Re: opcodes: more build fixes

[Alan Modra <amodra at bigpond dot net dot au>]

On Mon, Aug 24, 2009 at 11:41:11PM +0200, Ralf Wildenhues wrote:
> * Alan Modra wrote on Mon, Aug 24, 2009 at 10:46:51AM CEST:
> > I think it's a bug that CC_FOR_BUILD isn't used here.
> 
> Well, but in that case there are more issues than just CC_FOR_BUILD.
> Using $(LINK) is wrong, too.  The inclusion of "config.h" from the
> current directory is wrong, as that will have settings for the host
> compiler $CC, not for $CC_FOR_BUILD.  LIBIBERTY is wrong, that should be
> BUILD_LIBIBERTY, and for that we'd need more machinery to get its value
> right.

I'd forgotten about the libiberty problem.  Please don't think my
comment that "it's a bug" means that I think you ought to fix it!

> And then there is DaveK's posted mail (without patch) to add LIBINTL,
> which would also be wrong in cross compile mode, because it would have
> to be BUILD_LIBINTL, and that would require even more other build
> machinery changes.  (BTW, maybe the easiest way out is to just not use
> i18n in generator binaries, like gcc/ does.)

Agreed.  We could also assume there are build libraries already
available on the build machine.  So, for example, just use -liberty
for the generator binaries when cross compiling.

> I don't mind whether this is applied now, or postponed until a more
> complete solution is available, but to me it looks like at least it
> shouldn't regress anywhere.

The patch is OK.

-- 
Alan Modra
Australia Development Lab, IBM