This is due to some kind of bug in ListView that never resettles
properly for bottom-to-top views. This can arise when the pinned message
is loaded (because that squishes the view) or the window is resized
(because that also resizes the view.)
We can work around it by assuming the following:
1. The RoomPage knows the window is resizing because it gets its height
changed before TimelineView.
2. The first height change can be a marker to position the view at the
beginning.
This fixes the issue for me, I did the following in order to test this:
* Switch between many rooms, especially ones with a pinned message. Now
all of them start at the bottom as they should.
* Resize the window, ensure that if you scrolled it stays around that
position - otherwise it sticks at the bottom.
This fixes the annoying "I just scrolled down to the bottom, how come
NeoChat doesn't think I did?"
From what I can tell this is also ListView bug (or something caused by
our style/Kirigami) that creates cases like contentY being -643.2 (for a
ListView of height 643) thus that's not "at Y's end". For our case
though, we don't care and can safely round it.
Title this adds a number of options for when messages should be automatically marked as read for the user to choose from.
{width=480 height=262}
This does some further cleanup of RoomPage, mostly removing all the vestigial bits from when we could have multiple windows. But also stuff is moved to TimelineView where possible.
The loading placeholder is removed as TimelineModel already has this built in.
TimelineView now gets room from it's model. This is to ensure we're always using the same room as it which may not be true momentarily when RoomManager.currentRoom changes as the model does it's own reset sequence. This will prevent some race conditions in future (and which I already hit creating other MRs)
Refactor TimelineView to make it more reliable and prepare for read marker choice. This is done by creating signalling from the mode when reset which can be used to move the scrollbar to the newest meassage.
Some of the spaghetti is also removed so there is no need for ChatBar and TimelineView to talk directly.
The code to mark messages as read if they are all visible after 10s has been removed infour of just marking as read on entry if all are visible. This is temporary until a follow up providing user options is finished (although it will be one of the options)
It turns out that for whatever reason ListView.posiitionViewAtEnd() ignores any whitespace. This means that when we use the function it goes to the last delegate. This is no good as we have some padding at the bottom to make space for the typing indicator.
So the fix for this is stupid and involves adding a "spacer" delegate to the timeline beginning model which is completely invisible but qml see as a delegate so we can both leave the space and properly position the view at the end.
BUG: 501075
This merge request addresses a minor papercut in NeoChat's layout when using a screen magnifier.
If you are zoomed in on a conversation, typing indicators are generally invisible. This merge request moves them to align with actual messages, making them more visible to screen magnifier users like myself. That way, you can do the text chat equivalent of waiting for someone else to finish talking before you ramble on.
### Before
### After
Replace HoverActions with an inline action component that appears on hover. There are only actions for reply and react if there is space the overflow button opens the normal message menu.
NOTE: the most recent update changes things slightly, from the images below the buttons are now top aligned because of potentially hige messages. The actions are also now disabled for compact mode as they never really made sense there anyway. The menu now has all options so no one is missing out.
For normal messages
{width=419 height=138}
When space is limited
{width=411 height=130}
User messages
{width=296 height=114}
BUG: 503784
Rework MessageDelegate to be mostly a cpp class. This allows us to only load the components that are actually needed saving memory.
In testing using memtest it saved ~30% versus the current implementation.
