By slightly changing the shape of the function used to process read
receipts, we can make it so that it's trivial to run concurrently, which
gives some nice speedups locally.
Distribution of 6 worst case processing time, initial response:
Before:
0.172524963s
0.216173016s
0.252289760s
0.257619156s
0.275838632s
0.280295891s
After:
0.083094692s
0.117074046s
0.130246646s
0.132577343s
0.138685246s
0.170287945s
This patch updates `LatestEvent::update` to call the new
`LatestEvent::store` method, which will store the new `LatestEventValue`
in the `RoomInfo` struct, and will persist it in the `StateStore`.
This patch also adds the test for this new feature.
This patch moves `LatestEventValue` into the
`matrix_sdk_base::latest_event` module. If we want to store this value
in `RoomInfo`, it must be in this crate.
Because it's not allowed to `impl T` where `T` lives in a different
crate, this patch changes the `impl LatestEventValue` to `struct
LatestEventValueBuilder` + `impl LatestEventValueBuilder`. Luckily, all
methods on `LatestEventValue` are only constructors, so the change is
super straightforward.
This patch replaces `OwnedRoomId` by `WeakRoom` in `LatestEvent`. Apart
from simplifying a couple of method' signatures, it also opens the road
for storing the `LatestEventValue` in `RoomInfo`.
Since the server will reject any duplicate one-time keys forever,
clients which encounter such an error will spam sentry with such
reports.
This patch ensures that we only send the sentry report once.
This patch revisits the `assert_latest_event_content` macro to not take
a `true` or `false` value. It feels a bit weird to read. Instead, `with
|factory| { … }, true` becomes `event |factory| { … } is a candidate`.
Same for the `false`case which becomes `is not a candidate`. No more
comma, it feels a bit more like a sentence.
This patch replaces `LocalLatestEventValue::content` and `…::event_type`
fields by using the existing `SerializableEventContent`. It does exactly
the same thing.
The problem is: `LatestEventContent` cannot be serialized. It's annoying
because it means we can't store a `LatestEventValue` (that wraps a
`LatestEventContent`) in the database.
This patch revisits `LatestEventValue`. Before we got:
```rust
pub enum LatestEventValue {
None,
Remote(LatestEventContent),
LocalIsSending(LatestEventContent),
LocalCannotBeSent(LatestEventContent),
}
pub enum LatestEventContent {
RoomMessage(RoomMessageEventContent),
Sticker(StickerEventContent),
Poll(UnstablePollStartEventContent),
CallInvite(CallInviteEventContent),
CallNotify(CallNotifyEventContent),
KnockedStateEvent(RoomMemberEventContent),
Redacted(AnySyncMessageLikeEvent),
}
```
`LatestEventContent::Redacted` contains an `AnySyncMessageLikeEvent`.
That's the part that is not serializable.
It appears that `LatestEventContent` isn't necessary! The only thing we
need is to _filter_ the events by their type, no need to _find and
map_. The `LatestEventValue` can contain the entry event directly (e.g.
a `TimelineEvent` for the event cache). Okay, let's do that.
```rust
pub enum LatestEventValue {
None,
Remote(RemoteLatestEventValue),
LocalIsSending(???),
LocalCannotBeSent(???),
}
type RemoteLatestEventValue = TimelineEvent;
```
What about the `Local*` variants? We can't use a `TimelineEvent`. We
need a new type for that:
```rust
pub enum LatestEventValue {
None,
Remote(RemoteLatestEventValue),
LocalIsSending(LocalLatestEventValue),
LocalCannotBeSent(LocalLatestEventValue),
}
pub struct LocalLatestEventValue {
pub timestamp: MilliSecondsSinceUnixEpoch,
pub content: Raw<AnyMessageLikeEventContent>,
pub event_type: String,
}
```
We don't need the event ID nor the transaction ID in
`LocalLatestEventValue`.
That's the only change. All the other changes are about the tests.
Implements support for decryption of state events
- [ ] Introduce a case for `AnySyncStateEvent::RoomEncrypted` to the
`state_events` sync response processor.
- [ ] Introduce modified `Room::decrypt_event` and
`::try_decrypt_room_event`.
- [ ] Introduce testing macro
`assert_let_decrypted_state_event_content`.
- [ ] Add casts and explicit type hints where necessary.
---------
Signed-off-by: Skye Elliot <actuallyori@gmail.com>
Valere