Stefan Sperling:

> Thank you! Here is a new diff with these problems fixed.
> 
> Your fixes don't make a huge difference but they do cause 'malloc'
> and 'fseek' calls reported by gprof to drop down by about 2% on a
> trivial test repository.

ok naddy@


I have a further suggestion, though:

        while (dt->blocks[i].len != 0) {
                if (r == 0) {
			... read block into buf ...
                }
                if (len == dt->blocks[i].len && h == dt->blocks[i].hash) {
			... read other block into buf2 and compare ...
                }
                i = (i + 1) % dt->nalloc;
        }

There's no point in reading a block until we know that we need to
compare it.  The "if (r == 0)" part should move under the following
if.

                if (len == dt->blocks[i].len && h == dt->blocks[i].hash) {
                        if (r == 0) {
                                ... read block into buf ...
                        }
                        ... read other block into buf2 and compare ...
                }

-- 
Christian "naddy" Weisgerber                          naddy@mips.inka.de