* Remove m.thread filter from relations API call
We used MSC3981 to pass the recurse param to the /relations
endpoint so that we could get relations to events in a thread, but
we kept the rel_type filter on (as m.thread) so no second-order relations
would ever have been returned (a nested thread isn't a thing).
This removes the filter and does some filtering on the client side to
remove any events that shouldn't live in the threaded timeline (ie.
non-thread relations to the thread root event).
This should help fix stuck unreads because it will avoid the event that
the receipt refers to going missing (but only on HSes that support MSC3981).
For https://github.com/vector-im/element-web/issues/26718
* Fix import cycle
* Remove params from expected calls in tests to match
* Unused import
* Comment explaining the purpose of RoomList
* Fix incorrect return type declaration on RoomList.getRoomEncryption
* Move RoomList out of MatrixClient, into legacy Crypto
* Initialise RoomList inside Crypto.init to allow us to await it
* initial commit
* new interation test
* more comments
* fix test, quick refactor on request version
* cleaning and logs
* fix type
* cleaning
* remove delegate stuff
* remove events and use timer mocks
* fix import
* ts ignore in tests
* Quick cleaning
* code review
* Use Errors instead of Results
* cleaning
* review
* remove forceCheck as not useful
* bad naming
* inline pauseLoop
* mark as paused in finally
* code review
* post merge fix
* rename KeyDownloadRateLimit
* use same config in loop and pass along
A method that we use for fetching recursive related events on homeservers
without MSC3981 support injects events into the timeline in timestamp
order using a special method on event-timeline-set. Injecting events using
this method could cause on-screen notifications because it incorrectly set
the 'liveEvent' flag to true if the events were added tio the live timeline.
These events are never live though as the point is that we're fetching them.
* don't back up keys that we got from backup
* lint
* lint again
* remove key source struct and add function for importing from backup
* apply changes from review
---------
Co-authored-by: Richard van der Hoff <1389908+richvdh@users.noreply.github.com>
* Test whether an event not in a room's timeline causes notification count increase
Commited separately to demonstrate test failing before.
* Don't fix up notification counts if event isn't in the room
As explained by the comment, hopefully.
* Fix other test
* Rust-crypto: fix `bootstrapCrossSigning` on second call
Currently, `bootstrapCrossSigning` raises an exception if it is called a second
time before secret storage is set up. It is easily fixed by checking that 4S is
set up before trying to export to 4S.
Also a few logging fixes while we're in the area.
* Factor out an `AccountDataAccumulator`
* Another test for bootstrapCrossSigning
* Bump `matrix-rust-sdk-crypto-wasm` to v3.2.0
* Reinstate timeout on `getUserDevices` call
Turns out that this used to have a timeout of 1 second in the wasm
bindings, which it no longer does. Reinstate it here.
MatrixClient API was exposing two methods that only worked for legacy crypto:
- getDeviceEd25519Key
- getDeviceCurve25519Key
=> These are used in the react-sdk for some functionality (rageshake, sentry, rendez-vous).
I have deprecated those calls from MatrixClient and created a new API in CryptoApi (where it belongs):
getOwnDeviceKeys(): Promise<OwnDeviceKeys>
If we call methods on `OlmMachine` after `MatrixClient.stopClient` is called,
we will end up with a "use of moved value" error. We can turn these into
something more useful with judicious use of `getOlmMachineOrThrow`.
Alternatively, we can sidestep the issue by bailing out sooner.
Andy Balaam