Match a string between two or more words regardless of order

Question

I need a regular expression that matches words regardless of order. As an example, these lines should match with the marked range,

A longword1 B longword2 C
  ^-------------------^

A longword2 B longword1 C
  ^-------------------^

while these shouldn't:

A longword1 B longword1 C
A longword2 B longword2 C
A longword1 B
A longword2 C

(A, B, C are fillers, they can be essentially any text)

It is possible to just use alternations, such as: \b((longword1).*?(longword2)|(longword2).*?(longword2))\b. But the regex would grow factorially, i.e. three words would need 3! alternates. It's also possible to use subroutines, e.g. \b((?'A'longword1).*?(?'B'longword2')|(?P>B).*?(?P>A))\b. Although shorter, I would still need to include all of its permutations.

Now I've read this post and this other one, but the accepted answers don't exactly solve my problem. Using \b(?=.*longword1)(?=.*longword2).*\b would match the whole line instead of the range I've shown.

I understand, that it would be much easier if I checked the sentence against a list of words. But my current use case prevents it from being possible; I can only use regexes.

Here are some links to demonstrate what I meant:

EXPECTED:

Using alternates: https://regexr.com/5b6pv
Using subroutines: https://regexr.com/5b6ss

INCORRECT:

Using positive lookahead (as linked): https://regexr.com/5b6q2

Are there any simpler regex(es) to tackle this?

@Barmar I can't find the solution you mention. What do you mean? — Wiktor Stribiżew, Sep 01 '20 at 08:19
@Barmar But it matches from `A` to `C`. OP only needs to match a substring from word1 to word2. `A` and `C` are just fillers, they denote any text. And the word 1 and word 2 can be swapped. And there can be more than two words. — Wiktor Stribiżew, Sep 01 '20 at 08:26

Wiktor Stribiżew · Accepted Answer · 2020-09-01T08:14:36.363

4

You may use a backreference + a subroutine:

\b(longword1|longword2)\b.*?\b(?!\1\b)(?1)\b

Expanding it for three alternatives:

\b(longword1|longword2|longword3)\b.*?\b(?!\1\b)((?1))\b.*?\b(?!(?:\1|\2)\b)(?1)\b

See the regex demo and this regex demo, too. So, the list of words will be in Group 1, and you will only need to add backreferences before the subsequent subroutines.

Details

\b(longword1|longword2)\b - a whole word, either longword1 or longword2
.*? - any 0 or more chars other than line break chars, as few as possible
\b - a word boundary
(?!\1\b) - there cannot be the same text as matched in Group 1 followed with a word boundary
(?1) - a subroutine that matches the same pattern as in Group 1
\b - a word boundary

edited Sep 01 '20 at 08:14

answered Sep 01 '20 at 08:07

Wiktor Stribiżew

607,720
39
448
563

Just curious, will this syntax heavily impact the performance of the regex? – Nicky Logan Sep 01 '20 at 09:03
@NickyLogan Sure it will. But alternation would cause even more backtracking. – Wiktor Stribiżew Sep 01 '20 at 09:08
1

Great! I thought I know regexes pretty good but when I stuck on a task like this and found your solution, I understood I don't. Thanks! – Konstantin Nov 13 '21 at 15:12

score 0 · Answer 2 · answered Dec 29 '22 at 16:02

If you are using Perl regular expressions (PCRE), you can make use of the \K operator to help with lookaround to get what you need. For example: (longword1)(longword2)|.*\K(?1).*(?2)|.*\K(?2).*(?1)

Here is an demo for you: https://regex101.com/r/Qo00Hk/1

Match a string between two or more words regardless of order

2 Answers2