Hi Pascal.

14-Apr-04 15:50:16, Pascal Schmidt wrote to Darin McBride


 PS> Hi Darin! :-)

 DM>> ;-)  Most modern compilers probably do understand strlen.  For
 DM>> example, I believe Pentiums have a single hardware instruction
 DM>> that performs the strlen function much faster than software can
 DM>> emulate, so you'll end up with something that looks like a
 DM>> function call becoming pure assembly ;->

 PS> Not in one instruction, though that is somewhat true for memcpy
 PS> and memset.

"repz scasb" is technically two instructiuons and is the basic engine
behind an implementation of strlen,

 PS> it has to work on

 PS> For strlen inside the condition of a for loop, the situation is
 PS> more complicated since the code inside the loop can modify the
 PS> string, and really complicated data flow analysis in the compiler
 PS> would be needed to prove otherwise. Especially if the code inside
 PS> the loop calls code in another compilation unit (from another file
 PS> or inside some library)

 PS> strlen depends on side effects (the contents of memory when it is
 PS> called), so multiple calls to it cannot easily be optimized away

 PS> gcc has a special attribute (pure) that one can attach to
 PS> functions that tells gcc that it does not depend on or cause side
 PS> effects. This helps the optimizer and then it *can* optimize away
 PS> multiple calls with the same argument(s)

 JB>>> for (n=0; target[n] != '\0' ; n++) {
 DM>> I prefer this one.  ;-)
 PS> I would write this one as (minor difference only):

 PS> for (n = 0; target[n] != 0; n++) { ... }

the extra " != 0 " ay result in unwanted code but optimising compilers
should be able to cut it out...

 JB>>> and possibly even better re-written to use a char* instead

 JB>>> for( s=target ; *s ; s++ ) {

 DM>> Yes, I like this even more ;-)

 PS> I think that's obfuscated enough to require several seconds of
 PS> thought when looking at it some time after writing it.

Use it 20 times and it'll not be at all hard to understand next time you
encounter it.

 PS> However, I never needed to traverse a string this way in all the C
 PS> programs I have written so far

I think I wrote a simple uudecoder that used that construct.

 PS> Ciao Pascal

 -=> Bye <=-

---
(3:640/1042)