[PATCH] Document arithmetic overflow semantics

[Robert Dewar]

> Including gdb support?
> If Java maps signed types to unsigned, how is gdb going to know about it?

Indeed the debugging information should represent these as signed, and it is
indeed problematic that gdb debugging information in gcc is tied so closely
to the types used for code generation purposes (we have bumped our noses
on this in Ada many times). I think this just goes again to show that the
issue of operation semantics should not depend on the type alone, but rather
on the operation, as was previously suggested in this thread.