Patch: gcj and rhug build regression

[Tom Tromey]

>>>>> "Andrew" == Andrew Haley <aph@redhat.com> writes:

Andrew> Yes.  Sorry, I never thought of the combination of -C -O.
Andrew> Silly me.

Yeah.  It's valid, but not useful, I think.  I suppose someday
somebody might want to make it useful.

Andrew> I would have used flag-syntax-only instead of
Andrew> flag_emit_class_files, but I don't suppose it matters.  We
Andrew> seem to randomly use one or the other.

I think the bug would still occur with flag-syntax-only, since in this
case that flag would be 0.  I haven't looked at most of the uses of
this flag, though I recall adding one or two to the bytecode front end
to fix problems there that occurred when not generating any output.
(`gcj --syntax-only foo.class' is surprisingly useful; I use it all the
time to verify bytecode.)

Tom