D-N7 puts the develop column under the photograph on a tall window, and the
axis it turns on is aspect rather than width: a 960-wide portrait tablet is
expanded by width and wants the dock, a 1500-wide landscape desktop is
expanded by width and does not. So this cannot be folded into the layout
class, and it is not remembered per class either — closing the column in
landscape closes the dock in portrait, because it is the same column.
`window-resized` reported width alone and now reports both, from a
`shell-height` that subtracts the safe-area insets exactly as `shell-width`
subtracts them: on Android the strips the status and navigation bars occupy
are on the axis being measured, so the aspect of the window and the aspect of
the space the interface actually gets are not the same number.
`column_below` is the decision, with two thresholds rather than one. It is
read on every resize event, and a single threshold means a window dragged
along its own diagonal crosses it several times a second while the pointer is
still down. Entering at 1.25 and leaving at 1.15 is a dead band no plausible
drag re-crosses.
The comment on EXPANDED_MIN_WIDTH claimed a tablet in portrait gets the
compact layout. It does not — its panel is about 960 logical pixels across,
which clears 820 — and that mistaken example is the one D-N2 reasoned from.
Corrected in the same breath, since this is the commit that says what
portrait actually changes.
dtourolle
0e065389db