intermittent failures in these tests
The guess that motivates this commit about what's going wrong is that
under some conditions that have to do with parallel tests that might
grab the OS-level focus, racket/gui is deciding that some of the two
keystroke combinations count as separate keystrokes, instead of being
combined in the search for the keymap handler. For example, if the
test suite tries to type c:c and then c:l, that will come in as two
separate events. If unlucky timing coincidence happens, then
racket/gui will decide that those should not count together in the
search for a keymap entry, but count as two separate ones (so we'd get
c:l centering the view instead of c:c;c:l inserting let-ans).
more specifically, every second it will check the memory use;
if the change is more than a megabyte it will update the display
(without some kind of buffer like that, the result will be
different each time it is polled; I tried that an it was
too distracting)
closesracket/drracket#92
GL contexts and GC blits don't work, dialog placement doesn't work,
and checkbox animations seem to interefere with updating --- but at
least things mostly work.
Relevant to racket/racket#1547
Adjust a workaround for versions before 3.22 when setting the font for
a control.
GTK+ version 3.22 starts paying attention to whether a font size for a
control is absolute (as opposed to being in points), so the workaround
that was put in place for earlier versions breaks.
In addition, some part of the drawing stack seems to round point sizes
to an integeral size after DPI conversion. Take that rounding into
account when setting the font size in `normal-control-font`.
Closes#1522
There's no run-time representation of `NSApplicationDelegate`, so
trying to declare it ands up adding NULL as a protocol to the
Objective-C class for the application delegate.
Adding NULL that way leads to a crash on 10.12.1+ with TouchBar
support.
Closesracket/racket#1520
