Re: strncmp->memcmp when we know the shorter length

On Tue, Dec 21, 2010 at 8:29 PM, Gurjeet Singh <singh.gurjeet@gmail.com> wrote:
> On Tue, Dec 21, 2010 at 6:24 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>>
>> On Mon, Dec 20, 2010 at 1:10 PM, Noah Misch <noah@leadboat.com> wrote:
>> > When the caller knows the smaller string length, memcmp and strncmp are
>> > functionally equivalent.  Since memcmp need not watch each byte for a
>> > NULL
>> > terminator, it often compares a CPU word at a time for better
>> > performance.  The
>> > attached patch changes use of strncmp to memcmp where we have the length
>> > of the
>> > shorter string.  I was most interested in the varlena.c instances, but I
>> > tried
>> > to find all applicable call sites.  To benchmark it, I used the attached
>> > "bench-texteq.sql".  This patch improved my 5-run average timing of the
>> > SELECT
>> > from 65.8s to 56.9s, a 13% improvement.  I can't think of a case where
>> > the
>> > change should be pessimal.
>>
>> This is a good idea.  I will check this over and commit it.
>
> Doesn't this risk accessing bytes beyond the shorter string?

If it's done properly, I don't see how this would be a risk.

> Look at the
> warning above the StrNCpy(), for example.

If you're talking about this comment:
*      BTW: when you need to copy a non-null-terminated string (like a text*      datum) and add a null, do not do it
withStrNCpy(..., len+1).  That*      might seem to work, but it fetches one byte more than there is in the*      text
object.

...then that's not applicable here.  It's perfectly safe to compare to
strings of length n using an n-byte memcmp().  The bytes being
compared are 0 through n - 1; the terminating null is in byte n, or
else it isn't, but memcmp() certainly isn't going to look at it.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company