Patch: gcj and rhug build regression

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:
 > This patch fixes a build regression in rhug.
 > 
 > Before this patch, some code in Tomcat fails to compile.  The problem
 > is that the bytecode back end already knows that `foo = bar', in an
 > expression context, should leave the value on the stack.  It doesn't
 > correctly handle the trees generated by this code in patch_assignment
 > -- code that isn't required in this case anyway.
 > 
 > Tested on x86 Red Hat Linux 7.3.  I rebuilt libgcj with this, ran the
 > test suite, and also rebuilt rhug (still doesn't build, but at least
 > we get further).
 > 
 > Ok for 3.3 and trunk?

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

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

Andrew.