15

I recently came across the branch specifier in Vim regex builtins. Vim's help section on \& contains this:

A branch is one or more concats, separated by "\&".  It matches the last
concat, but only if all the preceding concats also match at the same
position.  Examples:
      "foobeep\&..." matches "foo" in "foobeep".
      ".*Peter\&.*Bob" matches in a line containing both "Peter" and "Bob"

It's not clear how it is used and what it is used for. A good explanation of what it does and how it is used would be great.

To be clear this is not the & (replace with whole match) used in a substitution, this is the \& used in a pattern.

Example usage: 

/\c\v([^aeiou]&\a){4}

Used to search for 4 consecutive consonants (Taken from vim tips).

Tom
  • 15,798
  • 4
  • 37
  • 48

2 Answers2

18

Explanation:

\& is to \|, what the and operator is to the or operator. Thus, both concats have to match, but only the last will be highlighted.

Example 1:

(The following tests assume :setlocal hlsearch.)

Imagine this string:

foo foobar

Now, /foo will highlight foo in both words. But sometimes you just want to match the foo in foobar. Then you have to use /foobar\&foo.

That's how it works anyway. Is it often used? I haven't seen it more than a few times so far. Most people will probably use zero-width atoms in such simple cases. E.g. the same as in this example could be done via /foo\zebar.

Example 2:

/\c\v([^aeiou]&\a){4}.

\c - ignore case

\v - "very magic" (-> you don't have to escape the & in this case)

(){4} - repeat the same pattern 4 times

[^aeiou] - exclude these characters

\a - alphabetic character

Thus, this, rather confusing, regexp would match xxxx, XXXX, wXyZ or WxYz but not AAAA or xxx1. Putting it in simple terms: Match any string of 4 alphabetic characters that doesn't contain either 'a', 'e', 'i', 'o' or 'u'.

mhinz
  • 3,291
  • 20
  • 20
  • @Sniffer Note: in all cases concat may be replaced with zero-width positive look-ahead (`a\&b\&c` is always `\%(a\)\@=\%(b\)\@=c`, wondering why you did not mention this, only have a few words about zero-width atoms). Look-aheads/look-behinds are more powerful then concats and it makes sense to get used to use only them because when you learn new regexp engine it is much more likely that it would support neither look-aheads nor concats or would support only look-aheads rather then have any support for concats. – ZyX Aug 19 '13 at 14:34
  • Also note that `\zs` used as zero-width is buggy: try searching for `.\zso` with `foo` in buffer and compare the result with `.\@<=o`. Do not know a bug for `\ze` though. – ZyX Aug 19 '13 at 14:39
  • @ZyX Thanks for the clarification. I don't work with VIM and don't know what it supports but mhinz explanation of operator seemed pretty logical so I up-voted his answer. – Ibrahim Najjar Aug 19 '13 at 14:43
  • @ZyX I read `:h /\@=`, which indicates that it works the same, but also that using `\&` is easier. No idea how the new engine is working internally. BTW, `.\zso` matches the first 'o' which is kind of what I expected. Or did I miss something here? – mhinz Aug 19 '13 at 14:49
  • @Sniffer His explanation is completely correct. I just pointed a detail about look-aheads because look-aheads are much more common and you may have already learned them. – ZyX Aug 19 '13 at 14:50
  • @ZyX Yes you are totally correct. I understood what you were trying to say. Thank you. – Ibrahim Najjar Aug 19 '13 at 14:51
  • 1
    @mhinz After some thought I do not think that behavior is a bug. But `.\@<=o` matches *two* o’s, `.\zso`, matches *one* o. This seems to be because one `o` is taken by the previous match and, unlike look-behind, `\zs` is applied after both zero-width and non-zero-width parts matched. And it is impossible to have non-zero-width match at the position already taken by previous match. – ZyX Aug 19 '13 at 14:53
  • @ZyX Ah, I see what you mean. Good hint! – mhinz Aug 19 '13 at 14:54
1

\& can be used to match a line containing two (or more) words in any order. For example,

/.*one\&.*two\&.*three

will find lines containing one, two and three in any order. The .* is necessary because each branch must start matching in the same place.

Note, the last branch is the one that participates in any substitution. For example, applying the following substitution:

s/.*one\&.*two\&.*three/<&>/

on the line

The numbers three, two, and one

results in

<The numbers three>, two, and one

Firstrock
  • 931
  • 8
  • 5