Norihiro Tanaka wrote:
> Perhaps, I may run accross it in order to use old GCC (4.1.2 on CentOS
> 5.10).

Thanks, it's lucky you did, so that we won't be inundated by other 
people reporting similar problems.  I see now that we were passing 
-Wno-pointer-sign to GCC, and that this suppressed useful diagnostics on 
newer GCC instances (but not on your older one).  I installed the 
attached patch to fix both the signedness problem, and the 
diagnostic-suppression problem.