It turns out that Chrome now implements scroll-anchoring itself (ie, content
stays in the same place when you add more stuff off-screen), which means we
cannot rely on back-pagination to make ScrollPanel do a scroll.
Instead, just do a scrollToToken. Which turns out to be considerably simpler
anyway.
- Instead of using one attribute, use one that might just contain one token
- Use the first token when tracking a child
- Mandate that no commas can be in individual tokens
It seems that a number of the tests had started failing when run in
Chrome. They were fine under PhantomJS, but the MegolmExport tests only work
under Chrome, and I need them to work...
Mostly the problems were timing-related, where assumptions made about how
quickly the `then` handler on a promise would be called were no longer
valid. Possibly Chrome 55 has made some changes to the relative priorities of
setTimeout and sendMessage calls.
One of the TimelinePanel tests was failing because it was expecting the contents
of a div to take up more room than they actually were. It's possible this is
something very environment-specific; hopefully the new value will work on a
wider range of machines.
Also some logging tweaks.
Under certain conditions, it was possible to get stuck in a state where any
user-initiated scroll would be met with "Working around
vector-im/vector-web#528" and overridden. Fix this by removing the duplication
between _lastSetScroll and recentEventScroll, and using _lastSetScroll which is
more reliable.
