude Code

[roger-datocms]

Should be ready to merge along with #171