This makes Kakoune behave more consistently when using it for Zig.
Empty comments are already auto-removed from the C-family and Python.
In this PR, string literals that start with \\ are not removed in purpose.
modeline-parse leads by matching an expensive regex against the entire buffer,
which can take a long time on huge files.
Perl takes too long on this regex and it seems not even ripgrep
optimizes the \z component
$ ruby -e '10000.times { puts "a" * 10000 }' > big
$ time rg --multiline --only-matching '\A(.+\n){1,5}|(.+\n){1,5}\z' big | wc -l
10
__________________________
Executed in 419.81 millis
usr time 399.84 millis
sys time 20.78 millis
where
$ time kak big -e q
__________________________
Executed in 179.19 millis
usr time 133.61 millis
sys time 53.50 millis
Let's lose the regex.
Fixes#4911
The x11-terminal command spawns a potentially long-lived terminal
process. The terminal can is completely independent of the Kakoune
session that created it.
Due to how it's implemented, the spawned terminals will have
environment variables "kak_opt_termcmd" and "kak_quoted_reg_a"
set. This can be surprising, especially since, by convention, the
environment contains no lowercase variables. Let's stop exporting them.
When launch matches using `id` kitty tries to match against tab id
before matching windows. When there are multiple tabs it's likely to
match a tab before matching a window.
Use `window_id` directly to avoid any possiblity of matching tabs.
This is only needed for `kitty @ launch` for other commands there is
no specific `window_id` field.
This continues the work started by 0a9c90fec (rc: use a separate
*-insert hook to auto-insert comments, 2021-04-17).
The one that's left is Rust but that one is trickier.
Why?
Most users who pass the current selection to grep likely do not intend to pass
the selection as a regex input string.
This makes the grepcmd use an additional -F flag to perform literal-string
matching for the current selection. The -F flag seems to be the standard flag
for literal-string matching in every grep implementation I've found.
Before, sed would add quotes to every line of the single multiline argument,
causing the final quoted argument to "split ... command" or screen to contain
unquoted newlines such as this (from kakoune-cr):
tell application "iTerm"
tell current session of current window
tell (split vertically with same profile command "env PATH='...' 'sh' '-c' ''
' export KAKOUNE_SESSION=$1'
' export KAKOUNE_CLIENT=$2'
' shift 3'
''
' [ $# = 0 ] && set \"$SHELL\"'
''
' \"sh\"'
' ' '--' '51909' 'client0' 'terminal' ") to select
end tell
end tell
Instead of using sed to do this which operates on single lines at a time, simply
use printf to insert ' quotes before and after the entire argument.
At the moment, inserting a new line while being in a comment result in a
"//<indentation>" instead of "<indentation>//".
To fix this, we just paste the comment and indent after the newline
initial indentation.
Just like the parent commit but for tmux-repl-* commands.
Note that tmux-repl-set-pane without arguments is kind of broken; it
increments the current pane ID, which only works in the most basic
scenarios. We should probably replace it with something better,
with menu completions etc.
The tmux-terminal-window command always spawns windows in the tmux
session where the Kakoune session was started - even if the calling
Kakoune client lives in a different tmux session.
Fix this by always creating the window in the tmux
session of the calling client. We already do the same for
tmux-terminal-{horizontal,vertical}.
I call tmux-terminal-impl with "new-window -a" (instead of
"new-window"), so make sure the fix works for my use case.
I considered retrieving the tmux session ID from the $TMUX environment
variable but the tmux manpage only specifies that $TMUX contains
"some internal data".
A pane's ID is immutable for the lifetime of the tmux server.
Same with window/session IDs.
When creating a new tmux repl, we record all three IDs to later use
them to send text to the repl.
The window/session IDs can be invalidating when a pane is moved to
a different window/session (via "tmux move-pane", "tmux move-window"
etc). This will cause repl-send-text to fail.
Fix this by dropping the redundant and potentially incorrect
window/session IDs. The immutable pane ID is enough.
tmux-send-text would silently fail when the repl is no more. Let's
instead print an error, pointing the user to the *debug* buffer which
has tmux' stderr.
At the moment, inserting a new line while being in a comment result in a
"//<indentation>" instead of "<indentation>//".
To fix this, we just re-order both InsertChar hooks.
Positional arguments in awk’s printf is a feature that is only available
in the GNU implementation of awk (gawk). So the ctags auto-completion feature
was broken in Kakoune if the installed version of awk was nawk or mawk.
This simple change makes it retro-compatible with those versions.
See https://www.gnu.org/software/gawk/manual/html_node/Printf-Ordering.html
When calling `:gopls definition`, the gopls LSP server returns the location of
the selected definition. Then, `gopls.kak` tries to parse this output to
feed the `:edit` command and open the file in Kakoune. To do this, it
uses `sed` to transform `<path>.go:<line>:<colstart>-<colend>` to `<path>.go
<line> <colstart>`. However, if the `<path>` contains a dash character,
the `sed` will fail and strip everything after this first dash, removing
the line and columns information.
Closes#4776
When running modeline-parse on this file:
# kakoune: filetype=ledger:indentwidth=4
I get this error from dash (and a similar one from bash):
sh: 53: readonly: key: is read only
This is because the readonly variable "key" is used elsewhere, both
times as global. Fix this by making both variables local. While
at it, remove an unused variable.
Fixes#4478
This is equivalent to a change to grep.kak in 649e252f7 (bring *grep*
buffer to front in context of toolsclient, 2020-08-14).
If a toolsclient is set, make-next-error (and make-previous-error) will
jump to %opt{make_current_error_line}. This is wrong if the toolsclient
does not show the *make* buffer. In that case make_current_error_line
is undefined and we end up showing the goto menu. This has occasionally
been annoying me for a long time but I never bothered investigating.
Fix this by switching to the *make* buffer. The potential downside
is if make-next-error is run from the toolsclient, where we no longer
jump to the error but that's fine because we can use <ret>.
We can maybe improve this later by extending the logic, see
https://github.com/mawww/kakoune/pull/3656#pullrequestreview-472052285
I can't think of a case where a URL would not start at a word boundary.
Let's add that to the regex. In addition to correctness, this also
slightly improves performance because matching can stop earlier.
$ HOME=$PWD hyperfine -w 1 'git checkout HEAD'{~,}' -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy'
Benchmark 1: git checkout HEAD~ -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy
Time (mean ± σ): 1.123 s ± 0.022 s [User: 1.100 s, System: 0.027 s]
Range (min … max): 1.093 s … 1.174 s 10 runs
Benchmark 2: git checkout HEAD -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy
Time (mean ± σ): 1.019 s ± 0.026 s [User: 1.001 s, System: 0.021 s]
Range (min … max): 0.984 s … 1.051 s 10 runs
Summary
'git checkout HEAD -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy' ran
1.10 ± 0.04 times faster than 'git checkout HEAD~ -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy'
There have been proposals to add more language aliases to markdown.kak
(#4592) and allow users to add their own aliases (#4489).
To recap: various markdown implementations allow specifying aliases
for languages. For example, here is a code block that should be
highlighted as filetype "haskell" but isn't:
```hs
-- highlight as haskell
```
There are lots of aliases out in the wild - "pygmentize -L" lists
some but I don't think there is a canonical list.
Today we have a hardcoded list of supported filetypes. This is hard
to mainta, extend, and it can impact performance.
This patch simply attempts to load the module "hs" and the shared
highlighter "hs". This means that users can use this (obvious?) snippet
to add their own aliases:
provide-module hs %{
require-module haskell
add-highlighter shared/hs ref haskell
}
Untrusted Markdown files can load arbitrary modules, but that was
already true before, and modules are assumed to be trusted anyway.
Since language highlighters are now loaded *after* the generic
code-block highlighter, we need to make sure the language highlighters
take precedence. Do this by making them sub-regions of the generic one.
Closes#4489
This improves performance on the [5MB Markdown
file](https://github.com/mawww/kakoune/issues/4685#issuecomment-1208129806).
$ HOME=$PWD hyperfine -w 1 'git checkout HEAD'{~,}' -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy'
Benchmark 1: git checkout HEAD~ -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy
Time (mean ± σ): 3.225 s ± 0.074 s [User: 3.199 s, System: 0.027 s]
Range (min … max): 3.099 s … 3.362 s 10 runs
Benchmark 2: git checkout HEAD -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy
Time (mean ± σ): 1.181 s ± 0.030 s [User: 1.162 s, System: 0.021 s]
Range (min … max): 1.149 s … 1.234 s 10 runs
Summary
'git checkout HEAD -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy' ran
2.73 ± 0.09 times faster than 'git checkout HEAD~ -- :/rc/filetype/markdown.kak && ./kak.opt big_markdown.md -e "hook global NormalIdle .* quit" -ui dummy'
(These numbers depend on another optimization.)
Filetypes markdown and restructuredtext reuse highlighters from other
filetypes to highlight code blocks. For example, to highlight a code
block of language foo they essentially do
require-module foo
add-highlighter [...] ref foo
This works great if the module name matches the shared
highlighter. This is the case almost all scripts in rc/filetype*.
The only exception is kakrc.kak: the highlighter is named "kakrc"
(just like the filetype) but the module is named "kak".
This requires weird hacks in markdown/restructuredtext. Ideally we
could remove this inconsistency by renaming both the filetype and the
highlighter to "kak" but that's a breaking change. Until we do that,
let's add an alias so we can treat filetypes uniformly. This helps
the following commits, which otherwise would need to add ugly extra
code for kakrc highlighters.
The following commit will generalize this approach, allowing users
to add arbitrary aliases.
As reported in
https://github.com/mawww/kakoune/issues/4685#issuecomment-1200530001
ledger.kak defines a region end that matches every character of
the buffer. This causes performance issues for large buffers.
Since the affected regions are only ever filled with a single color,
just use a regex highlighter instead of a region highlighter.
This improves performance when loading the file for the first time.
Speedup on [example.journal.txt](https://github.com/mawww/kakoune/issues/4685#issuecomment-1193243588)
$ HOME=$PWD hyperfine -w 1 'git checkout HEAD'{~,}' -- :/rc/filetype/ledger.kak && ./kak.opt example.journal.txt -e "modeline-parse; hook global NormalIdle .* quit" -ui dummy'
Benchmark 1: git checkout HEAD~ -- :/rc/filetype/ledger.kak && ./kak.opt example.journal.txt -e "modeline-parse; hook global NormalIdle .* quit" -ui dummy
Time (mean ± σ): 362.1 ms ± 5.1 ms [User: 336.6 ms, System: 30.2 ms]
Range (min … max): 352.6 ms … 369.1 ms 10 runs
Benchmark 2: git checkout HEAD -- :/rc/filetype/ledger.kak && ./kak.opt example.journal.txt -e "modeline-parse; hook global NormalIdle .* quit" -ui dummy
Time (mean ± σ): 271.2 ms ± 16.7 ms [User: 252.8 ms, System: 24.0 ms]
Range (min … max): 253.9 ms … 305.0 ms 10 runs
Summary
'git checkout HEAD -- :/rc/filetype/ledger.kak && ./kak.opt example.journal.txt -e "modeline-parse; hook global NormalIdle .* quit" -ui dummy' ran
1.33 ± 0.08 times faster than 'git checkout HEAD~ -- :/rc/filetype/ledger.kak && ./kak.opt example.journal.txt -e "modeline-parse; hook global NormalIdle .* quit" -ui dummy'
After opening a markdown file
```b
```
```c
int main() {}
```
markdown-load-languages will run an "evaluate-commands -itersel".
The first selection makes us run "require-module b", which fails
because that module can't be found. Since -itersel only ignores the
"no selection remaining" error we fail to run "require-module c". Fix
this by ignoring errors.
Dockerfiles of the form `Dockerfile.foo-bar` were not detected for syntax
highlighting.
Mainly meaning for this to capture _ and -, but I don't see why we wouldn't
capture any special character.
We often use the pattern «map global normal ": foo"». The space
after the colon is unnecessary since execution of the mapping won't
add to history anyway, since 217dd6a1d (Disable history when executing
maps, 2015-11-10).
With the parent commit, the space is no longer necessary for user
mappings, so there is no reason to continue the cargo-cult.
Remove the space from mappings to set a good example.
If I use "man dirname" and select "basename"
SEE ALSO
basename(1), readlink(1)
^------^
then pressing <ret> to trigger man-jump selects everything from "ALSO"
until "basename(1)" Obviously that's not the name of a man page,
so it fails. When I select only "asename" it works.
The bad selection happens because we use a combination of <a-?> and ?
to extend the selection to a full link. This is more complicated than
it needs to be; let's just select the surrounding WORD. This works
fine because man page links never start mid-word, and trailing
characters are ignored anyway.
`x` is often criticized as hard to predict due to its slightly complex
behaviour of selecting next line if the current one is fully selected.
Change `x` to use the previous `<a-x>` behaviour, and change `<a-x>` to
trim to fully selected lines as `<a-X>` did.
Adapt existing indentation script to the new behaviour
The canonical way to disable all auto-insertion hooks is
set-option global disabled_hooks .*-insert
A recent change allowed to disable hooks that insert ) and }
independent of hooks that insert // (a step in the right
direction, we should do it for more filetypes).
Since the new hook ("go-insert-closing-delimiter") doesn't match
.*-insert, it broke the above snippet. Fix this by renaming it to
"go-closing-delimiter-insert".
This makes it a bit less obvious how to disable only comment insertion.
Not sure if there's interest in that, but make it easier by renaming
"go-insert" to "go-comment-insert".
In TOML's triple-quoted strings, it's allowed to place quotes just
next to the triple quotes.
foo = '''bar''''
^ part of the string contents
^^^ closing sequence
We greedily interpret the first three single quotes as closing
sequence. As a result the remaining single quote wrongly opens a new
string. Fix this by only treating triple quotes as closing if they
are not followed by another quote.
I don't think there is another possible interpretation of quadruple
quotes in TOML, so this should not affect other valid inputs.
Fixes#4143
Changed the indentation behavior such that an extra level of
indentation is added after a line containing a ( or { that is
not closed on the same line instead of aligning to the unclosed
( or {. This is consistent with how `go fmt` formats source code.
Added regression tests.
When indenting on newline in Go files, only remove trailing whitespace
on the previous line and copy indentation of the previous line if in
comment context.
Added regression tests.
Add a separate hook group for inserting ) and } on newline because the
current implementation does not work in 100% of cases and should be
able to be disabled independently of copying comment characters (which
is much easier in comparison to get right) if one does not care about
this feature.
While Wayland offers nothing general to help us support `focus` on all
window managers, WM-specific implementations are generally possible.
Sway is a tiling window manager that mimics i3, and has a reasonably
powerful CLI that can help us achieve this.
In addition to supporting `focus` for Sway, this change paves the way
for additional WM-specific Wayland functionality by adding a detection
step to wayland.kak, in a similar fashion to detection.kak.
An indent hook automatically adds whitespace, so it seems prudent to
add the hook to remove unwanted whitespace again. This is what we do
in most languages already.
Some languages have a trim-indent command but don't use it (for no
apparent reason). Make them trim trailing spaces when exiting insert
mode, like most other languages support scripts do.
An OCaml comment is `(* Some comment *)`. Like the C-family this can be
a multiline comment.
Recognize when the user is trying to commence a comment when they type `(*` and
then automatically insert `*)` on behalf of the user. A small convenience.
Co-authored-by: Maxime Coste <mawww@kakoune.org>
Fixes the handling of multiple backslashes before gitignore * and ?
glob patterns. Adds character classes to gitignore highlighting.
Co-authored-by: Johannes Altmanninger <aclopte@gmail.com>
Commit 85b78dda (src: Select the data inserted by `!` and `<a-!>`,
merged on 2021-03-06) broke autorestore by making it delete the
restored content. I've been using it for 6 months but never noticed
since I didn't use autorestore
Reproducer:
HOME=$PWD kak -s foo README.asciidoc -e 'exec iUNSAVED-CONTENT'
# In another terminal:
ps aux | awk '/kak -s foo/ {print $2; exit}' | xargs kill -HUP
HOME=$PWD kak -s foo README.asciidoc
Delete the trailing newline instead of the restored content.
While at it, remove some <space> commands from execute-keys, to make
it work on the breaking-cleanups branch which swaps <space> and ",".
Closes#4335
Add a group to the `file-detection` hooks.
There's no way to remove hooks without a group. With this patch, you'll be able to remove those
`file-detection` hooks manually. There's no need for two separate groups since if you wanted to
remove only one, you could run `remove-hooks` and then only add one again.
Related: #3670
There is a bug that causes `:git show-diff` to fail when using an external diff, for example difftastic.
This change ensures that we don't use an external diff tool when diffing the current buffer.
