On Thu, 2004-07-08 at 11:40, Georg Schwarz wrote:
> > Georg Schwarz wrote:
> > > env AR=/usr/local/mips-sgi-irix5.3/bin/AR
> > > NM=/usr/local/mips-sgi-irix5.3/bin/nm
> > > RANLIB=/usr/local/mips-sgi-irix5.3/bin/ranlib CONFIG_SHELL=/bin/ksh
> > > /usr/people/schwarz/gcc-3.4.0/configure --prefix=/usr/local
> > > --with-gnu-as --with-gnu-ld --with-as=/usr/local/mips-sgi-irix5.3/bin/as
> > > --with-ld=/usr/local/mips-sgi-irix5.3/bin/ld
> > > --enable-languages=c,c++,ada

I didn't get around to creating a bugzilla bug report, but I have done
quite a bit more investigation.

When you configure this way, the toplevel Makefile gets the new NM
values written into it.  The gcc Makefile does not.  When you do "make
all-gcc", the toplevel value of NM is passed down to the gcc Makefile,
and it works as expected.  When you do "make bootstrap", the toplevel
value of NM is passed down to the gcc Makefile, and then the gcc
Makefile does a recursive make without passing down any arguments and
this value gets lost.  We do export NM_FOR_TARGET which seems like it
should work, but the NM_FOR_TARGET value that gets exported is the value
defined in the gcc Makefile, not the value that was passed down from the
toplevel.  I don't know why exports are working this way.

Assuming this is the way things are supposed to work in make, then
solving this problem seems to require autoconf tests for AR and NM in
the gcc directory, so that we get the environment variable values
written into the gcc Makefile.  That attached patch does this, and with
this patch I get the expected result for your configure/make commands.

I'm not really a configure/make expert, so if someone could confirm that
this is the right approach I would appreciate that.
-- 
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com