Main highlighter run on itself, on the optimized version:
- optimized (8 runs, 3 last noted):
1.1201650000
1.1074430000
1.1263810000
- unoptimized (8 runs, 3 last noted):
1.5746400000
1.5115250000
1.5155440000
Average difference: 0.415907
Running main highlighter on itself (both runs are on the optimized
version):
- clean (8 runs, last 3 noted):
1.7007670000
1.7330720000
1.7038810000
- optimized (8 runs, last 3 noted):
1.5007230000
1.5142960000
1.4973320000
Average difference: 0.208456
When parsing main-highlighter with itself, the *_check_path function is
called 426 times. Note that there are 686 region_highlight resulting
entries.
This is a prerequisite for the next commit.
The incumbent code was wrong: the test of $in_redirection in the first hunk
would never have seen «(( in_redirection == 2 ))». That had no visible
effect since options to sudo don't look like redirection operators.
* danielsh/m0vie-i288-v2:
driver: Don't highlight in isearch
driver: Always bind zle-line-finish and use it instead of accept-*
driver: Widget binding: Support binding incomplete/nonexistent widgets
Special handling for cursor imprint or partial path highlighting
is needed in more cases than accept-*. For example when accepting
a line from isearch, no accept-* widget is invoked.
The proper way is to use zle-line-finish.
Trumps #259.
Fixes#284.
Merge remote-tracking branch 'upstream/pr/342'
* upstream/pr/342:
'main': Highlight a broken symlink as a file.
tests: Add an XFail test for issue #342.
Merge remote-tracking branch 'danielshahaf/i267-linewise-region-v1'
* danielsh/i267-linewise-region-v1:
tests: Add a regression test for issue #267, concerning highlighting a vi linewise region.
Support linewise region.
If the separator feature is disabled, this makes it possible to
change the main 'path' styles in a running session without the
need to touch the '_pathseparator' styles, too.
Merge remote-tracking branch 'upstream/pr/331'
* upstream/pr/331:
'main': Try the "non-command word" codepath when a word may be either a command word or a non-command word.
* upstream/pr/319:
tests: Extend issue #290 test with another variant.
'main': Fix issue #290, «Mishighlights "longloops" 'repeat'».
tests: New test for issue #290, «Mishighlights "longloops" 'repeat'».
Assignments to positional parameters using scalar assignment syntax is
a feature: it is covered by upstream's test suite ("append to positional
parameter" in Test/A06assign.ztst).
This commit is based on the work done by
Jorge Israel Peña (blaenk) in #136.
Changes:
* Adjusted to changes on the latest master branch.
* Use regular path highlighter colors by default.
* Break out early if the separator color is the same to
improve performance.
* Tests.
If possible, try to use the zsh/parameter module to get
information about a shell words.
This avoids subshells and is a huge speed improvement
on systems such as cygwin.
Note 1:
$commands does not know about PATH_DIRS. So in case
PATH_DIRS is set, 'type -w' is still used if nothing
else matches.
Note 2:
zsh/parameter can't distinguish between 'command' and
'hashed'. Adjusted the test for that case to XFAIL.
The ideal solution would be if whence had an option to
put the result in REPLY instead of printing it to stdout.
* upstream/pr/308:
'main': Fix issue #302, "Process substitution misparsed as redirection".
'main': Factor out common code. No functional change.
'main': Add a regression test for issue #302, "Process substitution misparsed as redirection".
Consider indexing long buffers (say 300 chars). E.g. following line:
integer offset=${${buf[start_pos+1,len]}[(i)$needle]}
will gradually go into large indices for first index [a,] and be kept
at large index for second index [,b]. Instead, we can remove already
processed characters from buf:
proc_buf="${proc_buf[offset + $#arg + 1,-1]}"
$offset and $#arg are small, first index will not be large. $proc_buf
will be smaller and smaller and the second index [,-1] will run shorter
and shorter.
Since the _zsh_highlight_add_highlight that the tests use ignores
ZSH_HIGHLIGHT_STYLES, we can test both an injective mapping and an empty
mapping at once.
_zsh_highlight_add_highlight appends to region_highlight $1 $2 and the
next non-null parameter. If there is no non-null parameter, do nothing,
This is so that highlighters can add a style with fallbacks if the most
specific style is not defined. If none of the applicable styles are
defined, do the right thing and don't add an invalid entry to
region_highlight.
The pattern highlighter doesn't use this function as it'd need too large
of an API change.
Compute «${#BUFFER}» just once. I haven't measured the effect of this, but
it might just avoid a strlen() every time around the loop, which could add
up for large ${BUFFER}s.
See issue #295.
zsh older than workers/28418 (zsh revision 23bdfc7fd2a012d5205ed22d18eb39e41c8fbc95)
doesn't support the «${name:offset:length}» syntax. Therefore, use the older
«$name[start,end]» syntax instead.
Followup to 3cb58fd7d7, "Optimization: -1 is
extra computation of string length internally".
Fixeszsh-users/zsh-syntax-highlighting#279.
./parse.zsh "zplugin.zsh" > out_norm.txt 106,17s user 7,84s system 98% cpu 1:55,85 total
./parse.zsh "zplugin.zsh" > out_opt.txt 51,91s user 7,44s system 99% cpu 59,912 total
% wc -l zplugin.zsh
3188 zplugin.zsh
