exclude character classes inside [] in a regex

Question

I want to write a regex that finds words contained in brackets, like [word]. The words are not allowed to contain spaces or the characters [ or ].

I tried with

\[[^\[\]\s]\+]

But applied to the string

'foo [first] bar [second] baz [third] and [fourth]'

it only finds [third] and [fourth].

I don't understand why. What is wrong with the regex? It is not totally wrong as it still finds some matches. But not all I would expect.

EDIT: Ok, I found it. \s is not working as expected. I intended to use it as a character class, but it is obiously interpreted as the separate characters \ and s. As first and second both contain the s they don't match.

Now: How do I match character classes inside a [] construct? (I updated the title accordingly.)

EDIT2: Regex in Vi: How does one use character classes inside groups? (e.g. [\s\w]) basically askes the same. But there is no answer about how to exclude character classes.

Answer 1

Character classes such as \s are not acceptable inside a [...] group (with or without the ^, but character class expressions such as [:space:] actually are. So you can use the following:

/\[[^\[\][:space:]]\+]

This will correctly match [first] and [second], besides [third] and [fourth] in your original regular expression.

Note that there's a small difference between \s and [:space:]. While the former only matches space and tab, the latter matches all whitespace characters: space, tab, CR, NL, vertical tab and form feed. For most cases (including this one), that doesn't really affect the end result. (In fact, you could argue you get a better regex from [:space:], since it matches more characters that should qualify as whitespace.)

There are similar [:...:] sequences for other corresponding matches, for example [:keyword:] which is the same as \k. But some, such as \w, don't have a direct counterpart, but it's actually just [:alnum:] plus underscore, so [[:alnum:]_] should do it, or [^[:alnum:]_] to match \W.