[Patch] Another fix for cycle detection

[Roland McGrath]

> I've already seen it overflow 8 bit signed, seems silly to flip to an
> unsigned which doesn't give us that much headroom.  But if that's what
> you'd prefer, I'll do it.

If it has exceeded 127 in real-world cases, then going to 65535 seems
reasonable.  But any real-world case having exceeded the limit indicates
that explicit overflow detection is warranted.

> In case it's not obvious, this patch is pointless without last week's
> fix to cycle detection:
> 
> http://sources.redhat.com/ml/libc-alpha/2012-01/msg00072.html

That's a sufficiently more complex and subtle change that it merits more
careful review than this one needed, and I haven't looked at it in detail.


Thanks,
Roland