Re: strcat+strcat+strcat == baaad

[Morten Welinder]

Unless we're talking an inner and very intensive loop, all this is a
bit silly.

However, assuming that it does make sense, sprintf variants are not
part of the answer.  They're interpreted and var-arg type, both of
which are little good for efficiency.  More like something along
these lines:

char *tmpdst = dst;

strcpy (tmpdst, src1); tmpdst += strlen (src1);
   strcpy (tmpdst, src2); tmpdst += strlen (src2);
   strcpy (tmpdst, src3); tmpdst += strlen (src3);
   ...
   strcpy (tmpdst, srcN); tmpdst += strlen (srcN);
   ...

This works even better when the strlen calls have known results.

Of course, nothing keeps the compiler from noticing a certain pattern
of sprintf calls can producing an equaivalent of the above, but right
now gcc will not do that if I recall correctly.

Morten

strcat+strcat+strcat == baaad

[Francois Gouget]

On 2 Dec 2002, Morten Welinder wrote:
[...]
> Unless we're talking an inner and very intensive loop, all this is a
> bit silly.
>
> However, assuming that it does make sense, sprintf variants are not
> part of the answer.  They're interpreted and var-arg type

Why would vararg be a problem performance-wise?
> More like something along these lines:
>
> char *tmpdst = dst;
>
> strcpy (tmpdst, src1); tmpdst += strlen (src1);

[...]
> This works even better when the strlen calls have known results.

Except usually src* are constant strings and it's a major pain from a
maintainability and readability point of vue to have to duplicate them.
And whenever src* is not a constant, the above is twice as slow as
cpycat because each string is read twice (once by strcpy and another
time by strlen).