Oracle REGEXP_LIKE and word boundaries

问题:

I am having a problem with matching word boundaries with REGEXP_LIKE. The following query returns a single row, as expected.

select 1 from dual
where regexp_like('DOES TEST WORK HERE','TEST');

But I want to match on word boundaries as well. So, adding the "\b" characters gives this query

select 1 from dual
where regexp_like('DOES TEST WORK HERE','\bTEST\b');

Running this returns zero rows. Any ideas?

回答1:

I believe you want to try

 select 1 from dual 
  where regexp_like ('does test work here', '(^|\s)test(\s|$)');

because the \b does not appear on this list: http://download.oracle.com/docs/cd/B19306_01/appdev.102/b14251/adfns_regexp.htm#i1007670

The \s makes sure that test starts and ends in a whitespace. This is not sufficient, however, since the string test could also appear at the very start or end of the string being matched. Therefore, I use the alternative (indicated by the |) ^ for start of string and $ for end of string.

Update (after 3 years+)... As it happens, I needed this functionality today, and it appears to me, that even better a regular expression is (^|\s|\W)test($|\s|\W) (The missing \b regular expression special character in Oracle).

回答2:

The shortest regex that can check for a whole word in Oracle is

(^|\W)test($|\W)

See the regex demo.

Details

• (^|\W) - a capturing group matching either
  • ^ - start of string
  • | - or
  • \W - a non-word char
• test - a word
• ($|\W) - a capturing group matching either
  • $ - end of string
  • | - or
  • \W - a non-word char.

Note that \W matches any chars but letters, digits and _. If you want to match a word that can appear in between _ (underscores), you need a bit different pattern:

(^|[^[:alnum:]])test($|[^[:alnum:]])

The [^[:alnum:]] negated bracket expression matches any char but alphanumeric chars, and matches _, so, _test_ will be matched with this pattern.

See this regex demo.

回答3:

In general, I would stick with René's solution, the exception being when you need the match to be zero-length. ie You don't want to actually capture the non-word character at the beginning/end.

For example, if our string is test test then (\b)test(\b) will match twice but (^|\s|\W)test($|\s|\W) will only match the first occurrence. At least, that's certainly the case if you try to use regexp_substr.

Example

SELECT regexp_substr('test test', '(^|\s|\W)test($|\s|\W)', 1, 1, 'i'), regexp_substr('test test', '(^|\s|\W)test($|\s|\W)', 1, 2, 'i') FROM dual;

Returns

test |NULL