Re: BUG #5478: ILIKE operator returns wrong result

Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > Tom Lane wrote:
> >> I have a feeling that this represents still another bug in the
> >> special-case path for % followed by _ (cf bug #4821). If so,
> >> maybe we ought to just toss out that optimization?
>
> > Yea, looks like it is this code in like_match.c:
>
> No, actually it's the bit right after that:
>
> /* Look for a place that matches the rest of the pattern */
> while (tlen > 0)
> {
> int matched = MatchText(t, tlen, p, plen);
>
> if (matched != LIKE_FALSE)
> return matched; /* TRUE or ABORT */
>
> NextChar(t, tlen);
> }
>
> If tlen == 0 when we reach this loop, we'll fall through and fail.
> But that is wrong since we need to consider the possibility that
> the remaining pattern can match a zero-length substring. So the
> loop needs to be changed to attempt a recursive MatchText for
> tlen equal to zero as well as greater than zero.

I took a different approach. I think the problem is that we check for
end of pattern without consuming '%' patterns. I copied that consume
loop from code above that where we also test for end of pattern.

With the attached patch (which includes a regression test addition), it
works fine:

test=> select 'ba' like '%__%';
?column?
----------
t
(1 row)

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com