vault/vrds

Watch

Fork 0

[breaking] chore: update rust dependencies (2 updates) #310

Closed

Wuast94 wants to merge 1 commit from updatarr/rust-dependencies-17605dfeaf into main

Wuast94 commented

2026-09-19 03:50:26 +02:00

Owner

Updates rust dependencies.

Breaking change detected. Review the release notes before merging.

Automerge: Disabled because breaking change release notes were detected

Updates

Dependency	From	To	Type	Locations
`matrix-sdk`	`0.19.0`	`0.19.1`	patch; breaking	[1] L24
`matrix-sdk-sqlite`	`0.19.0`	`0.19.1`	patch; breaking	[1] L95

Files

[1] Cargo.toml

Release notes

matrix-sdk (0.19.0 → 0.19.1)

Source comparison unavailable: source refs for this transition are not known.

`matrix-sdk-0.19.0`: matrix-sdk 0.19.0

Breaking change detected.

Source

Source ref: main

What's Changed

Removed

EventCache::save_events has been removed. This API was ideal to create deadlocks in the Event Cache if not used with special care. Moreover, it breaks the isolation between components and was violating the clear separation of responsibility between components.

The implications are the following: Room::messages, Room::event, Room::event_with_context and Room::relations no longer save fetched events in the Event Cache. (#6629)

Added

The ThreadEventCache has a persistent storage! All events received from the server can now be found locally without reaching the server anymore, which means: thread offline support!

Moreover, ThreadEventCache contains more events, like edits and redactions! Previously, these events were located in the RoomEventCache for technical reasons. This is now solved and works as expected. (#6517)
The pinned-events cache, aka PinnedEventsCache, now understands annotations, aka reactions! All annotations to a pinned-event are now part of the pinned-events cache, making it possible to display the reactions live.

At the same time, pinned-events are now deduplicated, along with all the related events, like edits, annotations etc. (#6568)
Add Encryption::dehydrated_devices(), a high-level [MSC3814] dehydrated-device manager that wraps the crypto-crate primitives: probe support, create, rehydrate, delete, weekly rotation, a Secret Storage round trip for the pickle key, and a lifecycle state stream. MatrixMockServer gains mock_* helpers for the four MSC3814 endpoints. (#6606)
Added Account::set_status, Account::clear_status, Account::set_call, and Account::clear_call to write [MSC4426] user-status profile fields (m.status and m.call). All four are gated behind the new unstable-msc4426 feature. (#6616)
Create a MediaFetcher trait that allows us to replace the existing DefaultMediaFetcher with any other implementation when building a Client. (#6625)
R2D2 (aka event_cache::Redecryptor) no longer decrypts an already decrypted event. It was taking its job too seriously. Can't blame it, but it was creating unnecessary update in the Event Cache and to all the listeners (Latest Event, Search, Timeline etc.). (#6629)
Global message search results are now sorted by full-text search relevance score, interleaving results from different rooms by relevance rather than grouping them per room. (#6645)
Add SlidingSync::set_room_subscriptions for updating room subscriptions non-destructively. This is similar to SlidingSync::reset_and_add_room_subscriptions but doesn't clear and then recreate all subscriptions. Instead, it removes existing subscriptions for rooms that are not resubscribed and adds new subscriptions for resubscribed rooms that don't already exist. Unlike SlidingSync::reset_and_add_room_subscriptions`, this method does not mark members as unsynced.(#6652)
Add client-owned presence configuration that can optionally send an immediate presence update. Generated sync requests use the client presence by default, which starts as unavailable, while classic sync settings can still override it per request. (#6672)
ThreadEventCache auto-shrinks when no more subscribers are alive. Subscribers are returned by [ThreadEventCache::subscribe]. When the last subscriber is dropped, it means no external API are listening to this cache, and thus it shrinks to its last chunk of events, to save memory. This behavior was already present for RoomEventCache. (#6690)
Added a new top-level Error variant, SignatureError representing errors that occurred signing or verifying cryptographic data. (#6715)
ClientBuilder::with_x509_signer and ClientBuilder::with_x509_verifier now allow configuring an X.509 signer and verifier for experimental X.509-based cross-signing identity verification.

Gated behind the experimental-x509-identity-verification feature. (#6727)
The new EventCache::forget_room method properly forget a specific room. It is different from EventCache::clear_all_events where (i) it's for a particular room instead of all rooms, (ii) it forgets the room, i.e. it's deleted from the Event Cache entirely instead of being cleared only.

If you want to forget the room client-wise, look at Client::forget_room (which calls EventCache::forget_room at the time of writing). (#6749)
Add Client::subscribe_to_own_profile, returning a stream of the current user's global profile. It emits the currently stored value and re-emits whenever the profile changes during sync, or an empty profile when none is stored. Note: Without the Profiles sliding sync extension enabled only an empty profile will be emitted and no updates will be published. (#6760)
Add Client::rtc_transports that queries the RTC transport advertised by the homeserver using the new authenticated GET /_matrix/client/v1/rtc/transports API. The value is cached for 1 day, and it is possible to force refresh with Client::refresh_rtc_transports, or to by-pass cache with Client::fetch_rtc_transports. [breaking] Client::rtc_foci has been deprecated and should only be used as a fallback for homeservers that do not support the new API (see Client::well_known_rtc_transports). (#6789)
Add support for MSC4515 and RTC transport discovery for widgets. (#6791)
AttachmentConfig gained an extra_content field and RoomSendQueue::send() now returns a builder-style future with with_extra_content(), allowing additional top-level fields to be included in an event's content — e.g. vendor-prefixed keys that server-side push rules can match on (in lieu of MSC3664). Extra fields never override the fields of the event itself. Supported by the send queue and the direct Room::send_attachment path. (#6812)
Added Client::enable_automatic_call_status(bool) an opt-in auto-sync that mirrors this device's MatrixRTC participation into the [MSC4426] m.call profile field. Gated behind the unstable-msc4426 feature.

Also adds Room::is_device_in_active_room_call(user_id, device_id) for device-scoped participation checks. (#6825)
Propagate the current tracing Span into the E2EE setup task. (#6833)
The event cache gained a shared BackPaginationQueue (EventCache::back_pagination_queue): a single background executor that runs back-pagination requests from every use case by Priority, with a bounded in-flight number (EventCacheConfig::max_concurrent_back_paginations) and one
run per room at a time. Enable it with ClientBuilder::with_enable_automatic_back_pagination(true). (#6838)
Add Client::disable_well_known_lookup, which disables every /.well-known/matrix/client request performed by the client, for users that must not emit any request to the well-known URI of their domain. When disabled, Client::tile_server returns None, Client::well_known_rtc_transports returns an empty list, and Client::discover_rtc_transports doesn't fall back to the well-known foci.

Add ClientBuilder::disable_well_known_lookup, which controls the initial value of the Client::disable_well_known_lookup flag and additionally disables any homeserver discovery performed by ClientBuilder::build. If true, the homeserver must then be given with ClientBuilder::homeserver_url. ClientBuilder::server_name and ClientBuilder::username can only be resolved through the well-known, so ClientBuilder::build fails with the new ClientBuildError::WellKnownLookupDisabled variant if they are called.

Add Client::discover_rtc_transports, which returns the MatrixRTC transports advertised by the homeserver (MSC4519), with a fallback to the well-known m.rtc_foci as long as well-known lookup is allowed by the client. Transports are exposed as the new RtcTransport enum, with a LiveKit variant and a Custom variant carrying the raw JSON for transport types the SDK doesn't know about. (#6845)
Added the msc_4388_rendezvous_server_supported() method to check if the homeserver supports MSC4388 for the QR code login. (#6871)
ThreadEventCache now applies new redactions and saves bundled latest thread events for events coming from the pagination, or that have been redecrypted (by the Redecryptor). (#6888)
ThreadEventCache gains 4 new methods:
1. read_receipts to get the ReadReceipts associated to the ThreadInfo,
2. num_unread_messages to get the number of unread messages (alias of
  read_receipts().await?.num_unread),
3. num_unread_notifications to get the number of unread notifications (alias
  of read_receipts().await?.num_notifications),
4. num_unread_mentions to get the number of unread mentions (alias of
  read_receipts().await?.num_mentions).
  (#6893)
Added Room::retention() to retrieve the current retention policy and
RoomPrivacySettings::update_room_retention() to set new values.

Leverages Ruma's unstable-msc1763 feature flag. (#6905)
Added Client::get_retention_configuration to retrieve the server level retention policy and per room overrides so a client can determine whether the server enforces any retention policy constraints. (#6908)
Media::get_media_preview() returns the OpenGraph-like data a homeserver generates for a URL. It uses the authenticated /_matrix/client/v1/media/preview_url endpoint when the homeserver supports
it, and falls back to the deprecated /_matrix/media/v3/preview_url otherwise. (#6918)
Added Room::effective_retention() to return the final retention policy for a room combining server-wide policy with room-specific. Clients can use this value to determine message retention. (#6919)
Add Room::human_member_ids and Room::human_member_ids_no_sync, which return the user IDs of the room members with the given memberships, without the service members declared by the io.element.functional_members state event. This is a convenient way to find the other party of a direct message. (#6925)
SendHandle::abort_with_reason() aborts a queued send like SendHandle::abort(), and attaches a reason to the redaction that materializes the abort when the event had already been sent by the time the abort was processed. The reason is persisted with the pending redaction, so it survives restarts like the rest of the send queue. (#6931)
ThreadEventCache gains a new method: subscribe_to_thread_info to subscribe to updates of the ThreadInfo. This is very similar to Room::subscribe_info. (#6993)
Add Client::total_unread_notifications, returning the sum of the client-side computed unread notification counts across all joined rooms, counting rooms marked as unread by hand as one each. (#7002)

Changed

RoomEventCacheUpdate::AddEphemeralEvents becomes RoomEventCacheUpdate::AddReadReceipt. It no longer contains an events: Vec<Raw<AnySyncEphemeralRoomEvent>> but a single event: ReceiptEventContent.

In the same vein, ThreadEventCacheUpdate is introduced! Thus, ThreadEventCache::subscribe no longer returns a Subscriber<TimelineVectorDiffs> but a Subscriber<ThreadEventCacheUpdate>. The TimelineVectorDiffs can be found in ThreadEventCacheUpdate::UpdateTimelineEvents, just like its sibling type RoomEventCacheUpdate. (#6038)

The thread cache, aka ThreadEventCache, has been extracted from the room cache, aka RoomEventCache. To get a thread cache, please use the EventCache::thread() method.

The RoomEventCache::thread_pagination and subscribe_to_thread methods have been removed.

Before:

let room_cache = event_cache.for_room(room_id).await?;
let thread_pagination = room_cache.thread_pagination(thread_id).await?;
let thread_subscriber = room_cache.subscribe_to_thread(thread_id)?;

After:

let thread_cache = event_cache.thread(room_id, thread_id).await?;
let thread_pagination = thread_cache.pagination(); // no async, no error!
let thread_subscriber = thread_cache.subscribe().await?;

Also, the EventCache::for_room method has been renamed room. (#6517)

The pinned-events cache, aka PinnedEventsCache, has been extracted from the room cache, aka RoomEventCache. To get a pinned-events cache, please use the EventCache::pinned_events() method.

The RoomEventCache::subscribe_to_pinned_events method have been removed and replaced by PinnedEventsCache::subscribe.

Before:

let room_cache = event_cache.room(room_id).await?;
let pinned_events_subscriber = room_cache.subscribe_to_pinned_events()?;

After:

let pinned_events_cache = event_cache.pinned_events(room_id).await?;
let pinned_events_subscriber = pinned_events_cache.subscribe().await?;
``` ([#6568](https://github.com/matrix-org/matrix-rust-sdk/pull/6568))

The event-focused cache, aka EventFocusedCache, has been extracted from the room cache, aka RoomEventCache. To get an event-focused cache, please use the EventCache::event_focused() method.

The RoomEventCache::get_or_create_event_focused_cache and get_event_focused_cache methods have been removed.

Before:

let room_cache = event_cache.for_room(room_id).await?;
let event_focused_cache = room_cache.get_or_create_event_focused_cache(thread_mode, number_of_initial_events).await?;

After:

let event_focused_cache = event_cache.event_focused(room_id, event_id, thread_mode, number_of_initial_events).await?;
``` ([#6603](https://github.com/matrix-org/matrix-rust-sdk/pull/6603))

The EventCacheStore::clear_all_linked_chunks trait method has been renamed clear_all_events to express its intent better. (#6629)
[breaking] The message search helpers now expose their results as Streams of pages instead of bespoke iterators. Room::search_messages returns a stream of pages of (score, event_id) pairs (use the new Room::search_messages_events for full TimelineEvents), and GlobalSearchBuilder::build/build_events replace the removed RoomSearchIterator/GlobalSearchIterator types. (#6645)
Put ClientInner::MediaFetcher behind a RwLock so it can be replaced at any time from any thread, without having a mutable ClientInner reference around. (#6689)
The event_cache::RoomEventCacheSubscriber type has been renamed to event_cache::Subscriber<RoomEventCacheUpdate>. The type has been made generic to be reused by ThreadEventCache. The behaviour is exactly the same as it keeps Deref and DerefMut to tokio::sync::mpsc::Receiver. (#6690)
Index media, stickers, polls and more message types. Move message parsing responsibility outside the search crate and into the main SDK one. (#6710)
[breaking] GrantLoginProgress::WaitingForAuth now has a continuation_sender: ContinuationMessageSender field. Applications must call continuation_sender.confirm() once the verification URI has been opened in the browser and the application is ready to proceed, or continuation_sender.cancel() to abort; previously it proceeded automatically as soon as this state was reached. This lets applications that suspend or navigate away while the verification URI is open to resume the process explicitly.

[breaking] qrcode::CheckCodeSenderError has been renamed to qrcode::SenderError, since it is now shared with ContinuationMessageSender. (#6711)
[breaking] RoomPreview::heroes is now Option<Vec<RoomHeroWithProfile>> instead of Option<Vec<RoomHero>>, exposing each hero's status and call fields from their global profile. These fields are only populated when syncing via sliding sync with the profiles extension enabled. (#6733)
Methods that delete key backup versions, such as resetting your cryptographic identity now iteratively delete all server-side key backup versions. (#6744)
[breaking] QRCodeGrantLoginError::UnexpectedMessage's received field type and ClientBuildError::AutoDiscovery's inner type are now Boxed to reduce the error enums' sizes.
(#6763)
To-device messages forwarded to widgets now include the MSC3819 encryptedboolean in their data, indicating whether the original to-device message was received encrypted. (MSC3819) (#6795)
[breaking] The check code in the QR code auth is now represented as an u8 directly instead of the CheckCode struct. (#6819)
Subscribing to an already subscribed room now refreshes its settings, e.g. its timeline_limit, instead of silently keeping the previous ones. (#6828)
[breaking] Background back-pagination is now driven by the shared BackPaginationQueue instead of a per-room credit system. (#6838)
The latest event of a room is now computed with automatic back-pagination: when no suitable candidate is in memory, the room's history is back-paginated until a candidate is found or the start of the timeline is reached. Requires ClientBuilder::with_enable_automatic_back_pagination(true). (#6840)
The send queue now sends requests strictly in the order they were queued, in a given room: a request that failed with an unrecoverable error (and got marked as wedged) blocks all the subsequent requests in the same room from being sent, until it's either retried with SendHandle::unwedge() or removed with SendHandle::abort(). Previously, a wedged request was skipped over, causing later messages to be sent before it, i.e. out of order.
- A failed media upload's send error is now reflected on the media event's
  local echo returned by RoomSendQueue::subscribe(), instead of the echo
  claiming the media is still being sent.
- An HTTP 520 status code (a non-standard "unknown server error", returned by
  Cloudflare and other reverse proxies) is now considered a transient error
  again, like the other 5xx status codes: requests failing with it will be
  retried, and the send queue treats it as a recoverable error (keeping the
  failed request in the queue) instead of wedging the request.
(#6843)
Widget's get_rtc_transports action is now using the Client::discover_rtc_transports API that checks the new authenticated GET /_matrix/client/v1/rtc/transports API, and falls back to the .well-known/matrix/client lookup if allowed by the client. Previously it was never falling back to the well-known lookup. (#6845)
The room subscription methods of SlidingSync are renamed after what they do to the set of subscriptions: subscribe_to_rooms becomes add_room_subscriptions, unsubscribe_to_rooms becomes remove_room_subscriptions, resubscribe_to_rooms becomes set_room_subscriptions, and clear_and_subscribe_to_rooms becomes reset_and_add_room_subscriptions. (#6927)
Account automatically updates the local copy of the user's global profile after updating (to act as a local echo). Requires support for global profile syncing. (#6984)
Fix flakiness in the cross-process store lock tests. Dropping a guard does not release the lock at once: the lease-renewal task publishes the release on its next tick. The tests now acquire with spin_lock_store, which retries with a backoff, instead of a single attempt after a fixed sleep. Each run also gets its own temporary directory for the sqlite database, instead of a predictable shared path under the system temporary directory that carried state into the next run. (#6899)
Add Client::send_encrypted_to_device (behind the experimental-send-custom-to-device feature) to Olm-encrypt a custom to-device message and send it to a set of recipient devices. (#6981)

Fixed

The ThreadSummary of a ThreadEventCache is computed by using the latest_events internal API, making it understand more events and delivering a more accurate result. This is still not perfect and complete but at least it's one step in the right direction! (#6517)
Resolve the room alias to a room id before evaluating pre-join info. This resolves the asymmetry when using join_room_by_id_or_alias, ensuring possibly crucial invite handling is handled correctly. (#6618)
The Event Cache has now a unique state for all the caches for all the rooms. It fixes numerous bugs but the most notable one is: when a cache has to be reloaded because its state has been invalidated, all the other caches are now reloaded at the same time. Previously, the invalidation was correctly detected, then a single cache was reloaded, and the invalidation was marked as done, thus the other caches couldn't detect it and reload accordingly. It was creating an inconsistency in the data. (#6629)
Prevent deadlocks when waking up the thread subscription catch-up task. (#6667)
Ensure a VectorDiff is not emitted to subscribers when an event cache is shrunk because no subscribers is listening to it. If there is no subscribers, it's useless to send it anyway, and if in the future, a race happens and a subscriber exists, it must not receive this diff. At the same time, the way the subscribers are tracked and counted have been revisited to use less space in memory and use less atomic operations. (#6684)
It is now possible to prevent matrix-sdk from pulling in aws-lc-rs as a transitive dependency by deactivating the newly introduced rustls-aws-lc-rs default feature. This is a breaking change for downstream users relying on matrix-sdk without default features, who now have to explicitly enable rustls-aws-lc-rs or register a rustls provider explicitly. (#6705)
CrossSigningResetHandle::auth now uses the common timeout helper (from matrix-sdk-base) instead of tokio::time directly. (#6719)
Export SupportedPathBuilder and SupportedAuthScheme Client::send uses these in it's trait bounds, but because they were private it wasn't possible to write dependent code that is generic over request types. (#6734)
Replace tokio::time::sleep with matrix_sdk_common::sleep::sleep to ensure QRCode auth compiles to WebAssembly. (#6743)
EventCache::clear_all_rooms no longer recreates an empty LinkedChunk per cache (RoomEventCache, ThreadEventCache etc.). The LinkedChunk's first chunk being now created lazily, it ensures clear_all_rooms can be used to clear or to forget data. (#6752)
The correct semantics for one-time key counts are now used when sync v2 is being used. This bug could have previously led to no one-time keys being uploaded to the homeserver. (#6780)
SlidingSync::set_room_subscriptions now cancels in-flight requests whenever the set of room subscriptions changes. It used to do so only when a subscription was added and another one was removed in the same call, so a first-time subscription or a singular removal was not sent until the current long poll expired, up to 30s later. (#6828)
The LatestEventValue of a room is no longer left unchanged when it moves between local states, e.g. when a second local event is queued, when the send queue wedges (LocalIsSending to LocalCannotBeSent), or when a send is retried. These values have no event ID, and were incorrectly considered equal to each other. (#6836)
Fix homeserver switching during QR code login when using rendezvous servers that advertise a valid .well-known configuration. (#6846)
The window in which an app can be signed out by another process rotating the OAuth 2.0 refresh token has been narrowed. The refresh now fetches the authorization server metadata before it reads the token, so a rotation during that request no longer makes the app send a stale token. Being suspended during the token exchange can still produce a sign-out. (#6860)
A rare deadlock has been fixed in event_cache::caches::aggregator::aggregate_timeline_for_threads.

When aggregating the timeline for each thread, we are doing look ups in the room cache, and in the thread caches. Every time we do a look up with find_event, we are acquiring a read lock over the state. Because look up in the room cache can happen frequently, we were acquiring a read lock over the state for the room cache for the entire lifetime of the aggregate_timeline_for_threads function. However, in some situations, it could lead to a deadlock.

Imagine task T1 is running handle_joined_room_update (which runs aggregate_timeline_for_threads): a read lock over the room state is acquired (L1), and the read lock over a thread state must be acquired for another look up (L2). Before L2 is acquired, task T2 is running a different part of the code, like a pagination, and waits to acquire a write lock (L3). It waits because L1 is still alive and L3 needs an exclusive access. Now, L2 enters the game and cannot be acquired because L3 is waiting. So we have:
- L1 is acquired but not released,
- L1 is waiting on L3 to release itself,
- L2 is waiting on L1,
- L3 is waiting on L2.
This is a deadlock.

The fix consists of acquiring a read lock over (all) the states once, to then project the read lock guard to the room and to each thread. This way, L1 is now a combination of L1 and L3. L2 will always wait on L1 to be released. (#6869)
Media::get_media_content now deterministically returns the content of a local media (i.e. a media sent by the user not yet received from the server). (#6896)
SlidingSync::reset_and_add_room_subscriptions now cancels the in-flight request when it removes the last subscriptions. Called with an empty list of rooms, it used to clear the subscriptions without cancelling the request, so the server kept sending the rooms until the current long poll expired, up to 30s later. (#6932)
When a sync contained more than one read receipts, only the first one was taken into account. Multiple problems arose from this, e.g. (i) perhaps the first selected receipt is not the correct one (RoomEventCache only cares about the Main or Unthreaded for example, and should ignore the Thread ones), (ii) or more simply the other read receipts were totally ignored! This is now fixed: ephemeral events are triaged similarly to timeline events, inside event_cache::aggregator. (#6938)
Deduplicate thread-subscription catch-up tokens to prevent duplicate catch-up requests. (#6955)
Fix unpinning an event not updating the pinned events timeline through pinned_event_listener_task.

Adding a pinned event updated the list, clearing it also did, but removing one or several items from it, as long as it wasn't empty, didn't. (#6991)
Local medias (as in attachments pending in the send queue) are always cached with MediaFormat::File, be it the file itself or its thumbnail, so requesting one with MediaFormat::Thumbnail was failing since the format started being taken into account. (#6996)
Restore checksums for UniFFI functions. They were removed to work around an issue where the checksums were incorrectly computed first for some 32bit ARM devices and then for some 64bit ones. That issue is now fixed so we should restore the checksums. (#7046)

matrix-sdk-sqlite (0.19.0 → 0.19.1)

Source comparison unavailable: source refs for this transition are not known.

`matrix-sdk-0.19.0`: matrix-sdk 0.19.0

Breaking change detected.

Source

Source ref: main

What's Changed

Removed

EventCache::save_events has been removed. This API was ideal to create deadlocks in the Event Cache if not used with special care. Moreover, it breaks the isolation between components and was violating the clear separation of responsibility between components.

The implications are the following: Room::messages, Room::event, Room::event_with_context and Room::relations no longer save fetched events in the Event Cache. (#6629)

Added

The ThreadEventCache has a persistent storage! All events received from the server can now be found locally without reaching the server anymore, which means: thread offline support!

Moreover, ThreadEventCache contains more events, like edits and redactions! Previously, these events were located in the RoomEventCache for technical reasons. This is now solved and works as expected. (#6517)
The pinned-events cache, aka PinnedEventsCache, now understands annotations, aka reactions! All annotations to a pinned-event are now part of the pinned-events cache, making it possible to display the reactions live.

At the same time, pinned-events are now deduplicated, along with all the related events, like edits, annotations etc. (#6568)
Add Encryption::dehydrated_devices(), a high-level [MSC3814] dehydrated-device manager that wraps the crypto-crate primitives: probe support, create, rehydrate, delete, weekly rotation, a Secret Storage round trip for the pickle key, and a lifecycle state stream. MatrixMockServer gains mock_* helpers for the four MSC3814 endpoints. (#6606)
Added Account::set_status, Account::clear_status, Account::set_call, and Account::clear_call to write [MSC4426] user-status profile fields (m.status and m.call). All four are gated behind the new unstable-msc4426 feature. (#6616)
Create a MediaFetcher trait that allows us to replace the existing DefaultMediaFetcher with any other implementation when building a Client. (#6625)
R2D2 (aka event_cache::Redecryptor) no longer decrypts an already decrypted event. It was taking its job too seriously. Can't blame it, but it was creating unnecessary update in the Event Cache and to all the listeners (Latest Event, Search, Timeline etc.). (#6629)
Global message search results are now sorted by full-text search relevance score, interleaving results from different rooms by relevance rather than grouping them per room. (#6645)
Add SlidingSync::set_room_subscriptions for updating room subscriptions non-destructively. This is similar to SlidingSync::reset_and_add_room_subscriptions but doesn't clear and then recreate all subscriptions. Instead, it removes existing subscriptions for rooms that are not resubscribed and adds new subscriptions for resubscribed rooms that don't already exist. Unlike SlidingSync::reset_and_add_room_subscriptions`, this method does not mark members as unsynced.(#6652)
Add client-owned presence configuration that can optionally send an immediate presence update. Generated sync requests use the client presence by default, which starts as unavailable, while classic sync settings can still override it per request. (#6672)
ThreadEventCache auto-shrinks when no more subscribers are alive. Subscribers are returned by [ThreadEventCache::subscribe]. When the last subscriber is dropped, it means no external API are listening to this cache, and thus it shrinks to its last chunk of events, to save memory. This behavior was already present for RoomEventCache. (#6690)
Added a new top-level Error variant, SignatureError representing errors that occurred signing or verifying cryptographic data. (#6715)
ClientBuilder::with_x509_signer and ClientBuilder::with_x509_verifier now allow configuring an X.509 signer and verifier for experimental X.509-based cross-signing identity verification.

Gated behind the experimental-x509-identity-verification feature. (#6727)
The new EventCache::forget_room method properly forget a specific room. It is different from EventCache::clear_all_events where (i) it's for a particular room instead of all rooms, (ii) it forgets the room, i.e. it's deleted from the Event Cache entirely instead of being cleared only.

If you want to forget the room client-wise, look at Client::forget_room (which calls EventCache::forget_room at the time of writing). (#6749)
Add Client::subscribe_to_own_profile, returning a stream of the current user's global profile. It emits the currently stored value and re-emits whenever the profile changes during sync, or an empty profile when none is stored. Note: Without the Profiles sliding sync extension enabled only an empty profile will be emitted and no updates will be published. (#6760)
Add Client::rtc_transports that queries the RTC transport advertised by the homeserver using the new authenticated GET /_matrix/client/v1/rtc/transports API. The value is cached for 1 day, and it is possible to force refresh with Client::refresh_rtc_transports, or to by-pass cache with Client::fetch_rtc_transports. [breaking] Client::rtc_foci has been deprecated and should only be used as a fallback for homeservers that do not support the new API (see Client::well_known_rtc_transports). (#6789)
Add support for MSC4515 and RTC transport discovery for widgets. (#6791)
AttachmentConfig gained an extra_content field and RoomSendQueue::send() now returns a builder-style future with with_extra_content(), allowing additional top-level fields to be included in an event's content — e.g. vendor-prefixed keys that server-side push rules can match on (in lieu of MSC3664). Extra fields never override the fields of the event itself. Supported by the send queue and the direct Room::send_attachment path. (#6812)
Added Client::enable_automatic_call_status(bool) an opt-in auto-sync that mirrors this device's MatrixRTC participation into the [MSC4426] m.call profile field. Gated behind the unstable-msc4426 feature.

Also adds Room::is_device_in_active_room_call(user_id, device_id) for device-scoped participation checks. (#6825)
Propagate the current tracing Span into the E2EE setup task. (#6833)
The event cache gained a shared BackPaginationQueue (EventCache::back_pagination_queue): a single background executor that runs back-pagination requests from every use case by Priority, with a bounded in-flight number (EventCacheConfig::max_concurrent_back_paginations) and one
run per room at a time. Enable it with ClientBuilder::with_enable_automatic_back_pagination(true). (#6838)
Add Client::disable_well_known_lookup, which disables every /.well-known/matrix/client request performed by the client, for users that must not emit any request to the well-known URI of their domain. When disabled, Client::tile_server returns None, Client::well_known_rtc_transports returns an empty list, and Client::discover_rtc_transports doesn't fall back to the well-known foci.

Add ClientBuilder::disable_well_known_lookup, which controls the initial value of the Client::disable_well_known_lookup flag and additionally disables any homeserver discovery performed by ClientBuilder::build. If true, the homeserver must then be given with ClientBuilder::homeserver_url. ClientBuilder::server_name and ClientBuilder::username can only be resolved through the well-known, so ClientBuilder::build fails with the new ClientBuildError::WellKnownLookupDisabled variant if they are called.

Add Client::discover_rtc_transports, which returns the MatrixRTC transports advertised by the homeserver (MSC4519), with a fallback to the well-known m.rtc_foci as long as well-known lookup is allowed by the client. Transports are exposed as the new RtcTransport enum, with a LiveKit variant and a Custom variant carrying the raw JSON for transport types the SDK doesn't know about. (#6845)
Added the msc_4388_rendezvous_server_supported() method to check if the homeserver supports MSC4388 for the QR code login. (#6871)
ThreadEventCache now applies new redactions and saves bundled latest thread events for events coming from the pagination, or that have been redecrypted (by the Redecryptor). (#6888)
ThreadEventCache gains 4 new methods:
1. read_receipts to get the ReadReceipts associated to the ThreadInfo,
2. num_unread_messages to get the number of unread messages (alias of
  read_receipts().await?.num_unread),
3. num_unread_notifications to get the number of unread notifications (alias
  of read_receipts().await?.num_notifications),
4. num_unread_mentions to get the number of unread mentions (alias of
  read_receipts().await?.num_mentions).
  (#6893)
Added Room::retention() to retrieve the current retention policy and
RoomPrivacySettings::update_room_retention() to set new values.

Leverages Ruma's unstable-msc1763 feature flag. (#6905)
Added Client::get_retention_configuration to retrieve the server level retention policy and per room overrides so a client can determine whether the server enforces any retention policy constraints. (#6908)
Media::get_media_preview() returns the OpenGraph-like data a homeserver generates for a URL. It uses the authenticated /_matrix/client/v1/media/preview_url endpoint when the homeserver supports
it, and falls back to the deprecated /_matrix/media/v3/preview_url otherwise. (#6918)
Added Room::effective_retention() to return the final retention policy for a room combining server-wide policy with room-specific. Clients can use this value to determine message retention. (#6919)
Add Room::human_member_ids and Room::human_member_ids_no_sync, which return the user IDs of the room members with the given memberships, without the service members declared by the io.element.functional_members state event. This is a convenient way to find the other party of a direct message. (#6925)
SendHandle::abort_with_reason() aborts a queued send like SendHandle::abort(), and attaches a reason to the redaction that materializes the abort when the event had already been sent by the time the abort was processed. The reason is persisted with the pending redaction, so it survives restarts like the rest of the send queue. (#6931)
ThreadEventCache gains a new method: subscribe_to_thread_info to subscribe to updates of the ThreadInfo. This is very similar to Room::subscribe_info. (#6993)
Add Client::total_unread_notifications, returning the sum of the client-side computed unread notification counts across all joined rooms, counting rooms marked as unread by hand as one each. (#7002)

Changed

RoomEventCacheUpdate::AddEphemeralEvents becomes RoomEventCacheUpdate::AddReadReceipt. It no longer contains an events: Vec<Raw<AnySyncEphemeralRoomEvent>> but a single event: ReceiptEventContent.

In the same vein, ThreadEventCacheUpdate is introduced! Thus, ThreadEventCache::subscribe no longer returns a Subscriber<TimelineVectorDiffs> but a Subscriber<ThreadEventCacheUpdate>. The TimelineVectorDiffs can be found in ThreadEventCacheUpdate::UpdateTimelineEvents, just like its sibling type RoomEventCacheUpdate. (#6038)

The thread cache, aka ThreadEventCache, has been extracted from the room cache, aka RoomEventCache. To get a thread cache, please use the EventCache::thread() method.

The RoomEventCache::thread_pagination and subscribe_to_thread methods have been removed.

Before:

let room_cache = event_cache.for_room(room_id).await?;
let thread_pagination = room_cache.thread_pagination(thread_id).await?;
let thread_subscriber = room_cache.subscribe_to_thread(thread_id)?;

After:

let thread_cache = event_cache.thread(room_id, thread_id).await?;
let thread_pagination = thread_cache.pagination(); // no async, no error!
let thread_subscriber = thread_cache.subscribe().await?;

Also, the EventCache::for_room method has been renamed room. (#6517)

The pinned-events cache, aka PinnedEventsCache, has been extracted from the room cache, aka RoomEventCache. To get a pinned-events cache, please use the EventCache::pinned_events() method.

The RoomEventCache::subscribe_to_pinned_events method have been removed and replaced by PinnedEventsCache::subscribe.

Before:

let room_cache = event_cache.room(room_id).await?;
let pinned_events_subscriber = room_cache.subscribe_to_pinned_events()?;

After:

let pinned_events_cache = event_cache.pinned_events(room_id).await?;
let pinned_events_subscriber = pinned_events_cache.subscribe().await?;
``` ([#6568](https://github.com/matrix-org/matrix-rust-sdk/pull/6568))

The event-focused cache, aka EventFocusedCache, has been extracted from the room cache, aka RoomEventCache. To get an event-focused cache, please use the EventCache::event_focused() method.

The RoomEventCache::get_or_create_event_focused_cache and get_event_focused_cache methods have been removed.

Before:

let room_cache = event_cache.for_room(room_id).await?;
let event_focused_cache = room_cache.get_or_create_event_focused_cache(thread_mode, number_of_initial_events).await?;

After:

let event_focused_cache = event_cache.event_focused(room_id, event_id, thread_mode, number_of_initial_events).await?;
``` ([#6603](https://github.com/matrix-org/matrix-rust-sdk/pull/6603))

The EventCacheStore::clear_all_linked_chunks trait method has been renamed clear_all_events to express its intent better. (#6629)
[breaking] The message search helpers now expose their results as Streams of pages instead of bespoke iterators. Room::search_messages returns a stream of pages of (score, event_id) pairs (use the new Room::search_messages_events for full TimelineEvents), and GlobalSearchBuilder::build/build_events replace the removed RoomSearchIterator/GlobalSearchIterator types. (#6645)
Put ClientInner::MediaFetcher behind a RwLock so it can be replaced at any time from any thread, without having a mutable ClientInner reference around. (#6689)
The event_cache::RoomEventCacheSubscriber type has been renamed to event_cache::Subscriber<RoomEventCacheUpdate>. The type has been made generic to be reused by ThreadEventCache. The behaviour is exactly the same as it keeps Deref and DerefMut to tokio::sync::mpsc::Receiver. (#6690)
Index media, stickers, polls and more message types. Move message parsing responsibility outside the search crate and into the main SDK one. (#6710)
[breaking] GrantLoginProgress::WaitingForAuth now has a continuation_sender: ContinuationMessageSender field. Applications must call continuation_sender.confirm() once the verification URI has been opened in the browser and the application is ready to proceed, or continuation_sender.cancel() to abort; previously it proceeded automatically as soon as this state was reached. This lets applications that suspend or navigate away while the verification URI is open to resume the process explicitly.

[breaking] qrcode::CheckCodeSenderError has been renamed to qrcode::SenderError, since it is now shared with ContinuationMessageSender. (#6711)
[breaking] RoomPreview::heroes is now Option<Vec<RoomHeroWithProfile>> instead of Option<Vec<RoomHero>>, exposing each hero's status and call fields from their global profile. These fields are only populated when syncing via sliding sync with the profiles extension enabled. (#6733)
Methods that delete key backup versions, such as resetting your cryptographic identity now iteratively delete all server-side key backup versions. (#6744)
[breaking] QRCodeGrantLoginError::UnexpectedMessage's received field type and ClientBuildError::AutoDiscovery's inner type are now Boxed to reduce the error enums' sizes.
(#6763)
To-device messages forwarded to widgets now include the MSC3819 encryptedboolean in their data, indicating whether the original to-device message was received encrypted. (MSC3819) (#6795)
[breaking] The check code in the QR code auth is now represented as an u8 directly instead of the CheckCode struct. (#6819)
Subscribing to an already subscribed room now refreshes its settings, e.g. its timeline_limit, instead of silently keeping the previous ones. (#6828)
[breaking] Background back-pagination is now driven by the shared BackPaginationQueue instead of a per-room credit system. (#6838)
The latest event of a room is now computed with automatic back-pagination: when no suitable candidate is in memory, the room's history is back-paginated until a candidate is found or the start of the timeline is reached. Requires ClientBuilder::with_enable_automatic_back_pagination(true). (#6840)
The send queue now sends requests strictly in the order they were queued, in a given room: a request that failed with an unrecoverable error (and got marked as wedged) blocks all the subsequent requests in the same room from being sent, until it's either retried with SendHandle::unwedge() or removed with SendHandle::abort(). Previously, a wedged request was skipped over, causing later messages to be sent before it, i.e. out of order.
- A failed media upload's send error is now reflected on the media event's
  local echo returned by RoomSendQueue::subscribe(), instead of the echo
  claiming the media is still being sent.
- An HTTP 520 status code (a non-standard "unknown server error", returned by
  Cloudflare and other reverse proxies) is now considered a transient error
  again, like the other 5xx status codes: requests failing with it will be
  retried, and the send queue treats it as a recoverable error (keeping the
  failed request in the queue) instead of wedging the request.
(#6843)
Widget's get_rtc_transports action is now using the Client::discover_rtc_transports API that checks the new authenticated GET /_matrix/client/v1/rtc/transports API, and falls back to the .well-known/matrix/client lookup if allowed by the client. Previously it was never falling back to the well-known lookup. (#6845)
The room subscription methods of SlidingSync are renamed after what they do to the set of subscriptions: subscribe_to_rooms becomes add_room_subscriptions, unsubscribe_to_rooms becomes remove_room_subscriptions, resubscribe_to_rooms becomes set_room_subscriptions, and clear_and_subscribe_to_rooms becomes reset_and_add_room_subscriptions. (#6927)
Account automatically updates the local copy of the user's global profile after updating (to act as a local echo). Requires support for global profile syncing. (#6984)
Fix flakiness in the cross-process store lock tests. Dropping a guard does not release the lock at once: the lease-renewal task publishes the release on its next tick. The tests now acquire with spin_lock_store, which retries with a backoff, instead of a single attempt after a fixed sleep. Each run also gets its own temporary directory for the sqlite database, instead of a predictable shared path under the system temporary directory that carried state into the next run. (#6899)
Add Client::send_encrypted_to_device (behind the experimental-send-custom-to-device feature) to Olm-encrypt a custom to-device message and send it to a set of recipient devices. (#6981)

Fixed

The ThreadSummary of a ThreadEventCache is computed by using the latest_events internal API, making it understand more events and delivering a more accurate result. This is still not perfect and complete but at least it's one step in the right direction! (#6517)
Resolve the room alias to a room id before evaluating pre-join info. This resolves the asymmetry when using join_room_by_id_or_alias, ensuring possibly crucial invite handling is handled correctly. (#6618)
The Event Cache has now a unique state for all the caches for all the rooms. It fixes numerous bugs but the most notable one is: when a cache has to be reloaded because its state has been invalidated, all the other caches are now reloaded at the same time. Previously, the invalidation was correctly detected, then a single cache was reloaded, and the invalidation was marked as done, thus the other caches couldn't detect it and reload accordingly. It was creating an inconsistency in the data. (#6629)
Prevent deadlocks when waking up the thread subscription catch-up task. (#6667)
Ensure a VectorDiff is not emitted to subscribers when an event cache is shrunk because no subscribers is listening to it. If there is no subscribers, it's useless to send it anyway, and if in the future, a race happens and a subscriber exists, it must not receive this diff. At the same time, the way the subscribers are tracked and counted have been revisited to use less space in memory and use less atomic operations. (#6684)
It is now possible to prevent matrix-sdk from pulling in aws-lc-rs as a transitive dependency by deactivating the newly introduced rustls-aws-lc-rs default feature. This is a breaking change for downstream users relying on matrix-sdk without default features, who now have to explicitly enable rustls-aws-lc-rs or register a rustls provider explicitly. (#6705)
CrossSigningResetHandle::auth now uses the common timeout helper (from matrix-sdk-base) instead of tokio::time directly. (#6719)
Export SupportedPathBuilder and SupportedAuthScheme Client::send uses these in it's trait bounds, but because they were private it wasn't possible to write dependent code that is generic over request types. (#6734)
Replace tokio::time::sleep with matrix_sdk_common::sleep::sleep to ensure QRCode auth compiles to WebAssembly. (#6743)
EventCache::clear_all_rooms no longer recreates an empty LinkedChunk per cache (RoomEventCache, ThreadEventCache etc.). The LinkedChunk's first chunk being now created lazily, it ensures clear_all_rooms can be used to clear or to forget data. (#6752)
The correct semantics for one-time key counts are now used when sync v2 is being used. This bug could have previously led to no one-time keys being uploaded to the homeserver. (#6780)
SlidingSync::set_room_subscriptions now cancels in-flight requests whenever the set of room subscriptions changes. It used to do so only when a subscription was added and another one was removed in the same call, so a first-time subscription or a singular removal was not sent until the current long poll expired, up to 30s later. (#6828)
The LatestEventValue of a room is no longer left unchanged when it moves between local states, e.g. when a second local event is queued, when the send queue wedges (LocalIsSending to LocalCannotBeSent), or when a send is retried. These values have no event ID, and were incorrectly considered equal to each other. (#6836)
Fix homeserver switching during QR code login when using rendezvous servers that advertise a valid .well-known configuration. (#6846)
The window in which an app can be signed out by another process rotating the OAuth 2.0 refresh token has been narrowed. The refresh now fetches the authorization server metadata before it reads the token, so a rotation during that request no longer makes the app send a stale token. Being suspended during the token exchange can still produce a sign-out. (#6860)
A rare deadlock has been fixed in event_cache::caches::aggregator::aggregate_timeline_for_threads.

When aggregating the timeline for each thread, we are doing look ups in the room cache, and in the thread caches. Every time we do a look up with find_event, we are acquiring a read lock over the state. Because look up in the room cache can happen frequently, we were acquiring a read lock over the state for the room cache for the entire lifetime of the aggregate_timeline_for_threads function. However, in some situations, it could lead to a deadlock.

Imagine task T1 is running handle_joined_room_update (which runs aggregate_timeline_for_threads): a read lock over the room state is acquired (L1), and the read lock over a thread state must be acquired for another look up (L2). Before L2 is acquired, task T2 is running a different part of the code, like a pagination, and waits to acquire a write lock (L3). It waits because L1 is still alive and L3 needs an exclusive access. Now, L2 enters the game and cannot be acquired because L3 is waiting. So we have:
- L1 is acquired but not released,
- L1 is waiting on L3 to release itself,
- L2 is waiting on L1,
- L3 is waiting on L2.
This is a deadlock.

The fix consists of acquiring a read lock over (all) the states once, to then project the read lock guard to the room and to each thread. This way, L1 is now a combination of L1 and L3. L2 will always wait on L1 to be released. (#6869)
Media::get_media_content now deterministically returns the content of a local media (i.e. a media sent by the user not yet received from the server). (#6896)
SlidingSync::reset_and_add_room_subscriptions now cancels the in-flight request when it removes the last subscriptions. Called with an empty list of rooms, it used to clear the subscriptions without cancelling the request, so the server kept sending the rooms until the current long poll expired, up to 30s later. (#6932)
When a sync contained more than one read receipts, only the first one was taken into account. Multiple problems arose from this, e.g. (i) perhaps the first selected receipt is not the correct one (RoomEventCache only cares about the Main or Unthreaded for example, and should ignore the Thread ones), (ii) or more simply the other read receipts were totally ignored! This is now fixed: ephemeral events are triaged similarly to timeline events, inside event_cache::aggregator. (#6938)
Deduplicate thread-subscription catch-up tokens to prevent duplicate catch-up requests. (#6955)
Fix unpinning an event not updating the pinned events timeline through pinned_event_listener_task.

Adding a pinned event updated the list, clearing it also did, but removing one or several items from it, as long as it wasn't empty, didn't. (#6991)
Local medias (as in attachments pending in the send queue) are always cached with MediaFormat::File, be it the file itself or its thumbnail, so requesting one with MediaFormat::Thumbnail was failing since the format started being taken into account. (#6996)
Restore checksums for UniFFI functions. They were removed to work around an issue where the checksums were incorrectly computed first for some 32bit ARM devices and then for some 64bit ones. That issue is now fixed so we should restore the checksums. (#7046)

Updates `rust dependencies`. **Breaking change detected.** Review the release notes before merging. Automerge: Disabled because breaking change release notes were detected ### Updates | Dependency | From | To | Type | Locations | | --- | --- | --- | --- | --- | | <code>matrix-sdk</code> | <code>0.19.0</code> | <code>0.19.1</code> | patch; **breaking** | [1] L24 | | <code>matrix-sdk-sqlite</code> | <code>0.19.0</code> | <code>0.19.1</code> | patch; **breaking** | [1] L95 | ### Files - [1] <code>Cargo.toml</code> ### Release notes <details> <summary>matrix-sdk (0.19.0 → 0.19.1)</summary> Source comparison unavailable: source refs for this transition are not known. ### [`matrix-sdk-0.19.0`](https://github.com/matrix-org/matrix-rust-sdk/releases/tag/matrix-sdk-0.19.0): matrix-sdk 0.19.0 **Breaking change detected.** [Source](https://github.com/matrix-org/matrix-rust-sdk) Source ref: <code>main</code> # What's Changed ## Removed - `EventCache::save_events` has been removed. This API was ideal to create deadlocks in the Event Cache if not used with special care. Moreover, it breaks the isolation between components and was violating the clear separation of responsibility between components. The implications are the following: `Room::messages`, `Room::event`, `Room::event_with_context` and `Room::relations` no longer save fetched events in the Event Cache. ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) ### Added - The `ThreadEventCache` has a persistent storage! All events received from the server can now be found locally without reaching the server anymore, which means: thread offline support! Moreover, `ThreadEventCache` contains more events, like edits and redactions! Previously, these events were located in the `RoomEventCache` for technical reasons. This is now solved and works as expected. ([#6517](https://github.com/matrix-org/matrix-rust-sdk/pull/6517)) - The pinned-events cache, aka `PinnedEventsCache`, now understands annotations, aka reactions! All annotations to a pinned-event are now part of the pinned-events cache, making it possible to display the reactions live. At the same time, pinned-events are now deduplicated, along with all the related events, like edits, annotations etc. ([#6568](https://github.com/matrix-org/matrix-rust-sdk/pull/6568)) - Add `Encryption::dehydrated_devices()`, a high-level [MSC3814] dehydrated-device manager that wraps the crypto-crate primitives: probe support, create, rehydrate, delete, weekly rotation, a Secret Storage round trip for the pickle key, and a lifecycle state stream. `MatrixMockServer` gains `mock_*` helpers for the four MSC3814 endpoints. ([#6606](https://github.com/matrix-org/matrix-rust-sdk/pull/6606)) - Added `Account::set_status`, `Account::clear_status`, `Account::set_call`, and `Account::clear_call` to write [MSC4426] user-status profile fields (`m.status` and `m.call`). All four are gated behind the new `unstable-msc4426` feature. ([#6616](https://github.com/matrix-org/matrix-rust-sdk/pull/6616)) - Create a `MediaFetcher` trait that allows us to replace the existing `DefaultMediaFetcher` with any other implementation when building a `Client`. ([#6625](https://github.com/matrix-org/matrix-rust-sdk/pull/6625)) - R2D2 (aka `event_cache::Redecryptor`) no longer decrypts an already decrypted event. It was taking its job too seriously. Can't blame it, but it was creating unnecessary update in the Event Cache and to all the listeners (Latest Event, Search, Timeline etc.). ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) - Global message search results are now sorted by full-text search relevance score, interleaving results from different rooms by relevance rather than grouping them per room. ([#6645](https://github.com/matrix-org/matrix-rust-sdk/pull/6645)) - Add `SlidingSync::set_room_subscriptions` for updating room subscriptions non-destructively. This is similar to `SlidingSync::reset_and_add_room_subscriptions` but doesn't clear and then recreate all subscriptions. Instead, it removes existing subscriptions for rooms that are not resubscribed and adds new subscriptions for resubscribed rooms that don't already exist. Unlike SlidingSync::reset_and_add_room_subscriptions`, this method does not mark members as unsynced.([#6652](https://github.com/matrix-org/matrix-rust-sdk/pull/6652)) - Add client-owned presence configuration that can optionally send an immediate presence update. Generated sync requests use the client presence by default, which starts as unavailable, while classic sync settings can still override it per request. ([#6672](https://github.com/matrix-org/matrix-rust-sdk/pull/6672)) - `ThreadEventCache` auto-shrinks when no more subscribers are alive. Subscribers are returned by [`ThreadEventCache::subscribe`]. When the last subscriber is dropped, it means no external API are listening to this cache, and thus it shrinks to its last chunk of events, to save memory. This behavior was already present for `RoomEventCache`. ([#6690](https://github.com/matrix-org/matrix-rust-sdk/pull/6690)) - Added a new top-level `Error` variant, `SignatureError` representing errors that occurred signing or verifying cryptographic data. ([#6715](https://github.com/matrix-org/matrix-rust-sdk/pull/6715)) - `ClientBuilder::with_x509_signer` and `ClientBuilder::with_x509_verifier` now allow configuring an X.509 signer and verifier for experimental X.509-based cross-signing identity verification. Gated behind the `experimental-x509-identity-verification` feature. ([#6727](https://github.com/matrix-org/matrix-rust-sdk/pull/6727)) - The new `EventCache::forget_room` method properly forget a specific room. It is different from `EventCache::clear_all_events` where (i) it's for a particular room instead of all rooms, (ii) it forgets the room, i.e. it's deleted from the Event Cache entirely instead of being cleared only. If you want to forget the room client-wise, look at `Client::forget_room` (which calls `EventCache::forget_room` at the time of writing). ([#6749](https://github.com/matrix-org/matrix-rust-sdk/pull/6749)) - Add `Client::subscribe_to_own_profile`, returning a stream of the current user's global profile. It emits the currently stored value and re-emits whenever the profile changes during sync, or an empty profile when none is stored. **Note:** Without the Profiles sliding sync extension enabled only an empty profile will be emitted and no updates will be published. ([#6760](https://github.com/matrix-org/matrix-rust-sdk/pull/6760)) - Add `Client::rtc_transports` that queries the RTC transport advertised by the homeserver using the new authenticated `GET /_matrix/client/v1/rtc/transports` API. The value is cached for 1 day, and it is possible to force refresh with `Client::refresh_rtc_transports`, or to by-pass cache with `Client::fetch_rtc_transports`. [**breaking**] `Client::rtc_foci` has been deprecated and should only be used as a fallback for homeservers that do not support the new API (see `Client::well_known_rtc_transports`). ([#6789](https://github.com/matrix-org/matrix-rust-sdk/pull/6789)) - Add support for [MSC4515](https://github.com/matrix-org/matrix-spec-proposals/pull/4515) and RTC transport discovery for widgets. ([#6791](https://github.com/matrix-org/matrix-rust-sdk/pull/6791)) - `AttachmentConfig` gained an `extra_content` field and `RoomSendQueue::send()` now returns a builder-style future with `with_extra_content()`, allowing additional top-level fields to be included in an event's content — e.g. vendor-prefixed keys that server-side push rules can match on (in lieu of [MSC3664](https://github.com/matrix-org/matrix-spec-proposals/pull/3664)). Extra fields never override the fields of the event itself. Supported by the send queue and the direct `Room::send_attachment` path. ([#6812](https://github.com/matrix-org/matrix-rust-sdk/pull/6812)) - Added `Client::enable_automatic_call_status(bool)` an opt-in auto-sync that mirrors this device's MatrixRTC participation into the [MSC4426] `m.call` profile field. Gated behind the `unstable-msc4426` feature. Also adds `Room::is_device_in_active_room_call(user_id, device_id)` for device-scoped participation checks. ([#6825](https://github.com/matrix-org/matrix-rust-sdk/pull/6825)) - Propagate the current tracing `Span` into the E2EE setup task. ([#6833](https://github.com/matrix-org/matrix-rust-sdk/pull/6833)) - The event cache gained a shared `BackPaginationQueue` (`EventCache::back_pagination_queue`): a single background executor that runs back-pagination requests from every use case by `Priority`, with a bounded in-flight number (`EventCacheConfig::max_concurrent_back_paginations`) and one run per room at a time. Enable it with `ClientBuilder::with_enable_automatic_back_pagination(true)`. ([#6838](https://github.com/matrix-org/matrix-rust-sdk/pull/6838)) - Add `Client::disable_well_known_lookup`, which disables every `/.well-known/matrix/client` request performed by the client, for users that must not emit any request to the well-known URI of their domain. When disabled, `Client::tile_server` returns `None`, `Client::well_known_rtc_transports` returns an empty list, and `Client::discover_rtc_transports` doesn't fall back to the well-known foci. Add `ClientBuilder::disable_well_known_lookup`, which controls the initial value of the `Client::disable_well_known_lookup` flag and additionally disables any homeserver discovery performed by `ClientBuilder::build`. If `true`, the homeserver must then be given with `ClientBuilder::homeserver_url`. `ClientBuilder::server_name` and `ClientBuilder::username` can only be resolved through the well-known, so `ClientBuilder::build` fails with the new `ClientBuildError::WellKnownLookupDisabled` variant if they are called. Add `Client::discover_rtc_transports`, which returns the MatrixRTC transports advertised by the homeserver (MSC4519), with a fallback to the well-known `m.rtc_foci` as long as well-known lookup is allowed by the client. Transports are exposed as the new `RtcTransport` enum, with a `LiveKit` variant and a `Custom` variant carrying the raw JSON for transport types the SDK doesn't know about. ([#6845](https://github.com/matrix-org/matrix-rust-sdk/pull/6845)) - Added the `msc_4388_rendezvous_server_supported()` method to check if the homeserver supports MSC4388 for the QR code login. ([#6871](https://github.com/matrix-org/matrix-rust-sdk/pull/6871)) - `ThreadEventCache` now applies new redactions and saves bundled latest thread events for events coming from the pagination, or that have been redecrypted (by the `Redecryptor`). ([#6888](https://github.com/matrix-org/matrix-rust-sdk/pull/6888)) - `ThreadEventCache` gains 4 new methods: 1. `read_receipts` to get the `ReadReceipts` associated to the `ThreadInfo`, 2. `num_unread_messages` to get the number of unread messages (alias of `read_receipts().await?.num_unread`), 3. `num_unread_notifications` to get the number of unread notifications (alias of `read_receipts().await?.num_notifications`), 4. `num_unread_mentions` to get the number of unread mentions (alias of `read_receipts().await?.num_mentions`). ([#6893](https://github.com/matrix-org/matrix-rust-sdk/pull/6893)) - Added `Room::retention()` to retrieve the current retention policy and `RoomPrivacySettings::update_room_retention()` to set new values. Leverages Ruma's `unstable-msc1763` feature flag. ([#6905](https://github.com/matrix-org/matrix-rust-sdk/pull/6905)) - Added `Client::get_retention_configuration` to retrieve the server level retention policy and per room overrides so a client can determine whether the server enforces any retention policy constraints. ([#6908](https://github.com/matrix-org/matrix-rust-sdk/pull/6908)) - `Media::get_media_preview()` returns the OpenGraph-like data a homeserver generates for a URL. It uses the authenticated `/_matrix/client/v1/media/preview_url` endpoint when the homeserver supports it, and falls back to the deprecated `/_matrix/media/v3/preview_url` otherwise. ([#6918](https://github.com/matrix-org/matrix-rust-sdk/pull/6918)) - Added `Room::effective_retention()` to return the final retention policy for a room combining server-wide policy with room-specific. Clients can use this value to determine message retention. ([#6919](https://github.com/matrix-org/matrix-rust-sdk/pull/6919)) - Add `Room::human_member_ids` and `Room::human_member_ids_no_sync`, which return the user IDs of the room members with the given memberships, without the service members declared by the `io.element.functional_members` state event. This is a convenient way to find the other party of a direct message. ([#6925](https://github.com/matrix-org/matrix-rust-sdk/pull/6925)) - `SendHandle::abort_with_reason()` aborts a queued send like `SendHandle::abort()`, and attaches a reason to the redaction that materializes the abort when the event had already been sent by the time the abort was processed. The reason is persisted with the pending redaction, so it survives restarts like the rest of the send queue. ([#6931](https://github.com/matrix-org/matrix-rust-sdk/pull/6931)) - `ThreadEventCache` gains a new method: `subscribe_to_thread_info` to subscribe to updates of the `ThreadInfo`. This is very similar to `Room::subscribe_info`. ([#6993](https://github.com/matrix-org/matrix-rust-sdk/pull/6993)) - Add `Client::total_unread_notifications`, returning the sum of the client-side computed unread notification counts across all joined rooms, counting rooms marked as unread by hand as one each. ([#7002](https://github.com/matrix-org/matrix-rust-sdk/pull/7002)) ### Changed - `RoomEventCacheUpdate::AddEphemeralEvents` becomes `RoomEventCacheUpdate::AddReadReceipt`. It no longer contains an `events: Vec<Raw<AnySyncEphemeralRoomEvent>>` but a single `event: ReceiptEventContent`. In the same vein, `ThreadEventCacheUpdate` is introduced! Thus, `ThreadEventCache::subscribe` no longer returns a `Subscriber<TimelineVectorDiffs>` but a `Subscriber<ThreadEventCacheUpdate>`. The `TimelineVectorDiffs` can be found in `ThreadEventCacheUpdate::UpdateTimelineEvents`, just like its sibling type `RoomEventCacheUpdate`. ([#6038](https://github.com/matrix-org/matrix-rust-sdk/pull/6038)) - The thread cache, aka `ThreadEventCache`, has been extracted from the room cache, aka `RoomEventCache`. To get a thread cache, please use the `EventCache::thread()` method. The `RoomEventCache::thread_pagination` and `subscribe_to_thread` methods have been removed. Before: ```rust let room_cache = event_cache.for_room(room_id).await?; let thread_pagination = room_cache.thread_pagination(thread_id).await?; let thread_subscriber = room_cache.subscribe_to_thread(thread_id)?; ``` After: ```rust let thread_cache = event_cache.thread(room_id, thread_id).await?; let thread_pagination = thread_cache.pagination(); // no async, no error! let thread_subscriber = thread_cache.subscribe().await?; ``` Also, the `EventCache::for_room` method has been renamed `room`. ([#6517](https://github.com/matrix-org/matrix-rust-sdk/pull/6517)) - The pinned-events cache, aka `PinnedEventsCache`, has been extracted from the room cache, aka `RoomEventCache`. To get a pinned-events cache, please use the `EventCache::pinned_events()` method. The `RoomEventCache::subscribe_to_pinned_events` method have been removed and replaced by `PinnedEventsCache::subscribe`. Before: ```rust let room_cache = event_cache.room(room_id).await?; let pinned_events_subscriber = room_cache.subscribe_to_pinned_events()?; ``` After: ```rust let pinned_events_cache = event_cache.pinned_events(room_id).await?; let pinned_events_subscriber = pinned_events_cache.subscribe().await?; ``` ([#6568](https://github.com/matrix-org/matrix-rust-sdk/pull/6568)) - The event-focused cache, aka `EventFocusedCache`, has been extracted from the room cache, aka `RoomEventCache`. To get an event-focused cache, please use the `EventCache::event_focused()` method. The `RoomEventCache::get_or_create_event_focused_cache` and `get_event_focused_cache` methods have been removed. Before: ```rust let room_cache = event_cache.for_room(room_id).await?; let event_focused_cache = room_cache.get_or_create_event_focused_cache(thread_mode, number_of_initial_events).await?; ``` After: ```rust let event_focused_cache = event_cache.event_focused(room_id, event_id, thread_mode, number_of_initial_events).await?; ``` ([#6603](https://github.com/matrix-org/matrix-rust-sdk/pull/6603)) - The `EventCacheStore::clear_all_linked_chunks` trait method has been renamed `clear_all_events` to express its intent better. ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) - [**breaking**] The message search helpers now expose their results as `Stream`s of pages instead of bespoke iterators. `Room::search_messages` returns a stream of pages of `(score, event_id)` pairs (use the new `Room::search_messages_events` for full `TimelineEvent`s), and `GlobalSearchBuilder::build`/`build_events` replace the removed `RoomSearchIterator`/`GlobalSearchIterator` types. ([#6645](https://github.com/matrix-org/matrix-rust-sdk/pull/6645)) - Put `ClientInner::MediaFetcher` behind a `RwLock` so it can be replaced at any time from any thread, without having a mutable `ClientInner` reference around. ([#6689](https://github.com/matrix-org/matrix-rust-sdk/pull/6689)) - The `event_cache::RoomEventCacheSubscriber` type has been renamed to `event_cache::Subscriber<RoomEventCacheUpdate>`. The type has been made generic to be reused by `ThreadEventCache`. The behaviour is exactly the same as it keeps `Deref` and `DerefMut` to `tokio::sync::mpsc::Receiver`. ([#6690](https://github.com/matrix-org/matrix-rust-sdk/pull/6690)) - Index media, stickers, polls and more message types. Move message parsing responsibility outside the search crate and into the main SDK one. ([#6710](https://github.com/matrix-org/matrix-rust-sdk/pull/6710)) - [**breaking**] `GrantLoginProgress::WaitingForAuth` now has a `continuation_sender: ContinuationMessageSender` field. Applications must call `continuation_sender.confirm()` once the verification URI has been opened in the browser and the application is ready to proceed, or `continuation_sender.cancel()` to abort; previously it proceeded automatically as soon as this state was reached. This lets applications that suspend or navigate away while the verification URI is open to resume the process explicitly. [**breaking**] `qrcode::CheckCodeSenderError` has been renamed to `qrcode::SenderError`, since it is now shared with `ContinuationMessageSender`. ([#6711](https://github.com/matrix-org/matrix-rust-sdk/pull/6711)) - [**breaking**] `RoomPreview::heroes` is now `Option<Vec<RoomHeroWithProfile>>` instead of `Option<Vec<RoomHero>>`, exposing each hero's status and call fields from their global profile. These fields are only populated when syncing via sliding sync with the profiles extension enabled. ([#6733](https://github.com/matrix-org/matrix-rust-sdk/pull/6733)) - Methods that delete key backup versions, such as resetting your cryptographic identity now iteratively delete all server-side key backup versions. ([#6744](https://github.com/matrix-org/matrix-rust-sdk/pull/6744)) - [**breaking**] `QRCodeGrantLoginError::UnexpectedMessage`'s `received` field type and `ClientBuildError::AutoDiscovery`'s inner type are now `Box`ed to reduce the error enums' sizes. ([#6763](https://github.com/matrix-org/matrix-rust-sdk/pull/6763)) - To-device messages forwarded to widgets now include the MSC3819 `encrypted`boolean in their `data`, indicating whether the original to-device message was received encrypted. ([MSC3819](https://github.com/matrix-org/matrix-spec-proposals/pull/3819)) ([#6795](https://github.com/matrix-org/matrix-rust-sdk/pull/6795)) - [**breaking**] The check code in the QR code auth is now represented as an `u8` directly instead of the `CheckCode` struct. ([#6819](https://github.com/matrix-org/matrix-rust-sdk/pull/6819)) - Subscribing to an already subscribed room now refreshes its settings, e.g. its `timeline_limit`, instead of silently keeping the previous ones. ([#6828](https://github.com/matrix-org/matrix-rust-sdk/pull/6828)) - [**breaking**] Background back-pagination is now driven by the shared `BackPaginationQueue` instead of a per-room credit system. ([#6838](https://github.com/matrix-org/matrix-rust-sdk/pull/6838)) - The latest event of a room is now computed with automatic back-pagination: when no suitable candidate is in memory, the room's history is back-paginated until a candidate is found or the start of the timeline is reached. Requires `ClientBuilder::with_enable_automatic_back_pagination(true)`. ([#6840](https://github.com/matrix-org/matrix-rust-sdk/pull/6840)) - The send queue now sends requests strictly in the order they were queued, in a given room: a request that failed with an unrecoverable error (and got marked as wedged) blocks all the subsequent requests in the same room from being sent, until it's either retried with `SendHandle::unwedge()` or removed with `SendHandle::abort()`. Previously, a wedged request was skipped over, causing later messages to be sent before it, i.e. out of order. - A failed media upload's send error is now reflected on the media event's local echo returned by `RoomSendQueue::subscribe()`, instead of the echo claiming the media is still being sent. - An HTTP 520 status code (a non-standard "unknown server error", returned by Cloudflare and other reverse proxies) is now considered a transient error again, like the other 5xx status codes: requests failing with it will be retried, and the send queue treats it as a recoverable error (keeping the failed request in the queue) instead of wedging the request. ([#6843](https://github.com/matrix-org/matrix-rust-sdk/pull/6843)) - Widget's `get_rtc_transports` action is now using the `Client::discover_rtc_transports` API that checks the new authenticated `GET /_matrix/client/v1/rtc/transports` API, and falls back to the `.well-known/matrix/client` lookup if allowed by the client. Previously it was never falling back to the well-known lookup. ([#6845](https://github.com/matrix-org/matrix-rust-sdk/pull/6845)) - The room subscription methods of `SlidingSync` are renamed after what they do to the set of subscriptions: `subscribe_to_rooms` becomes `add_room_subscriptions`, `unsubscribe_to_rooms` becomes `remove_room_subscriptions`, `resubscribe_to_rooms` becomes `set_room_subscriptions`, and `clear_and_subscribe_to_rooms` becomes `reset_and_add_room_subscriptions`. ([#6927](https://github.com/matrix-org/matrix-rust-sdk/pull/6927)) - `Account` automatically updates the local copy of the user's global profile after updating (to act as a local echo). Requires support for global profile syncing. ([#6984](https://github.com/matrix-org/matrix-rust-sdk/pull/6984)) - Fix flakiness in the cross-process store lock tests. Dropping a guard does not release the lock at once: the lease-renewal task publishes the release on its next tick. The tests now acquire with `spin_lock_store`, which retries with a backoff, instead of a single attempt after a fixed sleep. Each run also gets its own temporary directory for the sqlite database, instead of a predictable shared path under the system temporary directory that carried state into the next run. ([#6899](https://github.com/matrix-org/matrix-rust-sdk/pull/6899)) - Add `Client::send_encrypted_to_device` (behind the `experimental-send-custom-to-device` feature) to Olm-encrypt a custom to-device message and send it to a set of recipient devices. ([#6981](https://github.com/matrix-org/matrix-rust-sdk/pulls/6981)) ### Fixed - The `ThreadSummary` of a `ThreadEventCache` is computed by using the `latest_events` internal API, making it understand more events and delivering a more accurate result. This is still not perfect and complete but at least it's one step in the right direction! ([#6517](https://github.com/matrix-org/matrix-rust-sdk/pull/6517)) - Resolve the room alias to a room id before evaluating pre-join info. This resolves the asymmetry when using `join_room_by_id_or_alias`, ensuring possibly crucial invite handling is handled correctly. ([#6618](https://github.com/matrix-org/matrix-rust-sdk/pull/6618)) - The Event Cache has now a unique state for all the caches for all the rooms. It fixes numerous bugs but the most notable one is: when a cache has to be reloaded because its state has been invalidated, **all** the other caches are now reloaded at the same time. Previously, the invalidation was correctly detected, then a single cache was reloaded, and the invalidation was marked as done, thus the other caches couldn't detect it and reload accordingly. It was creating an inconsistency in the data. ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) - Prevent deadlocks when waking up the thread subscription catch-up task. ([#6667](https://github.com/matrix-org/matrix-rust-sdk/pull/6667)) - Ensure a `VectorDiff` is not emitted to subscribers when an event cache is shrunk because no subscribers is listening to it. If there is no subscribers, it's useless to send it anyway, and if in the future, a race happens and a subscriber exists, it must not receive this diff. At the same time, the way the subscribers are tracked and counted have been revisited to use less space in memory and use less atomic operations. ([#6684](https://github.com/matrix-org/matrix-rust-sdk/pull/6684)) - It is now possible to prevent `matrix-sdk` from pulling in `aws-lc-rs` as a transitive dependency by deactivating the newly introduced `rustls-aws-lc-rs` default feature. This is a breaking change for downstream users relying on `matrix-sdk` without default features, who now have to explicitly enable `rustls-aws-lc-rs` or register a rustls provider explicitly. ([#6705](https://github.com/matrix-org/matrix-rust-sdk/pull/6705)) - `CrossSigningResetHandle::auth` now uses the common timeout helper (from `matrix-sdk-base`) instead of `tokio::time` directly. ([#6719](https://github.com/matrix-org/matrix-rust-sdk/pull/6719)) - Export SupportedPathBuilder and SupportedAuthScheme Client::send uses these in it's trait bounds, but because they were private it wasn't possible to write dependent code that is generic over request types. ([#6734](https://github.com/matrix-org/matrix-rust-sdk/pull/6734)) - Replace `tokio::time::sleep` with `matrix_sdk_common::sleep::sleep` to ensure QRCode auth compiles to WebAssembly. ([#6743](https://github.com/matrix-org/matrix-rust-sdk/pull/6743)) - `EventCache::clear_all_rooms` no longer recreates an empty `LinkedChunk` per cache (`RoomEventCache`, `ThreadEventCache` etc.). The `LinkedChunk`'s first chunk being now created lazily, it ensures `clear_all_rooms` can be used to clear _or_ to forget data. ([#6752](https://github.com/matrix-org/matrix-rust-sdk/pull/6752)) - The correct semantics for one-time key counts are now used when sync v2 is being used. This bug could have previously led to no one-time keys being uploaded to the homeserver. ([#6780](https://github.com/matrix-org/matrix-rust-sdk/pull/6780)) - `SlidingSync::set_room_subscriptions` now cancels in-flight requests whenever the set of room subscriptions changes. It used to do so only when a subscription was added and another one was removed in the same call, so a first-time subscription or a singular removal was not sent until the current long poll expired, up to 30s later. ([#6828](https://github.com/matrix-org/matrix-rust-sdk/pull/6828)) - The `LatestEventValue` of a room is no longer left unchanged when it moves between local states, e.g. when a second local event is queued, when the send queue wedges (`LocalIsSending` to `LocalCannotBeSent`), or when a send is retried. These values have no event ID, and were incorrectly considered equal to each other. ([#6836](https://github.com/matrix-org/matrix-rust-sdk/pull/6836)) - Fix homeserver switching during QR code login when using rendezvous servers that advertise a valid `.well-known` configuration. ([#6846](https://github.com/matrix-org/matrix-rust-sdk/pull/6846)) - The window in which an app can be signed out by another process rotating the OAuth 2.0 refresh token has been narrowed. The refresh now fetches the authorization server metadata before it reads the token, so a rotation during that request no longer makes the app send a stale token. Being suspended during the token exchange can still produce a sign-out. ([#6860](https://github.com/matrix-org/matrix-rust-sdk/pull/6860)) - A rare deadlock has been fixed in `event_cache::caches::aggregator::aggregate_timeline_for_threads`. When aggregating the timeline for each thread, we are doing look ups in the room cache, and in the thread caches. Every time we do a look up with `find_event`, we are acquiring a read lock over the state. Because look up in the room cache can happen frequently, we were acquiring a read lock over the state for the room cache for the entire lifetime of the `aggregate_timeline_for_threads` function. However, in some situations, it could lead to a deadlock. Imagine task T1 is running `handle_joined_room_update` (which runs `aggregate_timeline_for_threads`): a read lock over the room state is acquired (L1), and the read lock over a thread state must be acquired for another look up (L2). Before L2 is acquired, task T2 is running a different part of the code, like a pagination, and waits to acquire a write lock (L3). It waits because L1 is still alive and L3 needs an exclusive access. Now, L2 enters the game and cannot be acquired because L3 is waiting. So we have: - L1 is acquired but not released, - L1 is waiting on L3 to release itself, - L2 is waiting on L1, - L3 is waiting on L2. This is a deadlock. The fix consists of acquiring a read lock over (all) the states once, to then project the read lock guard to the room and to each thread. This way, L1 is now a combination of L1 and L3. L2 will always wait on L1 to be released. ([#6869](https://github.com/matrix-org/matrix-rust-sdk/pull/6869)) - `Media::get_media_content` now deterministically returns the content of a local media (i.e. a media sent by the user not yet received from the server). ([#6896](https://github.com/matrix-org/matrix-rust-sdk/pull/6896)) - `SlidingSync::reset_and_add_room_subscriptions` now cancels the in-flight request when it removes the last subscriptions. Called with an empty list of rooms, it used to clear the subscriptions without cancelling the request, so the server kept sending the rooms until the current long poll expired, up to 30s later. ([#6932](https://github.com/matrix-org/matrix-rust-sdk/pull/6932)) - When a sync contained more than one read receipts, only the first one was taken into account. Multiple problems arose from this, e.g. (i) perhaps the first selected receipt is not the correct one (`RoomEventCache` only cares about the `Main` or `Unthreaded` for example, and should ignore the `Thread` ones), (ii) or more simply the other read receipts were totally ignored! This is now fixed: ephemeral events are triaged similarly to timeline events, inside `event_cache::aggregator`. ([#6938](https://github.com/matrix-org/matrix-rust-sdk/pull/6938)) - Deduplicate thread-subscription catch-up tokens to prevent duplicate catch-up requests. ([#6955](https://github.com/matrix-org/matrix-rust-sdk/pull/6955)) - Fix unpinning an event not updating the pinned events timeline through `pinned_event_listener_task`. Adding a pinned event updated the list, clearing it also did, but removing one or several items from it, as long as it wasn't empty, didn't. ([#6991](https://github.com/matrix-org/matrix-rust-sdk/pull/6991)) - Local medias (as in attachments pending in the send queue) are always cached with `MediaFormat::File`, be it the file itself or its thumbnail, so requesting one with `MediaFormat::Thumbnail` was failing since the format started being taken into account. ([#6996](https://github.com/matrix-org/matrix-rust-sdk/pull/6996)) - Restore checksums for UniFFI functions. They were removed to work around an issue where the checksums were incorrectly computed first for some 32bit ARM devices and then for some 64bit ones. That issue is now fixed so we should restore the checksums. ([#7046](https://github.com/matrix-org/matrix-rust-sdk/pull/7046)) </details> <details> <summary>matrix-sdk-sqlite (0.19.0 → 0.19.1)</summary> Source comparison unavailable: source refs for this transition are not known. ### [`matrix-sdk-0.19.0`](https://github.com/matrix-org/matrix-rust-sdk/releases/tag/matrix-sdk-0.19.0): matrix-sdk 0.19.0 **Breaking change detected.** [Source](https://github.com/matrix-org/matrix-rust-sdk) Source ref: <code>main</code> # What's Changed ## Removed - `EventCache::save_events` has been removed. This API was ideal to create deadlocks in the Event Cache if not used with special care. Moreover, it breaks the isolation between components and was violating the clear separation of responsibility between components. The implications are the following: `Room::messages`, `Room::event`, `Room::event_with_context` and `Room::relations` no longer save fetched events in the Event Cache. ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) ### Added - The `ThreadEventCache` has a persistent storage! All events received from the server can now be found locally without reaching the server anymore, which means: thread offline support! Moreover, `ThreadEventCache` contains more events, like edits and redactions! Previously, these events were located in the `RoomEventCache` for technical reasons. This is now solved and works as expected. ([#6517](https://github.com/matrix-org/matrix-rust-sdk/pull/6517)) - The pinned-events cache, aka `PinnedEventsCache`, now understands annotations, aka reactions! All annotations to a pinned-event are now part of the pinned-events cache, making it possible to display the reactions live. At the same time, pinned-events are now deduplicated, along with all the related events, like edits, annotations etc. ([#6568](https://github.com/matrix-org/matrix-rust-sdk/pull/6568)) - Add `Encryption::dehydrated_devices()`, a high-level [MSC3814] dehydrated-device manager that wraps the crypto-crate primitives: probe support, create, rehydrate, delete, weekly rotation, a Secret Storage round trip for the pickle key, and a lifecycle state stream. `MatrixMockServer` gains `mock_*` helpers for the four MSC3814 endpoints. ([#6606](https://github.com/matrix-org/matrix-rust-sdk/pull/6606)) - Added `Account::set_status`, `Account::clear_status`, `Account::set_call`, and `Account::clear_call` to write [MSC4426] user-status profile fields (`m.status` and `m.call`). All four are gated behind the new `unstable-msc4426` feature. ([#6616](https://github.com/matrix-org/matrix-rust-sdk/pull/6616)) - Create a `MediaFetcher` trait that allows us to replace the existing `DefaultMediaFetcher` with any other implementation when building a `Client`. ([#6625](https://github.com/matrix-org/matrix-rust-sdk/pull/6625)) - R2D2 (aka `event_cache::Redecryptor`) no longer decrypts an already decrypted event. It was taking its job too seriously. Can't blame it, but it was creating unnecessary update in the Event Cache and to all the listeners (Latest Event, Search, Timeline etc.). ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) - Global message search results are now sorted by full-text search relevance score, interleaving results from different rooms by relevance rather than grouping them per room. ([#6645](https://github.com/matrix-org/matrix-rust-sdk/pull/6645)) - Add `SlidingSync::set_room_subscriptions` for updating room subscriptions non-destructively. This is similar to `SlidingSync::reset_and_add_room_subscriptions` but doesn't clear and then recreate all subscriptions. Instead, it removes existing subscriptions for rooms that are not resubscribed and adds new subscriptions for resubscribed rooms that don't already exist. Unlike SlidingSync::reset_and_add_room_subscriptions`, this method does not mark members as unsynced.([#6652](https://github.com/matrix-org/matrix-rust-sdk/pull/6652)) - Add client-owned presence configuration that can optionally send an immediate presence update. Generated sync requests use the client presence by default, which starts as unavailable, while classic sync settings can still override it per request. ([#6672](https://github.com/matrix-org/matrix-rust-sdk/pull/6672)) - `ThreadEventCache` auto-shrinks when no more subscribers are alive. Subscribers are returned by [`ThreadEventCache::subscribe`]. When the last subscriber is dropped, it means no external API are listening to this cache, and thus it shrinks to its last chunk of events, to save memory. This behavior was already present for `RoomEventCache`. ([#6690](https://github.com/matrix-org/matrix-rust-sdk/pull/6690)) - Added a new top-level `Error` variant, `SignatureError` representing errors that occurred signing or verifying cryptographic data. ([#6715](https://github.com/matrix-org/matrix-rust-sdk/pull/6715)) - `ClientBuilder::with_x509_signer` and `ClientBuilder::with_x509_verifier` now allow configuring an X.509 signer and verifier for experimental X.509-based cross-signing identity verification. Gated behind the `experimental-x509-identity-verification` feature. ([#6727](https://github.com/matrix-org/matrix-rust-sdk/pull/6727)) - The new `EventCache::forget_room` method properly forget a specific room. It is different from `EventCache::clear_all_events` where (i) it's for a particular room instead of all rooms, (ii) it forgets the room, i.e. it's deleted from the Event Cache entirely instead of being cleared only. If you want to forget the room client-wise, look at `Client::forget_room` (which calls `EventCache::forget_room` at the time of writing). ([#6749](https://github.com/matrix-org/matrix-rust-sdk/pull/6749)) - Add `Client::subscribe_to_own_profile`, returning a stream of the current user's global profile. It emits the currently stored value and re-emits whenever the profile changes during sync, or an empty profile when none is stored. **Note:** Without the Profiles sliding sync extension enabled only an empty profile will be emitted and no updates will be published. ([#6760](https://github.com/matrix-org/matrix-rust-sdk/pull/6760)) - Add `Client::rtc_transports` that queries the RTC transport advertised by the homeserver using the new authenticated `GET /_matrix/client/v1/rtc/transports` API. The value is cached for 1 day, and it is possible to force refresh with `Client::refresh_rtc_transports`, or to by-pass cache with `Client::fetch_rtc_transports`. [**breaking**] `Client::rtc_foci` has been deprecated and should only be used as a fallback for homeservers that do not support the new API (see `Client::well_known_rtc_transports`). ([#6789](https://github.com/matrix-org/matrix-rust-sdk/pull/6789)) - Add support for [MSC4515](https://github.com/matrix-org/matrix-spec-proposals/pull/4515) and RTC transport discovery for widgets. ([#6791](https://github.com/matrix-org/matrix-rust-sdk/pull/6791)) - `AttachmentConfig` gained an `extra_content` field and `RoomSendQueue::send()` now returns a builder-style future with `with_extra_content()`, allowing additional top-level fields to be included in an event's content — e.g. vendor-prefixed keys that server-side push rules can match on (in lieu of [MSC3664](https://github.com/matrix-org/matrix-spec-proposals/pull/3664)). Extra fields never override the fields of the event itself. Supported by the send queue and the direct `Room::send_attachment` path. ([#6812](https://github.com/matrix-org/matrix-rust-sdk/pull/6812)) - Added `Client::enable_automatic_call_status(bool)` an opt-in auto-sync that mirrors this device's MatrixRTC participation into the [MSC4426] `m.call` profile field. Gated behind the `unstable-msc4426` feature. Also adds `Room::is_device_in_active_room_call(user_id, device_id)` for device-scoped participation checks. ([#6825](https://github.com/matrix-org/matrix-rust-sdk/pull/6825)) - Propagate the current tracing `Span` into the E2EE setup task. ([#6833](https://github.com/matrix-org/matrix-rust-sdk/pull/6833)) - The event cache gained a shared `BackPaginationQueue` (`EventCache::back_pagination_queue`): a single background executor that runs back-pagination requests from every use case by `Priority`, with a bounded in-flight number (`EventCacheConfig::max_concurrent_back_paginations`) and one run per room at a time. Enable it with `ClientBuilder::with_enable_automatic_back_pagination(true)`. ([#6838](https://github.com/matrix-org/matrix-rust-sdk/pull/6838)) - Add `Client::disable_well_known_lookup`, which disables every `/.well-known/matrix/client` request performed by the client, for users that must not emit any request to the well-known URI of their domain. When disabled, `Client::tile_server` returns `None`, `Client::well_known_rtc_transports` returns an empty list, and `Client::discover_rtc_transports` doesn't fall back to the well-known foci. Add `ClientBuilder::disable_well_known_lookup`, which controls the initial value of the `Client::disable_well_known_lookup` flag and additionally disables any homeserver discovery performed by `ClientBuilder::build`. If `true`, the homeserver must then be given with `ClientBuilder::homeserver_url`. `ClientBuilder::server_name` and `ClientBuilder::username` can only be resolved through the well-known, so `ClientBuilder::build` fails with the new `ClientBuildError::WellKnownLookupDisabled` variant if they are called. Add `Client::discover_rtc_transports`, which returns the MatrixRTC transports advertised by the homeserver (MSC4519), with a fallback to the well-known `m.rtc_foci` as long as well-known lookup is allowed by the client. Transports are exposed as the new `RtcTransport` enum, with a `LiveKit` variant and a `Custom` variant carrying the raw JSON for transport types the SDK doesn't know about. ([#6845](https://github.com/matrix-org/matrix-rust-sdk/pull/6845)) - Added the `msc_4388_rendezvous_server_supported()` method to check if the homeserver supports MSC4388 for the QR code login. ([#6871](https://github.com/matrix-org/matrix-rust-sdk/pull/6871)) - `ThreadEventCache` now applies new redactions and saves bundled latest thread events for events coming from the pagination, or that have been redecrypted (by the `Redecryptor`). ([#6888](https://github.com/matrix-org/matrix-rust-sdk/pull/6888)) - `ThreadEventCache` gains 4 new methods: 1. `read_receipts` to get the `ReadReceipts` associated to the `ThreadInfo`, 2. `num_unread_messages` to get the number of unread messages (alias of `read_receipts().await?.num_unread`), 3. `num_unread_notifications` to get the number of unread notifications (alias of `read_receipts().await?.num_notifications`), 4. `num_unread_mentions` to get the number of unread mentions (alias of `read_receipts().await?.num_mentions`). ([#6893](https://github.com/matrix-org/matrix-rust-sdk/pull/6893)) - Added `Room::retention()` to retrieve the current retention policy and `RoomPrivacySettings::update_room_retention()` to set new values. Leverages Ruma's `unstable-msc1763` feature flag. ([#6905](https://github.com/matrix-org/matrix-rust-sdk/pull/6905)) - Added `Client::get_retention_configuration` to retrieve the server level retention policy and per room overrides so a client can determine whether the server enforces any retention policy constraints. ([#6908](https://github.com/matrix-org/matrix-rust-sdk/pull/6908)) - `Media::get_media_preview()` returns the OpenGraph-like data a homeserver generates for a URL. It uses the authenticated `/_matrix/client/v1/media/preview_url` endpoint when the homeserver supports it, and falls back to the deprecated `/_matrix/media/v3/preview_url` otherwise. ([#6918](https://github.com/matrix-org/matrix-rust-sdk/pull/6918)) - Added `Room::effective_retention()` to return the final retention policy for a room combining server-wide policy with room-specific. Clients can use this value to determine message retention. ([#6919](https://github.com/matrix-org/matrix-rust-sdk/pull/6919)) - Add `Room::human_member_ids` and `Room::human_member_ids_no_sync`, which return the user IDs of the room members with the given memberships, without the service members declared by the `io.element.functional_members` state event. This is a convenient way to find the other party of a direct message. ([#6925](https://github.com/matrix-org/matrix-rust-sdk/pull/6925)) - `SendHandle::abort_with_reason()` aborts a queued send like `SendHandle::abort()`, and attaches a reason to the redaction that materializes the abort when the event had already been sent by the time the abort was processed. The reason is persisted with the pending redaction, so it survives restarts like the rest of the send queue. ([#6931](https://github.com/matrix-org/matrix-rust-sdk/pull/6931)) - `ThreadEventCache` gains a new method: `subscribe_to_thread_info` to subscribe to updates of the `ThreadInfo`. This is very similar to `Room::subscribe_info`. ([#6993](https://github.com/matrix-org/matrix-rust-sdk/pull/6993)) - Add `Client::total_unread_notifications`, returning the sum of the client-side computed unread notification counts across all joined rooms, counting rooms marked as unread by hand as one each. ([#7002](https://github.com/matrix-org/matrix-rust-sdk/pull/7002)) ### Changed - `RoomEventCacheUpdate::AddEphemeralEvents` becomes `RoomEventCacheUpdate::AddReadReceipt`. It no longer contains an `events: Vec<Raw<AnySyncEphemeralRoomEvent>>` but a single `event: ReceiptEventContent`. In the same vein, `ThreadEventCacheUpdate` is introduced! Thus, `ThreadEventCache::subscribe` no longer returns a `Subscriber<TimelineVectorDiffs>` but a `Subscriber<ThreadEventCacheUpdate>`. The `TimelineVectorDiffs` can be found in `ThreadEventCacheUpdate::UpdateTimelineEvents`, just like its sibling type `RoomEventCacheUpdate`. ([#6038](https://github.com/matrix-org/matrix-rust-sdk/pull/6038)) - The thread cache, aka `ThreadEventCache`, has been extracted from the room cache, aka `RoomEventCache`. To get a thread cache, please use the `EventCache::thread()` method. The `RoomEventCache::thread_pagination` and `subscribe_to_thread` methods have been removed. Before: ```rust let room_cache = event_cache.for_room(room_id).await?; let thread_pagination = room_cache.thread_pagination(thread_id).await?; let thread_subscriber = room_cache.subscribe_to_thread(thread_id)?; ``` After: ```rust let thread_cache = event_cache.thread(room_id, thread_id).await?; let thread_pagination = thread_cache.pagination(); // no async, no error! let thread_subscriber = thread_cache.subscribe().await?; ``` Also, the `EventCache::for_room` method has been renamed `room`. ([#6517](https://github.com/matrix-org/matrix-rust-sdk/pull/6517)) - The pinned-events cache, aka `PinnedEventsCache`, has been extracted from the room cache, aka `RoomEventCache`. To get a pinned-events cache, please use the `EventCache::pinned_events()` method. The `RoomEventCache::subscribe_to_pinned_events` method have been removed and replaced by `PinnedEventsCache::subscribe`. Before: ```rust let room_cache = event_cache.room(room_id).await?; let pinned_events_subscriber = room_cache.subscribe_to_pinned_events()?; ``` After: ```rust let pinned_events_cache = event_cache.pinned_events(room_id).await?; let pinned_events_subscriber = pinned_events_cache.subscribe().await?; ``` ([#6568](https://github.com/matrix-org/matrix-rust-sdk/pull/6568)) - The event-focused cache, aka `EventFocusedCache`, has been extracted from the room cache, aka `RoomEventCache`. To get an event-focused cache, please use the `EventCache::event_focused()` method. The `RoomEventCache::get_or_create_event_focused_cache` and `get_event_focused_cache` methods have been removed. Before: ```rust let room_cache = event_cache.for_room(room_id).await?; let event_focused_cache = room_cache.get_or_create_event_focused_cache(thread_mode, number_of_initial_events).await?; ``` After: ```rust let event_focused_cache = event_cache.event_focused(room_id, event_id, thread_mode, number_of_initial_events).await?; ``` ([#6603](https://github.com/matrix-org/matrix-rust-sdk/pull/6603)) - The `EventCacheStore::clear_all_linked_chunks` trait method has been renamed `clear_all_events` to express its intent better. ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) - [**breaking**] The message search helpers now expose their results as `Stream`s of pages instead of bespoke iterators. `Room::search_messages` returns a stream of pages of `(score, event_id)` pairs (use the new `Room::search_messages_events` for full `TimelineEvent`s), and `GlobalSearchBuilder::build`/`build_events` replace the removed `RoomSearchIterator`/`GlobalSearchIterator` types. ([#6645](https://github.com/matrix-org/matrix-rust-sdk/pull/6645)) - Put `ClientInner::MediaFetcher` behind a `RwLock` so it can be replaced at any time from any thread, without having a mutable `ClientInner` reference around. ([#6689](https://github.com/matrix-org/matrix-rust-sdk/pull/6689)) - The `event_cache::RoomEventCacheSubscriber` type has been renamed to `event_cache::Subscriber<RoomEventCacheUpdate>`. The type has been made generic to be reused by `ThreadEventCache`. The behaviour is exactly the same as it keeps `Deref` and `DerefMut` to `tokio::sync::mpsc::Receiver`. ([#6690](https://github.com/matrix-org/matrix-rust-sdk/pull/6690)) - Index media, stickers, polls and more message types. Move message parsing responsibility outside the search crate and into the main SDK one. ([#6710](https://github.com/matrix-org/matrix-rust-sdk/pull/6710)) - [**breaking**] `GrantLoginProgress::WaitingForAuth` now has a `continuation_sender: ContinuationMessageSender` field. Applications must call `continuation_sender.confirm()` once the verification URI has been opened in the browser and the application is ready to proceed, or `continuation_sender.cancel()` to abort; previously it proceeded automatically as soon as this state was reached. This lets applications that suspend or navigate away while the verification URI is open to resume the process explicitly. [**breaking**] `qrcode::CheckCodeSenderError` has been renamed to `qrcode::SenderError`, since it is now shared with `ContinuationMessageSender`. ([#6711](https://github.com/matrix-org/matrix-rust-sdk/pull/6711)) - [**breaking**] `RoomPreview::heroes` is now `Option<Vec<RoomHeroWithProfile>>` instead of `Option<Vec<RoomHero>>`, exposing each hero's status and call fields from their global profile. These fields are only populated when syncing via sliding sync with the profiles extension enabled. ([#6733](https://github.com/matrix-org/matrix-rust-sdk/pull/6733)) - Methods that delete key backup versions, such as resetting your cryptographic identity now iteratively delete all server-side key backup versions. ([#6744](https://github.com/matrix-org/matrix-rust-sdk/pull/6744)) - [**breaking**] `QRCodeGrantLoginError::UnexpectedMessage`'s `received` field type and `ClientBuildError::AutoDiscovery`'s inner type are now `Box`ed to reduce the error enums' sizes. ([#6763](https://github.com/matrix-org/matrix-rust-sdk/pull/6763)) - To-device messages forwarded to widgets now include the MSC3819 `encrypted`boolean in their `data`, indicating whether the original to-device message was received encrypted. ([MSC3819](https://github.com/matrix-org/matrix-spec-proposals/pull/3819)) ([#6795](https://github.com/matrix-org/matrix-rust-sdk/pull/6795)) - [**breaking**] The check code in the QR code auth is now represented as an `u8` directly instead of the `CheckCode` struct. ([#6819](https://github.com/matrix-org/matrix-rust-sdk/pull/6819)) - Subscribing to an already subscribed room now refreshes its settings, e.g. its `timeline_limit`, instead of silently keeping the previous ones. ([#6828](https://github.com/matrix-org/matrix-rust-sdk/pull/6828)) - [**breaking**] Background back-pagination is now driven by the shared `BackPaginationQueue` instead of a per-room credit system. ([#6838](https://github.com/matrix-org/matrix-rust-sdk/pull/6838)) - The latest event of a room is now computed with automatic back-pagination: when no suitable candidate is in memory, the room's history is back-paginated until a candidate is found or the start of the timeline is reached. Requires `ClientBuilder::with_enable_automatic_back_pagination(true)`. ([#6840](https://github.com/matrix-org/matrix-rust-sdk/pull/6840)) - The send queue now sends requests strictly in the order they were queued, in a given room: a request that failed with an unrecoverable error (and got marked as wedged) blocks all the subsequent requests in the same room from being sent, until it's either retried with `SendHandle::unwedge()` or removed with `SendHandle::abort()`. Previously, a wedged request was skipped over, causing later messages to be sent before it, i.e. out of order. - A failed media upload's send error is now reflected on the media event's local echo returned by `RoomSendQueue::subscribe()`, instead of the echo claiming the media is still being sent. - An HTTP 520 status code (a non-standard "unknown server error", returned by Cloudflare and other reverse proxies) is now considered a transient error again, like the other 5xx status codes: requests failing with it will be retried, and the send queue treats it as a recoverable error (keeping the failed request in the queue) instead of wedging the request. ([#6843](https://github.com/matrix-org/matrix-rust-sdk/pull/6843)) - Widget's `get_rtc_transports` action is now using the `Client::discover_rtc_transports` API that checks the new authenticated `GET /_matrix/client/v1/rtc/transports` API, and falls back to the `.well-known/matrix/client` lookup if allowed by the client. Previously it was never falling back to the well-known lookup. ([#6845](https://github.com/matrix-org/matrix-rust-sdk/pull/6845)) - The room subscription methods of `SlidingSync` are renamed after what they do to the set of subscriptions: `subscribe_to_rooms` becomes `add_room_subscriptions`, `unsubscribe_to_rooms` becomes `remove_room_subscriptions`, `resubscribe_to_rooms` becomes `set_room_subscriptions`, and `clear_and_subscribe_to_rooms` becomes `reset_and_add_room_subscriptions`. ([#6927](https://github.com/matrix-org/matrix-rust-sdk/pull/6927)) - `Account` automatically updates the local copy of the user's global profile after updating (to act as a local echo). Requires support for global profile syncing. ([#6984](https://github.com/matrix-org/matrix-rust-sdk/pull/6984)) - Fix flakiness in the cross-process store lock tests. Dropping a guard does not release the lock at once: the lease-renewal task publishes the release on its next tick. The tests now acquire with `spin_lock_store`, which retries with a backoff, instead of a single attempt after a fixed sleep. Each run also gets its own temporary directory for the sqlite database, instead of a predictable shared path under the system temporary directory that carried state into the next run. ([#6899](https://github.com/matrix-org/matrix-rust-sdk/pull/6899)) - Add `Client::send_encrypted_to_device` (behind the `experimental-send-custom-to-device` feature) to Olm-encrypt a custom to-device message and send it to a set of recipient devices. ([#6981](https://github.com/matrix-org/matrix-rust-sdk/pulls/6981)) ### Fixed - The `ThreadSummary` of a `ThreadEventCache` is computed by using the `latest_events` internal API, making it understand more events and delivering a more accurate result. This is still not perfect and complete but at least it's one step in the right direction! ([#6517](https://github.com/matrix-org/matrix-rust-sdk/pull/6517)) - Resolve the room alias to a room id before evaluating pre-join info. This resolves the asymmetry when using `join_room_by_id_or_alias`, ensuring possibly crucial invite handling is handled correctly. ([#6618](https://github.com/matrix-org/matrix-rust-sdk/pull/6618)) - The Event Cache has now a unique state for all the caches for all the rooms. It fixes numerous bugs but the most notable one is: when a cache has to be reloaded because its state has been invalidated, **all** the other caches are now reloaded at the same time. Previously, the invalidation was correctly detected, then a single cache was reloaded, and the invalidation was marked as done, thus the other caches couldn't detect it and reload accordingly. It was creating an inconsistency in the data. ([#6629](https://github.com/matrix-org/matrix-rust-sdk/pull/6629)) - Prevent deadlocks when waking up the thread subscription catch-up task. ([#6667](https://github.com/matrix-org/matrix-rust-sdk/pull/6667)) - Ensure a `VectorDiff` is not emitted to subscribers when an event cache is shrunk because no subscribers is listening to it. If there is no subscribers, it's useless to send it anyway, and if in the future, a race happens and a subscriber exists, it must not receive this diff. At the same time, the way the subscribers are tracked and counted have been revisited to use less space in memory and use less atomic operations. ([#6684](https://github.com/matrix-org/matrix-rust-sdk/pull/6684)) - It is now possible to prevent `matrix-sdk` from pulling in `aws-lc-rs` as a transitive dependency by deactivating the newly introduced `rustls-aws-lc-rs` default feature. This is a breaking change for downstream users relying on `matrix-sdk` without default features, who now have to explicitly enable `rustls-aws-lc-rs` or register a rustls provider explicitly. ([#6705](https://github.com/matrix-org/matrix-rust-sdk/pull/6705)) - `CrossSigningResetHandle::auth` now uses the common timeout helper (from `matrix-sdk-base`) instead of `tokio::time` directly. ([#6719](https://github.com/matrix-org/matrix-rust-sdk/pull/6719)) - Export SupportedPathBuilder and SupportedAuthScheme Client::send uses these in it's trait bounds, but because they were private it wasn't possible to write dependent code that is generic over request types. ([#6734](https://github.com/matrix-org/matrix-rust-sdk/pull/6734)) - Replace `tokio::time::sleep` with `matrix_sdk_common::sleep::sleep` to ensure QRCode auth compiles to WebAssembly. ([#6743](https://github.com/matrix-org/matrix-rust-sdk/pull/6743)) - `EventCache::clear_all_rooms` no longer recreates an empty `LinkedChunk` per cache (`RoomEventCache`, `ThreadEventCache` etc.). The `LinkedChunk`'s first chunk being now created lazily, it ensures `clear_all_rooms` can be used to clear _or_ to forget data. ([#6752](https://github.com/matrix-org/matrix-rust-sdk/pull/6752)) - The correct semantics for one-time key counts are now used when sync v2 is being used. This bug could have previously led to no one-time keys being uploaded to the homeserver. ([#6780](https://github.com/matrix-org/matrix-rust-sdk/pull/6780)) - `SlidingSync::set_room_subscriptions` now cancels in-flight requests whenever the set of room subscriptions changes. It used to do so only when a subscription was added and another one was removed in the same call, so a first-time subscription or a singular removal was not sent until the current long poll expired, up to 30s later. ([#6828](https://github.com/matrix-org/matrix-rust-sdk/pull/6828)) - The `LatestEventValue` of a room is no longer left unchanged when it moves between local states, e.g. when a second local event is queued, when the send queue wedges (`LocalIsSending` to `LocalCannotBeSent`), or when a send is retried. These values have no event ID, and were incorrectly considered equal to each other. ([#6836](https://github.com/matrix-org/matrix-rust-sdk/pull/6836)) - Fix homeserver switching during QR code login when using rendezvous servers that advertise a valid `.well-known` configuration. ([#6846](https://github.com/matrix-org/matrix-rust-sdk/pull/6846)) - The window in which an app can be signed out by another process rotating the OAuth 2.0 refresh token has been narrowed. The refresh now fetches the authorization server metadata before it reads the token, so a rotation during that request no longer makes the app send a stale token. Being suspended during the token exchange can still produce a sign-out. ([#6860](https://github.com/matrix-org/matrix-rust-sdk/pull/6860)) - A rare deadlock has been fixed in `event_cache::caches::aggregator::aggregate_timeline_for_threads`. When aggregating the timeline for each thread, we are doing look ups in the room cache, and in the thread caches. Every time we do a look up with `find_event`, we are acquiring a read lock over the state. Because look up in the room cache can happen frequently, we were acquiring a read lock over the state for the room cache for the entire lifetime of the `aggregate_timeline_for_threads` function. However, in some situations, it could lead to a deadlock. Imagine task T1 is running `handle_joined_room_update` (which runs `aggregate_timeline_for_threads`): a read lock over the room state is acquired (L1), and the read lock over a thread state must be acquired for another look up (L2). Before L2 is acquired, task T2 is running a different part of the code, like a pagination, and waits to acquire a write lock (L3). It waits because L1 is still alive and L3 needs an exclusive access. Now, L2 enters the game and cannot be acquired because L3 is waiting. So we have: - L1 is acquired but not released, - L1 is waiting on L3 to release itself, - L2 is waiting on L1, - L3 is waiting on L2. This is a deadlock. The fix consists of acquiring a read lock over (all) the states once, to then project the read lock guard to the room and to each thread. This way, L1 is now a combination of L1 and L3. L2 will always wait on L1 to be released. ([#6869](https://github.com/matrix-org/matrix-rust-sdk/pull/6869)) - `Media::get_media_content` now deterministically returns the content of a local media (i.e. a media sent by the user not yet received from the server). ([#6896](https://github.com/matrix-org/matrix-rust-sdk/pull/6896)) - `SlidingSync::reset_and_add_room_subscriptions` now cancels the in-flight request when it removes the last subscriptions. Called with an empty list of rooms, it used to clear the subscriptions without cancelling the request, so the server kept sending the rooms until the current long poll expired, up to 30s later. ([#6932](https://github.com/matrix-org/matrix-rust-sdk/pull/6932)) - When a sync contained more than one read receipts, only the first one was taken into account. Multiple problems arose from this, e.g. (i) perhaps the first selected receipt is not the correct one (`RoomEventCache` only cares about the `Main` or `Unthreaded` for example, and should ignore the `Thread` ones), (ii) or more simply the other read receipts were totally ignored! This is now fixed: ephemeral events are triaged similarly to timeline events, inside `event_cache::aggregator`. ([#6938](https://github.com/matrix-org/matrix-rust-sdk/pull/6938)) - Deduplicate thread-subscription catch-up tokens to prevent duplicate catch-up requests. ([#6955](https://github.com/matrix-org/matrix-rust-sdk/pull/6955)) - Fix unpinning an event not updating the pinned events timeline through `pinned_event_listener_task`. Adding a pinned event updated the list, clearing it also did, but removing one or several items from it, as long as it wasn't empty, didn't. ([#6991](https://github.com/matrix-org/matrix-rust-sdk/pull/6991)) - Local medias (as in attachments pending in the send queue) are always cached with `MediaFormat::File`, be it the file itself or its thumbnail, so requesting one with `MediaFormat::Thumbnail` was failing since the format started being taken into account. ([#6996](https://github.com/matrix-org/matrix-rust-sdk/pull/6996)) - Restore checksums for UniFFI functions. They were removed to work around an issue where the checksums were incorrectly computed first for some 32bit ARM devices and then for some 64bit ones. That issue is now fixed so we should restore the checksums. ([#7046](https://github.com/matrix-org/matrix-rust-sdk/pull/7046)) </details>

Wuast94 added 1 commit

2026-09-19 03:50:26 +02:00

[breaking] chore: update rust dependencies (0.19.0 → 0.19.1)

CI / Format (pull_request) Successful in 7m48s

Details

CI / Dependency Audit (pull_request) Successful in 11m43s

Details

CI / Clippy (pull_request) Successful in 14m59s

Details

CI / Logic Tests (pull_request) Successful in 18m39s

Details

CI / Production Build (pull_request) Successful in 24m24s

Details

17a0a97095

Wuast94 changed title from ~~[breaking] chore: update rust dependencies (0.19.0 → 0.19.1)~~ to [breaking] chore: update rust dependencies (2 updates)

2026-09-19 09:11:41 +02:00

Wuast94 force-pushed updatarr/rust-dependencies-17605dfeaf from 17a0a97095

CI / Format (pull_request) Successful in 7m48s

Details

CI / Dependency Audit (pull_request) Successful in 11m43s

Details

CI / Clippy (pull_request) Successful in 14m59s

Details

CI / Logic Tests (pull_request) Successful in 18m39s

Details

CI / Production Build (pull_request) Successful in 24m24s

Details

to 7e6c498b11

CI / Format (pull_request) Successful in 9m38s

Details

CI / Dependency Audit (pull_request) Successful in 11m52s

Details

CI / Clippy (pull_request) Successful in 16m47s

Details

CI / Logic Tests (pull_request) Successful in 22m5s

Details

CI / Production Build (pull_request) Successful in 32m4s

Details

2026-09-19 09:11:43 +02:00

Compare

Wuast94 force-pushed updatarr/rust-dependencies-17605dfeaf from 7e6c498b11

CI / Format (pull_request) Successful in 9m38s

Details

CI / Dependency Audit (pull_request) Successful in 11m52s

Details

CI / Clippy (pull_request) Successful in 16m47s

Details

CI / Logic Tests (pull_request) Successful in 22m5s

Details

CI / Production Build (pull_request) Successful in 32m4s

Details

to 8928665d05

CI / Dependency Audit (pull_request) Successful in 56s

Details

CI / Logic Tests (pull_request) Successful in 1m7s

Details

CI / Clippy (pull_request) Successful in 1m8s

Details

CI / Format (pull_request) Successful in 1m17s

Details

CI / Production Build (pull_request) Successful in 26m30s

Details

2026-09-20 01:54:14 +02:00

Compare

Wuast94 referenced this pull request from a commit

2026-09-20 16:33:00 +02:00

chore: update rust dependencies and lock file maintenance (#310, #312)

Wuast94 closed this pull request

2026-09-20 16:33:16 +02:00

CI / Dependency Audit (pull_request) Successful in 56s

Details

CI / Logic Tests (pull_request) Successful in 1m7s

Details

CI / Clippy (pull_request) Successful in 1m8s

Details

CI / Format (pull_request) Successful in 1m17s

Details

CI / Production Build (pull_request) Successful in 26m30s

Details

CI / Quality (pull_request)

Required

CI / Build production (pull_request)

Required

Pull request closed

This pull request cannot be reopened because the branch was deleted.

No reviewers

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

vault/vrds!310

No description provided.

Rows
Columns

[breaking] chore: update rust dependencies (2 updates) #310

Updates

Files

Release notes

matrix-sdk-0.19.0: matrix-sdk 0.19.0

What's Changed

Removed

Added

Changed

Fixed

matrix-sdk-0.19.0: matrix-sdk 0.19.0

What's Changed

Removed

Added

Changed

Fixed

Pull request closed

`matrix-sdk-0.19.0`: matrix-sdk 0.19.0

`matrix-sdk-0.19.0`: matrix-sdk 0.19.0