Done.

Raising the limit to 16MB seems about 50% faster.

Since the recommended memory limit is 64MB should we use utf8_encode 
for <= 16MB instead of just 1MB?

Ping?

No, the logic is: if the string is small, OR neither mbstring or iconv 
are available, then try utf8-en/de-code

Shouldn't this be an && operator?

Can you both test this patch, please?
http://lists.horde.org/archives/cvs/Week-of-Mon-20080428/077940.html

mb_convert_encoding is way slower than utf8_encode on my system.  To 
convert a 117 bytes (with 1,000,000 iterations) from ASCII to UTF-8 
takes:

utf8_encode time: 15 seconds
mb_convert_encoding time: 46 seconds

Howerver, mb_convert_encoding() is able to deal with larger strings.   
Eg: 50% of max memory instead of %25 max memory.

Do you have any speed numbers for mbstring vs. utf8-en/decode? If it's 
more efficient with memory and the same or faster for speed, we should 
just use it first, period.

Hello,

I got a problem today with a very big text mail (~4Mo) due to an 
infinite loop in a log deamon. This causes my whole mailbox page to 
crash with an PHP Fatal error:

PHP Fatal error:  Allowed memory size of 33554432 bytes exhausted 
(tried to allocate 16187725 bytes) in 
/usr/share/horde3/lib/Horde/String.php on line 128

I managed to prevent the error with the following hack:

         /* HACK by Elektordi 2008-04-28: Prevent crashs caused by 
very big text mails */
         if(strlen($input)>1048576 && String::extensionExists('mbstring')) {
                 return @mb_convert_encoding($input, $to, $from);
         }

in file String.php and function _convertCharset, just before the line 
/* Use utf8_[en|de]code() if possible. */

I did this because Mbstring seems to allocate much less memory than 
utf8_xxcode functions. But a better test in the "if" is to check the 
free memory instead of using the constant 1048576 (=1024*1024)

Hope this will help you!

Regards,

Elektordi