Fixes#1433 although in a slightly different way than requested:
We ensure that scrolloff *displayed* lines are visible below the
cursor, not scrolloff *buffer* lines.
Homebrew advises not using homebrew/dupes anymore:
% brew install homebrew/dupes/ncurses
Warning: homebrew/dupes was deprecated. This tap is now empty as all its formulae were migrated.
...
Warning: Use ncurses instead of deprecated homebrew/dupes/ncurses
John Doe Copyright Waiver
I dedicate any and all copyright interest in this software to the
public domain. I make this dedication for the benefit of the public at
large and to the detriment of my heirs and successors. I intend this
dedication to be an overt act of relinquishment in perpetuity of all
present and future rights to this software under copyright law.
Until now, buffer had multiple recognized end coordinates, either
{ line_count, 0 } or { line_count - 1, line[line_count - 1].length }.
Now the only correct end coord is { line_count, 0 }, removing the need
for various special cases.
The previous solution, ad9b090ddf, was
failing when completion was not replacing any text, as going through
SelectionList was not permitting to replace empty ranges.
Handle replacement as first a deletion then an insert to get fast
update, as having a list of forward deletions and then a list
for forwad insertion is much faster to update than a list of
interleaved delete/insert.
Instead of highlighting full lines and then trim them to make them
fit in the window, highlight only the visible portion, and rely on
the compute_display_setup system introduced for wrapping to setup
our buffer range correctly
Use the fast, linear time SelectionList::insert algorithm instead of
an ad-hoc one that needs to call SelectionList::update after each
modification of the buffer.
Fixes#1417
There is no need to trigger that event on every keystroke, we can
trigger it only when we hit the idle timeout, avoiding computations
when input keys are pasted.
A hook execution triggered by another hook execution would change the
shared MatchResults object, which is a wrong behaviour and makes it
point to dead string data.
The existing HookManager was able to run arbitrary hook functions,
but in practice was only used for user provided textual hooks.
That separation was causing some suboptimal performances, by moving
that logic directly in the hook manager we can improve hook filtering
performance which is a big part of startup time when opening lots
of files.
