Someone who understands the compiler might want to take a look at
the PR lib/14485 - I suspect that the underlying problem is more
likely with the compiler than with all the code I have been submitting
patches for.

That is, given

	u_long var;

and the code (fragment)

	var & 0xFF000000

(the value of the const doesn't matter, though having the 0x80000000 bit set
might) the current toolchain, when compiling for an i386 target on an alpha
host, generates warnings about extra bits being set to 0 (and warnings abort
the compile by default of course).

Explicitly making the constant be u_long avoids the problem, but that
shouldn't really be necessary - and doesn't seem to be required when
compiling the same source native on the alpha (at least, no-one has complained
that I know of, I'm yet to actually try that ... is the alpha one of th
ports using the new toolchain??).

kre