Synapse returns unspecced property `original_event` from `/relations`

[richvdh]

Synapse's implementation of /relations returns an unspecced property original_event:

https://github.com/matrix-org/synapse/blob/v1.60.0/synapse/handlers/relations.py#L166

This is not defined in the MSC so should be removed.

[clokep]

This is not defined in the MSC so should be removed.

Or should it be defined in MSC2676? See #5626 for why this was originally included. Maybe this is no longer necessary with #12476 though?

[revidee]

Another use-case to consider:
With it removed, there is no "easy" way to retrieve the most up-to-date version of an event in all cases. the /event/ endpoint currently does not aggregate edits and returns the original event (by design in MSC2676). The server should bundle all relations via m.relations in the unsigned content, however, for e2ee rooms, this is not the case for m.replace events, since get_bundled_aggregations calls get_applicable_edits which only returns applicable edits of type m.room.message, but not m.room.encrypted (SQL 1, SQL 2) (This may be a bug in of itself, I'm not sure). Therefore, the client then needs to also make an additional request to /relations/ (with an filter on m.replace) in order to fetch the most recent applicable edit (decided client-sided). Maybe more, if the client needs to paginate through the relations.

When does this happen? After an initial sync without the whole history loaded and in the case client needs to render some out-of-timeline event due to a reference to an (potentially very) old event (m.reference, m.in_reply_to), the client needs to load this single event from the server, and preferably with all edits applied to it.

So in short, in order to then query the most recent state of an event, one would have to call the /event/ endpoint to get the original message, then, call /relations/ to get all edits and apply the most recent applicable.

With the original_event still included in the /relations/ result, the client would save 1 RTT and only needs to call /relations/.

[richvdh]

This is not defined in the MSC so should be removed.

Or should it be defined in MSC2676? See #5626 for why this was originally included. Maybe this is no longer necessary with #12476 though?

Well, MSC2676 is concluded and in the process of being specced, so it would need to be a new MSC if there is a viable usecase for it.

[richvdh]

With it removed, there is no "easy" way to retrieve the most up-to-date version of an event in all cases.

I haven't fully grokked your comment, but this part is incorrect: you can use /context for this.

[revidee]

With it removed, there is no "easy" way to retrieve the most up-to-date version of an event in all cases.

I haven't fully grokked your comment, but this part is incorrect: you can use /context for this.

Unfortunately, not. In the case of e2ee rooms, the last applicable edit will not return anything, since the query (SQL 1, 2 in my comment) only checks for m.room.message. Hence, the server won't return the most-up-to-date-version.

[richvdh]

since the query (SQL 1, 2 in my comment) only checks for m.room.message

well, that sounds like a separate bug: #12793 Edit: #12503 actually I think

[clokep]

This is not defined in the MSC so should be removed.

Or should it be defined in MSC2676? See #5626 for why this was originally included. Maybe this is no longer necessary with #12476 though?

Well, MSC2676 is concluded and in the process of being specced, so it would need to be a new MSC if there is a viable usecase for it.

Ah, I had missed this already had FCP proposed on it.

I think you're correct though that with a combination of using /event or /context and fixing #12503, this becomes unnecessary.

[richvdh]

Well, MSC2676 is concluded and in the process of being specced, so it would need to be a new MSC if there is a viable usecase for it.

Sorry, this is incorrect; I must have been confusing it with some other MSC. (MSC2675, possibly, which specs /relations).

[clokep]

It looks like the matrix-js-sdk, matrix-android-sdk2 and matrix-ios-sdk all use this field in the response. matrix-js-sdk actually uses the field, I haven't really checked the other SDKs.

[richvdh]

Thanks for checking into that, @clokep. Would you be able to raise issues against the relevant projects?