Fix improper or missing room_stats updates when handling state events (deltas).

[reivilibre]

Pull Request Checklist

• Pull request is based on the develop branch
• Pull request includes a changelog file
• Pull request includes a sign off

[codecov]

Codecov Report

Merging #5691 into develop will increase coverage by 0.02%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           develop    #5691      +/-   ##
===========================================
+ Coverage    63.32%   63.35%   +0.02%     
===========================================
  Files          331      331              
  Lines        36140    36144       +4     
  Branches      5954     5956       +2     
===========================================
+ Hits         22887    22900      +13     
+ Misses       11622    11613       -9     
  Partials      1631     1631

[reivilibre]

Hrmm. It has come to my attention that the fix proposed here for #5690 isn't quite correct.

For rooms joined over federation, (e.g. librepush.net joining a room on matrix.org), state_events in librepush.net's room_stats will not count state events that have been overwritten in the past – I guess these don't cause deltas.

Edit: On the other hand, if a user makes the membership state transition leave→join and then you join a room over federation, the left_members stat is still sane and isn't negative – somehow. I was expecting it to go to -1 since the user left a leave state…

Edit: This state_events discrepancy will appear during backfill, as deltas seem to be coalesced.
e.g. if we have server S and T in federation, the following steps:

• Take T offline
• On S, send state events (transitions) A→B and B→C from S with the same state key/type. (A, B and C here represent some state content.)
• Bring T online again and wait for it to backfill.

Lead to T handling the delta A→C and increasing state_events by 1 in its room_stats whereas S has increased its by 2 (the actual № state events).

I'm not (yet) sure how to address this.

[reivilibre]

The above issue has now been addressed. I am happy for this to be reviewed. (Maybe will save its merge to be done alongside some other upcoming PRs…?)

[erikjohnston]

This seems legit to me, but I'd like a quick check from @hawkowl too

[reivilibre]

N.B. This PR may (or may not – we will have to see) be more or less obsolete given that I am going to make quite a bit of change to the room stats stuff. I'll leave it here for now.

[reivilibre]

Obsolete in the face of #5847