* Re-scan the slate document tree on each emoji replacement since
doing a replacement will invalidate all the offsets we have.
* Reset the emoji regex each time we use it.
Fixes https://github.com/vector-im/riot-web/issues/7550
Lots of fixes here as a lot of the API has changed (eg. anchorKey /
offsetKey are now anchor.key and offset.key, and collapseFocusToThing
is moveFocusToThing).
Also changes the ref to a function (sorry for lumping this into the
same PR).
Hopefully will fix https://github.com/vector-im/riot-web/issues/7105
With this more of the
controls that look like buttons can be operated via the keyboard and
navigated to by screen reader users. This includes editor buttons such
as File upload, Audio / Video call, Right pannel hide button, Jump to
the bottom timeline button, and some more buttons found in the user
settings.
Also I have added alt texts to some images that in turn label buttons
which these happen to be packed in and removed some untranslated alt
texts from decorative non-actionable images that might add more
verbosity when talking about screen reader user experience.
I haven't found anyone who can justify to me why we need
more complicated plurals for i18n (even in Polish) for
%(senderName)s added %(addedAddresses)s and removed %(removedAddresses)s as addresses for this room.
This is more reliable with LL enabled as the syncing user is
only known when it was active in the current timeline
or when the members have been loaded
when peeking, the members weren't being loaded at all because
the room wasn't available yet,
and the need for loading the members was never re-evaluated after that.
This only loads the members once the user has joined the room,
which also helps to avoid load all the members before an invite
is accepted.
the memberlist gets updated constantly while the LL members
are being added to the room state. this slows things down unneed, and is one of the
main cause of dropping frames when member arrive.
