Repository navigation
Read whole DMs and end a channel's manual unread at the bottom - #616
Conversation
wesbillman
left a comment
wesbillman
left a comment
There was a problem hiding this comment.
Carl, an automated reviewer, commenting via Wes’s GitHub account.
Two P2 correctness blockers; see the inline findings. The change is small and reuses the right owner, but does not yet meet the 9/10 correctness bar. Merge criteria: transaction-safe automatic clearing of local intent, and one verified DM cutoff per earned dwell, with regression tests for both races.
Reviewed head ed7c046148269f5cbceb7a7e7c3d0991d4e1a466 against base a244e3cb3c3eaebac9d8daacdc9028b3eb35bab8. Existing CI, including Chromium/WebKit journeys and DCO, is green. Additional deterministic reproductions use the real read-state owners plus a mounted useReading hook: the changed code loses a newer cross-window manual mark and acknowledges post-dwell replies. Corresponding controls pass with the base version of unread.ts. These are service/hook tests with modeled geometry, not native/mobile or live-browser acceptance; broad CI was not rerun locally.
Reading a DM now reads all of it: any earned dwell advances the DM's channel frontier through the newest retained message, replies included, and ends a manual unread on the DM. Reaching the live bottom of a channel ends a manual unread on the channel. It does not move the channel frontier, so mentions, replies and marked messages stay unread. This matches the mobile app. Signed-off-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>
After reload, webkit does not always restore the live bottom, so the bottom catch-up never ran. Keep Channel settings open and scroll the timeline to the bottom first. Signed-off-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>
Automatic reads checked manual-unread intent against this window's cached journal, so an older dwell in one window could erase a newer Mark unread saved by another window. The lease's validity now receives the journal being replaced inside the storage transaction. A dwell's catchUp and observe each took a fresh whole-DM cutoff, so a message arriving between them was read without being seen. A lease now takes one cutoff, witnessed by a verified message, and an empty observe no longer reads the DM. Signed-off-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>
ed7c046 to
5c60392
Compare
wesbillman
left a comment
wesbillman
left a comment
There was a problem hiding this comment.
No further changes requested: the transaction-current manual-intent fence and single witnessed DM cutoff address both findings in the previous review. I found no additional actionable regression in the fixes or public PR material.
Star Lord’s automated source review via Wes’s account; head 5c603922cdfd71889bec78654b1af6d5a9f5cb15, base 0bbf40aaaf0eb1fdb21a08c3500079922c812df8. Source-only: no tests or apps run; the CI required snapshot is successful (Windows skipped), while native/mobile and human acceptance remain unverified.
wesbillman
left a comment
wesbillman
left a comment
There was a problem hiding this comment.
Carl, an automated reviewer, commenting via Wes’s GitHub account.
No remaining blockers from my previous review. Both races are fixed, and I found no actionable regression in the fixes. The change now meets the 9/10 correctness/minimalness bar for this follow-up. This is a review comment, not an approval.
Reviewed head 5c603922cdfd71889bec78654b1af6d5a9f5cb15 against base 0bbf40aaaf0eb1fdb21a08c3500079922c812df8. Seven focused checks passed: the original three cross-window cases, mounted-reader late-arrival cases with fully visible and tall rows, pending-witness deletion, and the submitted mounted DM regression. These exercised unchanged production files with local reviewer probes and modeled geometry. An independent transaction/cancellation source review found no blockers. Existing CI is green, with Windows native validation skipped; broad suites were not duplicated locally. Native/mobile and human acceptance remain unverified.
Main now reads a whole DM on catch-up (#616), so a DM never gets an activity: mark. Update the DM case and the coverage note to match. Signed-off-by: Larry <627498bd4bd1f281a16431e3c6cce3b5c25b6692798c78672298aefbf2f8f8b5@buzz.block.builderlab.xyz>
* origin/main: fix(messages): show real upload progress for composer images (#662) chore: Use redirect_uri for oauth callback and enable Builderlab state param (#643) test(browser): measure the search palette after its entrance (#653) Cache coverage checks during read-state retention (#576) docs: define shared host boundaries and Canvas contract cases (#658) feat(plugins): allow bounded output for agent tool queries (#649) test(browser): measure GitHub panel actions after its enter motion (#625) Read whole DMs and end a channel's manual unread at the bottom (#616) docs: update Goose session viewer and add Pi session viewer (#651) Add Hermes Agent as a Tier 2 harness (#629) Signed-off-by: Luna <0828be588d5c9c4d092c5b1cf01761d5aef63d6b104807fd8e8284d517dd1a99@buzz.block.builderlab.xyz>
2 participants
🤖
Summary
Two places where Buzz kept showing something as unread after you had read it:
What does not change:
The mobile app gets the same two changes, so both apps behave the same.
Details
src/features/relay/unread.ts: in a DM, the reading logic now moves the DM's "read up to" point to the newest loaded message, and clears the DM's manual unread. In a DM it writes no per-message read records. The "read up to" time always comes from a message's own timestamp, never from the device clock.docs/unread.mddescribes the new rules, including the table of which actions clear which unread states.