By setting the ncurses_builtin_key_parser ui_option to true, we
can disable ncurses parsing of key strokes to get less portable
parsing but support for more complex modifiers.
Probably, the extra «s» at the end of «exist» was added accidentally. A verb after «does not» in Present Simple definitely shouldn't have this extra «s».
Ideally, something better should be done (re #2554) but this is a decent
intermediate step for some useful keys.
Note: NCurses supports parsing these keys when shifted (KEY_SR,
_SLEFT, S_RIGHT, etc), but it does not do the same thing for the other
modifiers.
Not having the `test` target (in the Makefile) depend on the `kak` one
prevents users from running commands that make use of parallelism, e.g.:
$ make -j all test
The above command sometimes results in the test suite running before
the binary has been compiled and symlinked, resulting in failures.
SIG_IGN is inherited after 'execve' and requires us to reset
that signal handler, which does not work well with vfork on
OSX. Using an empty function does the trick and will be auto
reset to default on exec.
When large buffers have been opened, copying Kakoune's memory
page descriptors can get pretty slow, making fork more expensive
than necessary.
vfork avoids that problem. While not strictly conforming, it seems
the few calls we do before execve (open, close, dup2 and
set_signal_handler) would not cause any problems on platforms we
care about.
the buffer name was not a very interesting information, whereas
the buffer range allows a hook to run only on the appended text
instead of all the buffer.
m_last_setup was not storing the actual position that was used to
redraw the window, but the previous one, leading to an additional
spurious redraw immediatly after (triggered by window position not
believed to be the one at last redraw).
Fixes#2562
As we do at most one push_next per step_thread, and we pop_current
before step_thread, we can avoid a branch there at the expense of
sometimes growing unecessarily (once).
It seems unlikely this would give performance gain, as buffer
lines are always accessed when we read that field, leading to
all the necessary data already being in memory. Removing it
reduces the size of a BufferIterator, which are already pretty
hefty objects.
The RegexHighlighter range cache can get pretty big in nested
regions use cases, and maintaining it can become pretty costly,
so if it hits a certain size, just drop it.
Should improve performances in #2454
This should greatly reduce memory usage by only caching matches
for ranges that needs to be highlighted, in the case where multiple
regions are nested, this means only the topmost region needs to parse
and cache the whole buffer, other regions highlighter will only ensure
the lines for the ranges they are called up are cached.
Fixes#2454
This adds a limitation that capture matching on regions only works
if the regions start/end/recurse match is less than 65635 byte long.
With this limitation we can reduce the RegexMatch struct size to 16
bytes instead of 32.
This is still not good enough,but should slightly improve high memory
usage as reported in #2454
This commit also introduces a regression in that I decided that the best way to
avoid overly long and confusing names was to rename the current shell-*
switches to script-*, and have the shell command completion be
shell-completion.
renamed script-{completion,candidates} to shell-script-*
Updated docs with new switch names
Added -shell-completion switch to x11-repl and kitty-repl
Final is more granular, it consists of FinalFg (f), FinalBg (g)
and FinalAttr (a) which control if a face's fg, bg, or attributes
fully overwrite the previous face (instead of merging) and if
following faces apply on top of this face or not.
Fixes#2388 if the Whitespace face has the FinalFg flag.
Add <c-w> and <a-d> (along with <c-W> and <a-D> that work on WORDs),
and <c-y> which pastes the transient clipboard contant (which saves
big erase, such as word erase and line end/begin erase).
Fixes#2355
In the end, no better solution materialized so far, and custom
Kakoune line editing bindings are hard to remember. Using well
known readline bindings seems just more convenient.
Closes#800, although it does not contain all the binding proposed
by it (I might accept a few additional ones, such as <c-w>, but not
too much, I still see that as a hack pending a nicer solution).
Previously, when wrapping lines at word boundaries, we would iterate forwards
for "wrap-width" characters, then iterate backwards until we found a word-break,
which was horribly slow.
Now we record the last word-boundary we saw as we iterate forwards, getting a
result in one pass.
Fixes#2339.
Due to a copy-paste mistake, the `:kill` command in a session with
multiple clients was the equivalent of a force-kill (`:kill!`).
This commit makes sure all buffers are saved before killing the
session, unless the force flag is specified.
Cast errors in RPC requests currently make the client quit with an
error saying "uncaught exception", since `Kakoune::bad_value_cast`
exceptions are not explicitely handled.
This commit tries to catch ill-formatted requests and return a more
human-friendly error message, without quitting the client.
This commit implements the -once flag on the `:hook` command, which
automatically removes a hook after it was run, to avoid having to
declare a group and remove it in the hook implementation.
Closes#2277
In particular, this make gathering a transformed range to a vector
faster because we can use the random access nature of underlying
iterator to get the size to allocate in the vector upfront.
Creating a window potentially runs hooks, which themselves could
trigger shell evaluation, which could handle urgent input events
such as a resize, while waiting for the shell to finish. When that
happens, the client had a temporarily null window as it had already
released its own window.
Fixes#2225
