I think the test is failing because the timeline created by Room::timeline is now not cached anymore in the Room instance; as a result, it could be that the fresh timeline reconstructed has minimal events, because sync sent a gap, and the timeline was reset as a result. If that's the case, then one of two things must be done about this:

• backpaginate with the fresh timeline recreated after the sending
• make sure to use the same timeline instance along the test

For what it's worth, it seems that Tests / Tests (viz., Ensuring client implementations work) now passes, after a first failed attempt.

I wonder if the live_events_rust test may be intermittently failing, because of the following:

• the Rust SDK is being created, and sync is started on it
• there's an RPC client that will create a room and send messages to it
• the Rust SDK is then being polled to see if it's received the events (through a timeline attached to the new room)

In some cases, it seems the Rust SDK doesn't emit timeline diff updates for all the events in the room. I think this may happen because the server marked a sliding sync as limited/gappy, so the SDK may have reset the timeline, resulting in a limited view over the timeline's events. It's not possible to tell if that happened, from looking at the current logs (and unfortunately I can't run this test locally, as I'm running into strange unmarshall errors, maybe related to using google-chrome at a different version than the one expected here).

Some events appear as missing, but they should be present in the persistent storage. If a subsequent back-pagination happened they'd be immediately reloaded from the store, if they've been seen; or correctly reloaded from the network, if the server decided to return only a subset of the events for any reason. The above test existing_events does run a back-pagination, with a comment meaning that it could be that some events are missing; I think that with the persistent storage, it's easier to assume the same thing, even in the case of live events, as a reset may happen, depending on how the events are received from the sync (in a single batch or in multiple batches, was the room subscribed to, etc.).

Of course, I might be wrong here, and there might be a bug where the client is missing some events, but this sounds rather unlikely. I'm tempted to add a back-pagination there too, in case we're hitting this edge-case, but maybe we can see how this works out first, in general.