Hi Ralf,
An excellent patch! I measure a speedup of "msgmerge af.po coreutils.pot"
from 376 sec. to 152 sec.
I changed your patch a bit
1. to avoid new function calls - more inlining.
2. to include a formal proof of the upper bound.
3. To leave the modified fstrcmp function the freedom what
* lib/fstrcmp.c (fstrcmp_internal): Static rename from fstrcmp,
also take string lengths as arguments.
(fstrcmp): New, thin public wrapper around fstrcmp_internal.
(fstrcmp_upper_bound): New function, for fast fstrcmp estimate.
(fstrcmp_if_higher): New function, to compute fstrcmp value only
if it
