> ((int)val >= 0) yields the correct result, but also causes that
> warning.

a-HA.  Granted, a compiler is entitled to emit as many spurious messages
as it likes, but this seems too clever by half...

> (((int)val & 0x100) == 0) also yields the correct result (for machines
> where char is 8 bits -- i.e. all machines that NetBSD supports), but
> does not cause that warning.  The expression is still always true on
> unsigned-char machines, but gcc isn't sophisticated enough to catch it.
                                        ^^^^^^^^^^^^^
To quote Rob Pike, "The technical term for that is ``brain damaged''."

*NOW* I understand.  "Never mind."

(One suggestion though:  add a comment to the code.)