Repository navigation
Replies: 2 comments
|
To make the remaining prototype behavior in this proposal explicit, here are synthetic acceptance scenarios beyond a successful snapshot fetch:
These describe tested fork protections and a proposed contract, not a claim that every interleaving was reproduced on clean upstream. Actual installed iOS PWA/OS badge behavior still needs device validation; WebKit viewport tests do not certify it. Category totals/unread presentation remains #597; readiness/retry remains #612. Public prototype: commit, INBOX action transitions, snapshot/write barriers, SW ordering, and real transport recovery. Fork CI is green. No private mailbox data is involved. Would this acceptance matrix be a useful basis for small upstream changes, or would you prefer some of it tracked separately? |
|
Thanks for the matrix. I checked it against current main. Most of the contract is already there: Scenario (1) found a real gap, fixed in d3102ae: the account badge counts INBOX, but the optimistic adjustments ignored the message's folder, so reading, deleting, archiving or moving unread mail in another folder dropped the Inbox badge for up to 15 s, and marking an Inbox conversation read also charged its unread copies elsewhere. Every adjustment now charges INBOX messages only, a move into INBOX raises the badge, and undo or failure gives back exactly what was taken. I'd rather not adopt the rest as a formal contract: the design is already in place, and further hardening is easier to justify one reproduction at a time. If you hit a count that stays wrong on main, please open an issue with the sequence and I'll treat it as a bug. |
Uh oh!
There was an error while loading. Please reload this page.
Would an authoritative INBOX count snapshot be a suitable upstream contract for convergence across tabs and push/reconnect events?
Proposal: retain immediate optimistic feedback, but settle it against a fresh server snapshot after writes finish. Old generations and snapshots taken before an active write should not overwrite newer counts. On resume or a missed WebSocket event, tabs should be able to converge without relying on the event delta history.
This describes a tested fork prototype and its recovery design; it is not a reproduced clean-upstream bug report. It is separate from category-tab total/unread semantics (#597) and row account scoping (#607), though those paths would consume the same scoped state.
The prototype includes an explicit fresh IMAP STATUS proof before releasing stranded optimistic state, plus guards for active writes and stale responses. Synthetic tests cover concurrent mutations and recovery. Reference: https://github.com/shady2k/mailflow/tree/integration/upstream-20261014 . No mailbox content or private logs are included. Feedback on the API and ownership of this state would help split a focused upstream change.
All reactions