d970501c85
In a previous commit I moved the auth chain calculation code to ruma because I thought I could optimize it by only taking auth chains from conflicted events instead of all events. It turned out that was wrong and now I removed that algorithm again (the full auth chains are now passed in as an argument to state_res::resolve again).
Matrix State Resolution in Rust!
/// Abstraction of a PDU so users can have their own PDU types.
pub trait Event {
/// The `EventId` of this event.
fn event_id(&self) -> &EventId;
/// The `RoomId` of this event.
fn room_id(&self) -> &RoomId;
/// The `UserId` of this event.
fn sender(&self) -> &UserId;
// and so on...
}
/// A mapping of event type and state_key to some value `T`, usually an `EventId`.
pub type StateMap<T> = BTreeMap<(EventType, Option<String>), T>;
/// A mapping of `EventId` to `T`, usually a `StateEvent`.
pub type EventMap<T> = BTreeMap<EventId, T>;
struct StateResolution {
// For now the StateResolution struct is empty. If "caching" `event_map`
// between `resolve` calls ends up being more efficient (probably not, as this would eat memory)
// it may have an `event_map` field. The `event_map` is all the events
// `StateResolution` has to know about to resolve state.
}
impl StateResolution {
/// The point of this all, resolve the possibly conflicting sets of events.
pub fn resolve<E: Event>(
room_id: &RoomId,
room_version: &RoomVersionId,
state_sets: &[StateMap<EventId>],
auth_events: Vec<Vec<EventId>>,
event_map: &mut EventMap<Arc<E>>,
) -> Result<StateMap<EventId>> {;
}
The StateStore trait is an abstraction around what ever database your server (or maybe even client) uses to store Persistant Data Units.
We use rumas types when deserializing any PDU or it's contents which helps avoid a lot of type checking logic synapse must do while authenticating event chains.