Interesting. I would have expected NetBSD/alpha to manifest the same
   symptoms, but a quick check shows that time_t is 4 bytes there. That'll
   teach me that all LP64 platforms are by no means created equal ;) In any
   case, thanks to ben@ for fixing this, and to you for pointing out that he
   had done so. Screen is working just fine now.

IIRC, the variable in question is a "long".  "long" on alpha is 8
bytes, but if the value fits inside a (4 byte) int, looking at the
long as an int "works" on little-endian systems.  doing it on big
endian systems means you get the top half of the long, which is
going to be zero likely.