* [dev] Don't use await for on stdout and stdin; pass local engine argument
1) Don't use await for on stdout followd by await for on stderr.
This can cause nothing to happen. E.g. if one didn't do something
like `flutter pub upgrade` dev/automated_tests and the version in the
old one was pre-nnbd (but flutter itself use lots of nnbd stuff) lots
of errors would be emitted, but because we basically only listen to
stdout nothing will happen (deadlock once stderr buffer runs out).
Note that it still awaits the exit-code below.
2) Pass --local-engine (if given) so there won't be any crashes because
of that.
* Create icons golden test
* Update icons_test.dart
* Update icons_test.dart
* Update icons_test.dart
* change test name
* Add ability to load material icon font for golden tests
* formatting
* Skip test for browser
* Add end of file newline
* Minor doc, style, and perf updates to Navigator/Routes
These are minor fixes I ended up making while working on a larger
project that never went anywhere.
- Used a ColoredBox instead of a DecoratedBox for ModalBarrier
(probably a trivial memory/perf win).
- A bunch of Navigator documentation fixes around when things rebuild.
- Mark routes dirty when the Navigator has a dependency change. I
cannot find a way to test this because as far as I can tell it makes
no actual difference to when things rebuild because whenever the
Navigator rebuilds the Overlay rebuilds and whenever that happens
every OverlayEntry rebuilds, but in theory that's not guaranteed so
this is sort of a correctness fix. It may even be a perf loss. We do
something similar in didUpdateWidget already. I could be convinced
to maybe remove these...
- Make ModalRoute.filter public like everything else.
- Made ModalRoute update its barrier when it gets an update, in case
e.g. the modal barrier depends on inherited widgets via the
navigator context. Again, not sure of any way to detect this, it
might actually be moot, but it seems to be the technically correct
solution?
- Minor style fixes.
All in all I couldn't figure out a way to test any of this (I wrote
multiple large tests but it turns out they all already pass on master
and are effectively redundant with existing tests).
* Remove extraneous blank line
