On Thu, Feb 17, 2011 at 12:06:30PM +0900, enami tsugutomo wrote:
>
> I just did `more fastgrep.c' and found following piece of code. The
> usage of wflag is an obvious regression from OpenBSD code.
>
> | if (fg->len >= 14 &&
> | strncmp(pat + (fg->bol ? 1 : 0), "[[:<:]]", 7) == 0 &&
> | strncmp(pat + (fg->bol ? 1 : 0) + fg->len - 7, "[[:>:]]", 7) == 0) {
> | fg->len -= 14;
> | /* Word boundary is handled separately in util.c */
> | wflag = true;
> | }
It looks like that might transform:
grep '[[:<:]]foo.*bar[[:>:]]'
to
grep -w 'foo.*bar'
which isn't a valid transform.
(I presume something else translated the original '\<foo.*bar\>' ...)
I've known grep where searches for '\<word\>' were a lot faster than
using -w. Presumably because -w locates the words (start and end)
rather than just the start!
David
--
David Laight: da...@l8s.co.uk
