Re: 3.1 bootstrap fails on sparcv9-sun-solaris2.8

[Rainer Orth <ro at TechFak dot Uni-Bielefeld dot DE>]

Brad Lucier <lucier@math.purdue.edu> writes:

> OK, I tried this with
> 
> ../configure sparcv9-sun-solaris2.8 --enable-checking=no --prefix=/pkgs/gcc-2.96 --enable-languages='c'
> make all
> make install
> 
> which worked, then, with new gcc being the default gcc, doing
> 
> ../configure sparcv9-sun-solaris2.8 --enable-checking=no --prefix=/pkgs/gcc-2.96
>  --enable-languages='c'
> make bootstrap
> 
> which failed.  (The stage1 xgcc segfaulted.)
> 
> But the solaris port has many more problems than this, and I don't think
> that you or I can fix them at this point.

I managed to bootstrap the 3.0 branch (20010402) with

	CC='cc -xarch=v9' configure sparcv9-sun-solaris2.8
	make bootstrap
	make RUNTESTFLAGS='--target_board "unix{-m32,}"' check

You need to apply

	http://gcc.gnu.org/ml/gcc-patches/2001-04/msg00250.html

(still unreviewed) for the -m32 tests to succeed, but with this setup, -m32
and -m64 testsuite failures are almost the same.  I still have to post a
complete analysis of the differences I encountered, but the 64x32 cross
compiler seems to work resonably.

Unfortunately, that's not what I really want: we won't be running 64-bit
Solaris 2.8 on all our machines, so either I need two different
installations of gcc: one sparcv9 for 64-bit (and 32-bit if desired)
compilation, and another sparc for 32-bit compilation.  Just having the
bi-arch 32x64 compiler instead of those (as Sun's cc does) would be much
more convenient.

	Rainer

-- 
-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Email: ro@TechFak.Uni-Bielefeld.DE