Passing large diff buffers via the environment can quickly result in
the error "execve failed: Argument list too long". Use a pipe like
in format.kak
When running | (or <a-|>), Kakoune does not use %arg{@} to populate
"$@" (missing feature?). Work around this by moving %arg{@} to a
temporary register. Apparently $kak_quoted_reg_a will never be an
empty list, so work around that too.
When diff parsing fails, we take care to run "fail" in the calling
client, unlike :format (probably a bug in format.kak).
(This patch is best viewed while ignoring whitespace changes (diff -w))
Mapping in the filetype hook matches others like grep.kak and man.kak.
Since we map in buffer scope, git diff buffers will override diff-jump
with git-diff-goto-source.
This means that the diff-jump binding applies here:
diff -u "$1" "$2" | kak -e 'set buffer filetype diff'
Reported in: https://github.com/mawww/kakoune/issues/153#issuecomment-1030643854
A recent commit wrapped diff.kak into a module. The module includes the
hook that adds diff highlighting to filetype=diff buffers. This means
that the hook is only loaded after opening the first diff buffer in a
Kakoune session, so it only actually fires for the second diff buffer.
Fix this by moving the hook out of the module.
Fixes#4525
The wrapper for "git blame" creates flags for each line of the buffer.
It parses the output from git and would send a flag (or a series of
flags) each time the commit to blame for a line differs from the
previous one. For files that were touched by a large number of commits,
this results in a high number of kakoune processes being launched, and
may take some time. This is visible in the session through the flags for
the different commits appearing on the lines one by one, possibly during
several seconds.
To speed up the process, batch flags before passing them to the kak
session. One solution could be to send all flags at once, but this might
delay the appearance of commit info for too long if "git blame" really
takes a long time. The alternative solution retained for this commit
consists in grouping as many flags as we can during one second
(roughly), to pass them to kakoune, and then to move on to the next
flags. This way, a new batch of commit information flags appears every
second or so in the client, until all information is added. This should
be much faster than lauching a kakoune process for each commit
reported by "git blame": tests have shown that blaming a large file in
the Linux repository goes 4.5 times faster when batching flags.
Co-authored-by: Johannes Altmanninger <aclopte@gmail.com>
This allows to jump from a mail buffer that contains an inline diff
to the source files (most accurate when the patch has been applied
locally).
This makes the diff module a mandatory dependency; we could relax that.
When reading and writing emails that contain patches (possibly
email-quoted), it can be convenient to the jump to the source file.
Allow this by making diff-jump (bound to <ret> in git-diff
buffers) ignore leading email quotes ("> "). A line that starts with
"> " should not occur in a unified diff, so this won't affect other
use cases.
Observe that diff-jump even works around interleaved replies; they
will not affect the computed line numbers because we ignore lines
that don't match ^(> )*[ +-].
git-diff-goto-source is specific to diffs produced by Git. This patch
generalizes the logic and moves it to a new diff-jump in diff.kak.
The main differences are:
- diff-jump handles plain file diffs (i.e. without the -r option). These
have no "diff" line. This means that it needs to parse +++/--- instead.
- diff-jump can go to the old file, not just the new one.
- diff-jump allows to override the base directory and the number of
directory components to strip.
git-diff-goto-source was implemented with several nested try/catch
blocks. Implementing the extra features would have added more
nesting, redundancy or hidden options. To avoid that, I ported the
parsing logic to Perl (which git.kak already depends on). Maybe
it's possible to do the same in awk.
Potential concerns:
- We could move diff-jump to a new rc/tools/diff.kak but then it's not
obvious where the "diff" module belongs to.
- Should diff "diff-jump -1" be spelled "diff-jump -p1"?
In future, the diff parser could be reused to implement a vimdiff-style
feature: given a diff and the "old" line number, we can compute the
corresponding "new" line number. Perhaps diff-jump should get a -client
argument.
We want to move git-diff-goto-source from rc/tools/git.kak
to rc/filetype/diff.kak (or should we could create
rc/tools/diff.kak?). Either way, create the diff module so we can
formalize this dependency.
Currently this module only provides highlighters, so require it
wherever we reference them.
Keep the diff-select-{file,hunk} commands outside the module because
people might already use them in git buffers.
When `set` or `se` is found at the start of the modeline, it should stop parsing options after `:`.
When `modeline-parse` is called in the following file, it should _not_ recognize `tabstop=4` and `invalid_option=3`.
```
# kak: set indentwidth=0 tabstop=16: tabstop=4 invalid_option=3
```
More info: http://vimdoc.sourceforge.net/htmldoc/options.html#modeline
adds the ability to press <ret> within a hunk and navigate to the original
source code. This can be useful because one often needs to go back and forth
between the diff and the full source code.
- You can press <ret> anywhere _within_ a hunk i.e. lines that start with
` `, `+`, `-`. You will be taken to the exact line in the source that corresponds
to where you pressed <ret> in the hunk. It actually does not make sense
to press <ret> on a `-` line because that does not exist anymore but
in that case you are taken to a nearby line in the hope this is still useful.
- You can also press <ret> on a range line (lines that
look like @@ ... @@). If you press <ret> on anywhere on a range line e.g.
```
@@ -120,3 +123,4 @@ fn some_function {
```
The code will try to navigate to the section heading "fn some_function {"
Note that the section heading is _not_ necessarily located at the
range line (in the above example the range line is 123).
- You can press <ret> on a +++ line also and you will be taken the first
line of the file
Caveats:
- Navigation to the original source file will be accurate only if any edits to
the original source file have been saved to disk, because otherwise
they will not be detected by the `:git diff` or `:git show` commands
- This feature should work well for most typical uses e.g. `:git diff`, `:git diff HEAD^`
`:git diff <some-sha1>`. In fact this feature should work in all scenarios when
the *current files* on disk are being compared _with_ some arbitrary git revision/staging.
It will be less useful in other scenarios when two arbitrary revisions are being
compared to each other or when you are trying to compare staging to some revision.
For example when you invoke `:git diff --staged` you are trying to compare staging
with HEAD but are navigating to what is currently on disk (which may be different
from staging).
Co-authored-by: Johannes Altmanninger <aclopte@gmail.com>
