Re: BUG #14512: Backslashes in LIKE

"David G. Johnston" <david(dot)g(dot)johnston(at)gmail(dot)com> writes:
> On Fri, Mar 17, 2017 at 11:16 AM, Alex Malek <magicagent(at)gmail(dot)com> wrote:
>> Given the current behavior the same query will work or raise on error
>> based on context.

> Recently discussed here:
> https://www.postgresql.org/message-id/flat/20170124172505(dot)1431(dot)56735%40wrigleys(dot)postgresql(dot)org#20170124172505(dot)1431(dot)56735(at)wrigleys(dot)postgresql(dot)org
> In short - preventing a "fails-to-fail" scenario here doesn't seem worthy
> of the effort and run-time cost doing so would entail.

BTW, looking again at the patch I suggested in
https://www.postgresql.org/message-id/10287.1485286334%40sss.pgh.pa.us
it strikes me that the check logic was unnecessarily stupid. What
we need to check is that there's not an odd number of backslashes at
the end of the pattern. Since we know that backend encodings are
ASCII-safe, the test logic could be changed to scan backwards,
something like

p += plen;
nbackslash = 0;
while (plen-- > 0)
{
if (*(--p) == '\\')
nbackslash++;
else
break;
}
if (nbackslash & 1)
ereport(ERROR, ...);

For patterns of practical interest this would be of small and nearly
constant cost. Maybe it is worth doing, especially since we've now
had two independent complaints about it.

regards, tom lane