[Rpm-maint] Problems in rpmstrPoolRehash function

[Michael Schroeder]

On Wed, Sep 11, 2013 at 07:09:29PM +0200, Michael Schroeder wrote:
> There are different (and easier) ways to fix this:
> 
> - you can always put an empty string into the pool, it would always
>   have id 1. This simplifies the dummy entry check to:
>       if (i != 1 && str[0] == 0)
> 
> - you could get rid of the dummy entries and remove the
>   rpmstrPoolStrlen() function. It's only used 5 times in the code
>   and calling strlen() on the returned string does not cost much.

Having slept one night I think the best way is really to remove the
dummy entries. You can still have rpmstrPoolStrlen() if you start
each chunk with a non-zero character, e.g.

\377first_string_in_chunk\0...\0last_string_in_chunk\0
    ^                           ^
You can then change rpmPoolStrlen to:

size_t rpmstrPoolStrlen(rpmstrPool pool, rpmsid sid)
{
    size_t slen = 0;
    if (pool && sid > 0 && sid <= pool->offs_size) {
	const char *str = pool->offs[sid];
	const char *nextstr = pool->offs[sid + 1];
	if (nextstr[-1] == 0)
	    slen = nextstr - str - 1;
	else
	    slen = strlen(str);		/* last string in chunk */
    }   
    return slen;
}

So rpmstrPoolStrlen() is as fast as before except for the last sting
in the chunk.

But maybe simply replacing the rpmstrPoolStrlen() calls with (inlined)
strlen() calls is even faster.

Cheers,
  Michael.

-- 
Michael Schroeder                                   mls at suse.de
SUSE LINUX Products GmbH,  GF Jeff Hawn, HRB 16746 AG Nuernberg
main(_){while(_=~getchar())putchar(~_-1/(~(_|32)/13*2-11)*13);}