Some further belt and braces to ensure that we don't crash in the timeline. This does the following:
- Checks all callback inputs
- Makes sure that any objects are cleaned up from incubators when deleting the delegate as it turns out that clear() doesn't do this. Hopefully this stops callbacks into an already deleted parent.
Hopefully helps with https://crash-reports.kde.org/organizations/kde/issues/261627/events/0b6bef500c5641e1924aa0bc2b0c11bd/
This fixes some glitches during animation, which had the background
disappearing when the pop animation was running,
and now uses always the pagerow for viewing the room.
This depends from the new PageRow animation on mobile,
https://invent.kde.org/frameworks/kirigami/-/merge_requests/1925
which makes all applications coherent with an application which
looks a bit more like the one on most android apps.
Also, the swipe back gesture now will work there too
Hopefully this stops any crashes around QuickActions.qml and EmojiDialog.qml. Best I can guess this is some race condition where QuickActions are deleted in the time it takes to instnatiate the EmojiDialog popup. I've also rearranged the updateQuickActions function to stop a possible race condition there.
BUG: 509484
When in the space overview page, direct chats were previously only
shown with their room ID. Fix this by getting the actual NeoChatRoom
and its displayName.
I broke this in 409b6da160 making it not
possible to join a succeeding room. I decided to keep the old behavior
of not prompting you to join, because:
1. You are intentionally choosing to tap the button.
2. Room successions aren't random, a privileged user in the room is
creating it and there's already some implicit existing trust
(Also the current join confirmation dialog isn't useful enough yet to
make an actual informed decision anyway.)
This wasn't a thing yet, so the search experience compared to something
like Element Web was much worse. Now you can scroll back as far as any
other client can!
I had to do a bit of refactoring to stop resetting the model all the
time, and append instead. Otherwise, it was quite straightforward to
implement.
