Dear folks,

i think the patch Izumi Tsutsui provided is correct. I stumbled on this 
same problem today with NetBSD/sparc 4.0_beta (!) so it has been dormant 
for quite some time. Propably gcc4 lossage/gain.

Only point of improval could be the usage of `uint8_t' instead of `char' 
since that has an implicit assumption that sizeof(char)=1. Not that i see 
that change in the future, but i'm just one of the folks who like stricter 
typing in such matters ;)

With regards,
Reinoud