Use internal types to track items within the
selector/mselector widgets, rather than pressing the
user-provided item structs into double-duty. With this
change, we can constify the strings within those
user-provided items. Do so, also removing the internal-side
elements. Update documentation. Closes#2267.
Add pixel offset fields to blitterargs in the pixel
portion of the union. Check in blitter_geom that
pixel offsets are not used with cell blitting.
Prepare the new blitterargs fields. Account for pixel
offsets when calculating cell geometry #1682.
if we crash while detecting kitty keyboard support, we
currently don't pop the keyboard state. since we set our
desired state before detecting support, this could leave
the terminal in an invalid state. now, we initialize
kbdlevel to UINT_MAX before detection, so that in a crash,
the pop always happens. we reset it at the end of
interrogation, assuming we didn't load it from a query
response. the downside is that a terminal which doesn't
consume this sequence will get hit with it in a crash, but
we're already crashing, so who cares? leaving the terminal
in an invalid state, however, is very annoying.
A Notcurses program might shift between the regular and
alternate screen freely. At all times, we want the Kitty
keyboard protocol to be locked in. At the same time, we
always want a screen's keyboard state at the time we start
using it preserved, and restored when we're done. Thus:
start: optionally smcup, push the state, set the state, query
leave alternate: pop state, rmcup, push state, set state
enter alternate: pop state, smcup, push state, set state
done: pop state, optionally rmcup
"set state" means both XTMODKEYS and kitty set.
Closes#2249
