When multiple m.replace edits arrive concurrently, getLastReplacement()
may block on decryption. If an older edit's decryption completes after a
newer edit has already been applied, the older async result overwrites
the target event with stale content.
Add a monotonic update counter (replacementUpdateId) and centralise all
replacement updates through updateTargetEventReplacement(). The method
captures the counter before awaiting and discards the result if a newer
update has started in the meantime.
This race is especially pronounced in encrypted rooms with rapid
streaming-style edits, where variable decryption timing causes
out-of-order promise resolution.
* Skip unwritten tests
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Tidy jest fake timers
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Remove unnecessary sessionStorage mock
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Improve types
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Improve async assertions
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Improve error assertions
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Improve object assertions
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Remove assertion testing unclear mock
This test failed when ran individually, same as after the clearAllMocks call
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Avoid awaiting non-thenables
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Pass nop function when stubbing out console, vitest won't accept it any other way
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Remove unnecessary mock which causes tests to fail after updating fetch-mock & fix typo
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix mistaken assertions not testing all values in array
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix hidden non-running tests in room.spec.ts
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Update fetch-mock-jest to @fetch-mock/jest
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Delint
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Make knip happier
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Make knip happier 2.0
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Delint
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Switch from Jest to Vitest
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Delint
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix CI
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Remove unnecessary fake timers
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Update vite
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Revert irrelevant changes
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix coverage spec paths
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix slow test reporter
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix bad merge conflict resolution
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix babel config
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
---------
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* WIP support for state_after
* Fix sliding sync sdk / embedded tests
* Allow both state & state_after to be undefined
Since it must have allowed state to be undefined previously: the test
had it as such.
* Fix limited sync handling
* Need to use state_after being undefined
if state can be undefined anyway
* Make sliding sync sdk tests pass
* Remove deprecated interfaces & backwards-compat code
* Remove useless assignment
* Use updates unstable prefix
* Clarify docs
* Remove additional semi-backwards compatible overload
* Update unstable prefixes
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix test
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Add test for MSC4222 behaviour
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Improve coverage
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Fix tests
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Tidy
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Add comments to explain why things work as they are.
* Fix sync accumulator for state_after sync handling
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Add tests
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Revert "Fix room state being updated with old (now overwritten) state and emitting for those updates. (#4242)"
This reverts commit 957329b218.
* Fix Sync Accumulator toJSON putting start timeline state in state_after field
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Update tests
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Add test case
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
* Iterate
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
---------
Signed-off-by: Michael Telatynski <7t3chguy@gmail.com>
Co-authored-by: Hugh Nimmo-Smith <hughns@matrix.org>
Co-authored-by: Michael Telatynski <7t3chguy@gmail.com>
Co-authored-by: Timo <toger5@hotmail.de>
* Refactor Relations to not be per-EventTimelineSet
* Fix comment and relations-container init
* Revert timing tweaks
* Fix relations order test
* Add test and simplify thread relations handling
* Fix order of initialising a room object
* Fix test
* Re-add thread handling for relations of unloaded threads
* Ditch confusing experimental getter `MatrixEvent::isThreadRelation`
* Fix room handling in RelationsContainer
* Iterate PR
* Tweak method naming to closer match spec
This changes the "relations created" event to ensure it is properly emitted even
if the target event is added to the timeline after the relation event. There was
perhaps always a risk of this happening in the past, but is seems more likely to
bite now with delayed decryption.
Part of https://github.com/vector-im/element-web/issues/17461
The `Relations` model was relying on object reference equality to prevent
duplicates, which breaks down if we ever have two objects that represent the
same event.
This fixes things to additionally track event IDs we've seen before and discard
any attempts to add them twice.
Fixes https://github.com/vector-im/element-web/issues/11161
Bas Nijholt