Key calls by account and sid, not by sid alone

asxpi commented

2026-10-17 07:52:52 +00:00

First-time contributor

One of our accounts calling another that is signed in on the same app
gives both legs one sid: our outgoing call and the other account's
ring. Calls kept them in one record keyed by the sid, so the ring took
the place of our call. When another device took the ring, its Ended
(other_device) closed our call screen, and our hang-up went out as
calls hangup for the account that was not in the call, a no-op in the
backend. The peer was never told and kept the call up with its video
frozen.

Give each call a key: its sid, or account and sid when another of our
accounts already holds that sid. The store, the sheets and the sinks
know a call by the key, and the backend is still sent the call's own
acc and sid, so a call nobody else shares behaves as before.

A test in calls.test.ts plays the case through: the ring does not take
our call, and our hang-up names our account.

🤖 Generated with Claude Code

One of our accounts calling another that is signed in on the same app gives both legs one sid: our outgoing call and the other account's ring. Calls kept them in one record keyed by the sid, so the ring took the place of our call. When another device took the ring, its Ended (other_device) closed our call screen, and our hang-up went out as calls hangup for the account that was not in the call, a no-op in the backend. The peer was never told and kept the call up with its video frozen. Give each call a key: its sid, or account and sid when another of our accounts already holds that sid. The store, the sheets and the sinks know a call by the key, and the backend is still sent the call's own acc and sid, so a call nobody else shares behaves as before. A test in calls.test.ts plays the case through: the ring does not take our call, and our hang-up names our account. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

asxpi added 1 commit

2026-10-17 07:52:52 +00:00

Key calls by account and sid, not by sid alone 31927bafea

One of our accounts calling another that is signed in on the same app
gives both legs one sid: our outgoing call and the other account's
ring. Calls kept them in one record keyed by the sid, so the ring took
the place of our call. When another device took the ring, its Ended
(other_device) closed our call screen, and our hang-up went out as
calls hangup for the account that was not in the call, a no-op in the
backend. The peer was never told and kept the call up with its video
frozen.

Give each call a key: its sid, or account and sid when another of our
accounts already holds that sid. The store, the sheets and the sinks
know a call by the key, and the backend is still sent the call's own
acc and sid, so a call nobody else shares behaves as before.

A test in calls.test.ts plays the case through: the ring does not take
our call, and our hang-up names our account.

Assisted-by: Claude:opus-5-5

ann commented

2026-10-17 14:59:12 +00:00

Owner

Merged by hand into develop (now at aa2fbcbf, on libtacky v1.0.0-rc10) for the 0.6.0 testing builds; thank you. One follow-up went in on top (c0a209eb): the camera action sent the call's key as the sid to the native engine on the leg with the composite key; it now sends call.sid, with a test. The Kotlin side still keys calls by sid alone (CallCoordinator.kt), noted for a later change. Closing, since a hand merge is not marked merged here.

Merged by hand into develop (now at aa2fbcbf, on libtacky v1.0.0-rc10) for the 0.6.0 testing builds; thank you. One follow-up went in on top (c0a209eb): the camera action sent the call's key as the sid to the native engine on the leg with the composite key; it now sends call.sid, with a test. The Kotlin side still keys calls by sid alone (CallCoordinator.kt), noted for a later change. Closing, since a hand merge is not marked merged here.

ann closed this pull request

2026-10-17 14:59:12 +00:00

ann referenced this pull request from a commit

2026-10-17 14:59:13 +00:00

The camera action tells the backend the call's sid, not its key: since PR #20 keys a call by account and sid (acc\nsid once another of our accounts holds the same sid), callAction gets the key, and the native engine's calls set_video was sent that key as the sid on the leg that had the composite one, so the camera toggle did nothing there -- the very case the PR is for. A calls.test.ts case rings another of our accounts under the same sid on a native engine and checks the set_video request names the sid.

ann referenced this pull request from a commit

2026-10-17 14:59:13 +00:00

libtacky pinned at v1.0.0-rc10 (8411e18f): rc9 plus the room mute (chats mute on a room turns its mentions off), PR #8 (Android frame streams renamed where link() is denied, so a video call on the phone shows both pictures), PR #9 (session-terminate when the media path closes or fails), PR #10 (SRV looked up through Android's own resolver) and the OMEMO trust-list republishing loop between two verified devices; the changelog's Unreleased section names what this build gains from the merged app PRs #19, #20 and #21, the libtacky fixes, and the OMEMO mode re-read, and says that another device's OMEMO on/off choice is taken only once that device is verified by hand, which is libtacky's rule.

ann referenced this pull request from a commit

2026-10-25 23:58:00 +00:00

A stranger's self-chosen name never stands alone (the user, 2026-10-26: "yes, show the address beside a stranger's name"): state/stranger-name.ts reads libtacky 1.1.3's ChatEntry.name_source and calls a name unvouched when it is `carried` (whatever the sender's message or subscribe said) or `pep` on someone with no roster item; the roster's own name and a roster contact's published nick stand alone as before (the user put that address in the roster themselves, and an address on every unnamed contact would be noise), rooms and PMs carry no source and are left alone (a room author's name has no source in libtacky's author <Changed> yet, audit #20). The address is the bare JID, "via Telegram" for a user behind a Telegram gateway (a numeric JID nobody reads; the transport is vouched by the roster, and the suffix still tells the stranger from the roster's contact of the same name), the Matrix ID for a Matrix gateway user (user%server@gw -> @user:server). Where it shows: the chat list row's title line, dim, after the name, middle-ellipsised (views/address-tag.ts: a head that ellipsises and an unshrinking tail, the address giving way first but never below its tail, the name keeping a few letters; one row height kept), and the row menu's subtitle; the chat header's line under the name, address first then the presence; the contact page, whole under the name with a hover saying why; notification titles "Mom (mallory@evil.example)", also with no row yet for the nick a 1:1 message carried, the Telegram secret chat nudge, the locked-time notice context and the push worker's mirrored names; the contacts-first gate's words; message search hits; the call screen under the caller's name; the call list. The request bar already showed the address, and the pickers' rows already carry the JID on their second line. Mock: Kadri Kask is a carried name, `mock stranger` takes { jid, name }, roster add/rename set name_source roster. i18n: "via {network}" and the contact page's hover in all six catalogues. Tests: stranger-name.test.ts (the rules, the address words, the split, the list row, the header, the contact page, notifications with and without a row, the notice context, the gate's words, search hits), call-views and calls-sheet cases; notifications, push and one-client expectations moved to the address now beside a nick no row vouches for. CHANGELOG Unreleased.

ann referenced this pull request from a commit