Re: Compute flags for target in configure, not Makefile

[Jason Merrill <jason at redhat dot com>]

>>>>> Alexandre Oliva <aoliva@redhat.com> writes:

 > --=-=-=
 > On Jul 24, 2000, Jason Merrill <jason@redhat.com> wrote:

 >>>>>>> Alexandre Oliva <aoliva@redhat.com> writes:
 >>> On Jul 22, 2000, Jason Merrill <jason@redhat.com> wrote:
 >>>> This seems wrong to me; the $(build_tooldir)/lib and
 >>>> $(build_tooldir)/include stuff should be used iff we aren't using
 >>>> newlib, but your patch causes it to always be used.

 >>> I think they should always be used, so that we get consistent
 >>> behavior.  Otherwise we'll have different behavior depending on
 >>> whether libsubdir exists or not.

 >> OK, I suppose, though they shouldn't ever find anything there.

 > Why?  If you've already got any headers installed there, you'd
 > probably like to use them, especially in case you're doing a glibc
 > build, not a newlib build.

That's why I mentioned newlib above, but OK.

 >>>> Why can't the build_tooldir stuff all be part of FLAGS_FOR_TARGET?

 >>> Of course it can, I just don't think it should.

 >> Why not?

 > Because it should only be used when we're building GCC.

I think you mean "when we're using gcc from the build directory"?  Yes, that
makes sense.  But in the past, all of this stuff was only used when we were
using gcc from the build directory; otherwise, we would just use $(CC) or
foo-bar-gcc.  Now I see that your patch changes that.

I suppose it might make sense to, say, build and install gcc and the
binutils, then build and install the various target libraries, in which
case we would need the handling for newlib et al to be separate for the
handling of gcc.  Is that what you had in mind?

A couple more nits:

-B$$r/gcc/ should come at the end, after the newlib stuff.

Testing for --with-gnu-ld is wrong; we can get that flag without having ld
in the source tree.

Jason