Re: [PATCH 3/4] Use pointers for traversing arrays in strstr, strcasestrand memmem.

[Eric Blake <eblake at redhat dot com>]

On 06/28/2012 08:00 AM, Eric Blake wrote:
> On 05/30/2012 03:11 AM, Maxim Kuvyrkov wrote:
>> [PATCH 3/4] Use pointers for traversing arrays in strstr, strcasestr and memmem.
>>
>> --
>> Maxim Kuvyrkov
>> CodeSourcery / Mentor Graphics
>>
>>
>> 	* string/str-two-way.h (two_way_short_needle): Use pointers instead of
>> 	array references.
>> 	* string/strcasestr.c (TOLOWER): Make side-effect safe.
> 
>> +++ b/string/strcasestr.c
>> @@ -36,7 +36,7 @@
>>  #include <stdbool.h>
>>  #include <strings.h>
>>  
>> -#define TOLOWER(Ch) (isupper (Ch) ? tolower (Ch) : (Ch))
>> +#define TOLOWER(Ch) tolower (Ch)
> 
> This optimization may work for glibc, but it violates POSIX.  I intended
> for the C implementation of strcasestr to be copied to other platforms
> (that is, gnulib intends to sync with glibc, and gnulib compiles this
> file as a replacement on other platforms where libc's strcasestr is
> sub-optimal).  For me to ack this part of the patch, you will need to
> use preprocessor guards to validate that you are compiling against a
> version of tolower() that has stronger guarantees about behavior than
> what POSIX states about out-of-range input.

Correcting myself, after re-reading POSIX:

I was thinking of _tolower(), not tolower(), as the function that POSIX
says has a limited input range.  If we accept that the cost of isupper()
is necessary, then we should be using _tolower() for speed.  On the
other hand, using tolower() in isolation should be about the same as
using isupper() to pre-filter whether _tolower() is safe.

So I think this optimization is safe, after all.

-- 
Eric Blake   eblake@redhat.com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

Attachment: signature.asc
Description: OpenPGP digital signature