Nvim documentation: pi_matchit

main help file
*pi_matchit.txt*   Extended "%" matching

For Vim version 6.3.  Last change:  2017 May 14

*matchit* *matchit.vim*

1. Extended matching with "%"				|matchit-intro|
2. Activation						|matchit-activate|
3. Configuration					|matchit-configure|
4. Supporting a New Language				|matchit-newlang|
5. Known Bugs and Limitations				|matchit-bugs|

The functionality mentioned here is a plugin, see |add-plugin|.
You can avoid loading this plugin by setting the "loaded_matchit" variable
in your |vimrc| file:
	:let loaded_matchit = 1


1. Extended matching with "%"				*matchit-intro*

%	Cycle forward through matching groups, such as "if", "else", "endif",
	as specified by |b:match_words|.

							*g%* *v_g%* *o_g%*
g%	Cycle backwards through matching groups, as specified by
	|b:match_words|.  For example, go from "if" to "endif" to "else".

							*[%* *v_[%* *o_[%*
[%	Go to [count] previous unmatched group, as specified by
	|b:match_words|.  Similar to |[{|.

							*]%* *v_]%* *o_]%*
]%	Go to [count] next unmatched group, as specified by
	|b:match_words|.  Similar to |]}|.

							*a%* *v_a%*
a%	In Visual mode, select the matching group, as specified by
	|b:match_words|, containing the cursor.  Similar to |v_a[|.
	A [count] is ignored, and only the first character of the closing
	pattern is selected.

In Vim, as in plain vi, the percent key, |%|, jumps the cursor from a brace,
bracket, or paren to its match.  This can be configured with the 'matchpairs'
option.  The matchit plugin extends this in several ways:

	    You can match whole words, such as "if" and "endif", not just
	single characters.  You can also specify a |regular-expression|.
	    You can define groups with more than two words, such as "if",
	"else", "endif".  Banging on the "%" key will cycle from the "if" to
	the first "else", the next "else", ..., the closing "endif", and back
	to the opening "if".  Nested structures are skipped.  Using |g%| goes
	in the reverse direction.
	    By default, words inside comments and strings are ignored, unless
	the cursor is inside a comment or string when you type "%".  If the
	only thing you want to do is modify the behavior of "%" so that it
	behaves this way, you do not have to define |b:match_words|, since the
	script uses the 'matchpairs' option as well as this variable.

See |matchit-details| for details on what the script does, and |b:match_words|
for how to specify matching patterns.

MODES:			*matchit-modes* *matchit-v_%* *matchit-o_%*

Mostly, % and related motions (|g%| and |[%| and |]%|) work just like built-in
motion commands in |Operator-pending| and |Visual| modes.  However, you
cannot make these motions |linewise| or |characterwise|, since the |:omap|s
that define them start with "v" in order to make the default behavior
inclusive.  (See |o_v|.)  In other words, "dV%" will not work.  The
work-around is to go through Visual mode:  "V%d" will work.

LANGUAGES:					*matchit-languages*

Currently, the following languages are supported:  Ada, ASP with VBS, Csh,
DTD, Entity, Essbase, Fortran, HTML, JSP (same as HTML), LaTeX, Lua, Pascal,
SGML, Shell, Tcsh, Vim, XML.  Other languages may already have support via
the default |filetype-plugin|s in the standard vim distribution.

To support a new language, see |matchit-newlang| below.

DETAILS:				*matchit-details* *matchit-parse*

Here is an outline of what matchit.vim does each time you hit the "%" key.  If
there are backrefs in |b:match_words| then the first step is to produce a
version in which these back references have been eliminated; if there are no
backrefs then this step is skipped.  This step is called parsing.  For
example, "\(foo\|bar\):end\1" is parsed to yield
"\(foo\|bar\):end\(foo\|bar\)".  This can get tricky, especially if there are
nested groups.  If debugging is turned on, the parsed version is saved as

Next, the script looks for a word on the current line that matches the pattern
just constructed.  It includes the patterns from the 'matchpairs' option.
The goal is to do what you expect, which turns out to be a little complicated.
The script follows these rules:

	Insist on a match that ends on or after the cursor.
	Prefer a match that includes the cursor position (that is, one that
		starts on or before the cursor).
	Prefer a match that starts as close to the cursor as possible.
	If more than one pattern in |b:match_words| matches, choose the one
		that is listed first.


	Suppose you
		:let b:match_words = '<:>,<tag>:</tag>'
 	and hit "%" with the cursor on or before the "<" in "a <tag> is born".
	The pattern '<' comes first, so it is preferred over '<tag>', which
	also matches.  If the cursor is on the "t", however, then '<tag>' is
	preferred, because this matches a bit of text containing the cursor.
	If the two groups of patterns were reversed then '<' would never be

	Suppose you
		:let b:match_words = 'if:end if'
 	(Note the space!) and hit "%" with the cursor at the end of "end if".
	Then "if" matches, which is probably not what you want, but if the
	cursor starts on the "end " then "end if" is chosen.  (You can avoid
	this problem by using a more complicated pattern.)

If there is no match, the cursor does not move.  (Before version 1.13 of the
script, it would fall back on the usual behavior of |%|).  If debugging is
turned on, the matched bit of text is saved as |b:match_match| and the cursor
column of the start of the match is saved as |b:match_col|.

Next, the script looks through |b:match_words| (original and parsed versions)
for the group and pattern that match.  If debugging is turned on, the group is
saved as |b:match_ini| (the first pattern) and |b:match_tail| (the rest).  If
there are backrefs then, in addition, the matching pattern is saved as
|b:match_word| and a table of translations is saved as |b:match_table|.  If
there are backrefs, these are determined from the matching pattern and
|b:match_match| and substituted into each pattern in the matching group.

The script decides whether to search forwards or backwards and chooses
arguments for the |searchpair()| function.  Then, the cursor is moved to the
start of the match, and |searchpair()| is called.  By default, matching
structures inside strings and comments are ignored.  This can be changed by
setting |b:match_skip|.


2. Activation						*matchit-activate*

For a new language, you can add a line such as
	let b:match_words = '\<foo\>:\<bar\>'
to the corresponding |filetype-plugin|.  See |b:match_words| below for how
this variable is interpreted.


3. Configuration					*matchit-configure*

There are several variables that govern the behavior of matchit.vim.  Note
that these are variables local to the buffer, not options, so use |:let| to
define them, not |:set|.  Some of these variables have values that matter; for
others, it only matters whether the variable has been defined.  All of these
can be defined in the |filetype-plugin| or autocommand that defines
|b:match_words| or "on the fly."

The main variable is |b:match_words|.  It is described in the section below on
supporting a new language.

				*MatchError* *matchit-hl* *matchit-highlight*
MatchError is the highlight group for error messages from the script.  By
default, it is linked to WarningMsg.  If you do not want to be bothered by
error messages, you can define this to be something invisible.  For example,
if you use the GUI version of Vim and your command line is normally white, you
can do
	:hi MatchError guifg=white guibg=white

If you
	:let b:match_ignorecase = 1
then matchit.vim acts as if 'ignorecase' is set: for example, "end" and "END"
are equivalent.  If you
	:let b:match_ignorecase = 0
then matchit.vim treats "end" and "END" differently.  (There will be no
b:match_infercase option unless someone requests it.)

Define b:match_debug if you want debugging information to be saved.  See
|matchit-debug|, below.

If b:match_skip is defined, it is passed as the skip argument to
|searchpair()|.  This controls when matching structures are skipped, or
ignored.  By default, they are ignored inside comments and strings, as
determined by the |syntax| mechanism.  (If syntax highlighting is turned off,
nothing is skipped.)  You can set b:match_skip to a string, which evaluates to
a non-zero, numerical value if the match is to be skipped or zero if the match
should not be skipped.  In addition, the following special values are
supported by matchit.vim:
	s:foo becomes (current syntax item) =~ foo
	S:foo becomes (current syntax item) !~ foo
	r:foo becomes (line before cursor) =~ foo
	R:foo becomes (line before cursor) !~ foo
(The "s" is meant to suggest "syntax", and the "r" is meant to suggest
"regular expression".)


	You can get the default behavior with
		:let b:match_skip = 's:comment\|string'
	If you want to skip matching structures unless they are at the start
	of the line (ignoring whitespace) then you can
		:let b:match_skip = 'R:^\s*'
 	Do not do this if strings or comments can span several lines, since
	the normal syntax checking will not be done if you set b:match_skip.

	In LaTeX, since "%" is used as the comment character, you can
		:let b:match_skip = 'r:%'
 	Unfortunately, this will skip anything after "\%", an escaped "%".  To
	allow for this, and also "\\%" (an escaped backslash followed by the
	comment character) you can
		:let b:match_skip = 'r:\(^\|[^\\]\)\(\\\\\)*%'
	See the $VIMRUNTIME/ftplugin/vim.vim for an example that uses both
	syntax and a regular expression.


4. Supporting a New Language				*matchit-newlang*

In order for matchit.vim to support a new language, you must define a suitable
pattern for |b:match_words|.  You may also want to set some of the
|matchit-configure| variables, as described above.  If your language has a
complicated syntax, or many keywords, you will need to know something about
Vim's |regular-expression|s.

The format for |b:match_words| is similar to that of the 'matchpairs' option:
it is a comma (,)-separated list of groups; each group is a colon(:)-separated
list of patterns (regular expressions).  Commas and backslashes that are part
of a pattern should be escaped with backslashes ('\:' and '\,').  It is OK to
have only one group; the effect is undefined if a group has only one pattern.
A simple example is
	:let b:match_words = '\<if\>:\<endif\>,'
		\ . '\<while\>:\<continue\>:\<break\>:\<endwhile\>'
(In Vim regular expressions, |/\<| and |/\>| denote word boundaries.  Thus "if"
matches the end of "endif" but "\<if\>" does not.)  Then banging on the "%"
key will bounce the cursor between "if" and the matching "endif"; and from
"while" to any matching "continue" or "break", then to the matching "endwhile"
and back to the "while".  It is almost always easier to use |literal-string|s
(single quotes) as above:  '\<if\>' rather than "\\<if\\>" and so on.

Exception:  If the ":" character does not appear in b:match_words, then it is
treated as an expression to be evaluated.  For example,
	:let b:match_words = 'GetMatchWords()'
allows you to define a function.  This can return a different string depending
on the current syntax, for example.

Once you have defined the appropriate value of |b:match_words|, you will
probably want to have this set automatically each time you edit the
appropriate file type.  The recommended way to do this is by adding the
definition to a |filetype-plugin| file.

Tips: Be careful that your initial pattern does not match your final pattern.
See the example above for the use of word-boundary expressions.  It is usually
better to use ".\{-}" (as many as necessary) instead of ".*" (as many as
possible).  See |/\{-|.  For example, in the string "<tag>label</tag>", "<.*>"
matches the whole string whereas "<.\{-}>" and "<[^>]*>" match "<tag>" and

				*matchit-spaces* *matchit-s:notend*
If "if" is to be paired with "end if" (Note the space!) then word boundaries
are not enough.  Instead, define a regular expression s:notend that will match
anything but "end" and use it as follows:
	:let s:notend = '\%(\<end\s\+\)\@<!'
	:let b:match_words = s:notend . '\<if\>:\<end\s\+if\>'

This is a simplified version of what is done for Ada.  The s:notend is a
|script-variable|.  Similarly, you may want to define a start-of-line regular
	:let s:sol = '\%(^\|;\)\s*'
if keywords are only recognized after the start of a line or after a
semicolon (;), with optional white space.

					*matchit-backref* *matchit-\1*
In any group, the expressions `\1`, `\2`, ..., `\9` (see |/\1|) refer to parts of the
INITIAL pattern enclosed in escaped parentheses.  These are referred to as
back references, or backrefs.  For example,
	:let b:match_words = '\<b\(o\+\)\>:\(h\)\1\>'
means that "bo" pairs with "ho" and "boo" pairs with "hoo" and so on.  Note
that "\1" does not refer to the "\(h\)" in this example.  If you have
"\(nested \(parentheses\)\) then "\d" refers to the d-th "\(" and everything
up to and including the matching "\)":  in "\(nested\(parentheses\)\)", "\1"
refers to everything and "\2" refers to "\(parentheses\)".  If you use a
variable such as `s:notend` or `s:sol` in the previous paragraph then remember
to count any "\(" patterns in this variable.  You do not have to count groups
defined by |/\%(\)|.

It should be possible to resolve back references from any pattern in the
group.  For example,
	:let b:match_words = '\(foo\)\(bar\):more\1:and\2:end\1\2'
would not work because "\2" cannot be determined from "morefoo" and "\1"
cannot be determined from "andbar".  On the other hand,
	:let b:match_words = '\(\(foo\)\(bar\)\):\3\2:end\1'
should work (and have the same effect as "foobar:barfoo:endfoobar"), although
this has not been thoroughly tested.

You can use |/zero-width| patterns such as |/\@<=| and |/\zs|.  (The latter has
not been thouroughly tested in matchit.vim.)  For example, if the keyword "if"
must occur at the start of the line, with optional white space, you might use
the pattern "\(^\s*\)\@<=if" so that the cursor will end on the "i" instead of
at the start of the line.  For another example, if HTML had only one tag then
one could
	:let b:match_words = '<:>,<\@<=tag>:<\@<=/tag>'
so that "%" can bounce between matching "<" and ">" pairs or (starting on
"tag" or "/tag") between matching tags.  Without the |/\@<=|, the script would
bounce from "tag" to the "<" in "</tag>", and another "%" would not take you
back to where you started.

DEBUGGING				*matchit-debug* *:MatchDebug*

If you are having trouble figuring out the appropriate definition of
|b:match_words| then you can take advantage of the same information I use when
debugging the script.  This is especially true if you are not sure whether
your patterns or my script are at fault!  To make this more convenient, I have
made the command :MatchDebug, which defines the variable |b:match_debug| and
creates a Matchit menu.  This menu makes it convenient to check the values of
the variables described below.  You will probably also want to read
|matchit-details| above.

Defining the variable |b:match_debug| causes the script to set the following
variables, each time you hit the "%" key.  Several of these are only defined
if |b:match_words| includes backrefs.

The b:match_pat variable is set to |b:match_words| with backrefs parsed.

The b:match_match variable is set to the bit of text that is recognized as a

The b:match_col variable is set to the cursor column of the start of the
matching text.

The b:match_wholeBR variable is set to the comma-separated group of patterns
that matches, with backrefs unparsed.

The b:match_iniBR variable is set to the first pattern in |b:match_wholeBR|.

The b:match_ini variable is set to the first pattern in |b:match_wholeBR|,
with backrefs resolved from |b:match_match|.

The b:match_tail variable is set to the remaining patterns in
|b:match_wholeBR|, with backrefs resolved from |b:match_match|.

The b:match_word variable is set to the pattern from |b:match_wholeBR| that
matches |b:match_match|.

The back reference '\'.d refers to the same thing as '\'.b:match_table[d] in


5. Known Bugs and Limitations				*matchit-bugs*

The various |:vmap|s defined in the script (%, |g%|, |[%|, |]%|, |a%|) may
have undesired effects in Select mode |Select-mode-mapping|.  At least, if you
want to replace the selection with any character in "ag%[]" there will be a
pause of |'updatetime'| first. E.g., "yV%" would normally work linewise, but
the plugin mapping makes it characterwise.

It would be nice if "\0" were recognized as the entire pattern.  That is, it
would be nice if "foo:\end\0" had the same effect as "\(foo\):\end\1".

top - main help file