Re: FreeBSD 5.0 cannot bootstrap: libmath/stubs.c: fabsl undeclared

[Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>]

In article <20021212201901.A20737@attbi.com> you write:

> I am still using FreeBSD 5.0, but now have downgraded to
> Bison 1.35.  I can get further along in the bootstrap, but
> I get a failure when compiling libstdc++-v3.
> This is from a fresh checkout of the gcc mainline.

Hi Craig,

Sorry to be a pest, but actually you are not using FreeBSD 5.0.  You
are using a prerelease of FreeBSD 5.0.  Depending upon how old it is,
it may not work at all with FSF sources.  We have made changes to the
FSF tree of gcc mainline to track FreeBSD-CURRENT (I catch it whenever
beast/ref5 are upgraded).  Until there is a released FreeBSD 5.0,
there is nothing to really lock against for those rare things that are
not fully autoconf'd.  Now, all that said, your exact error made no
sense to me until I looked at it closely.

>From today's ref5 (which is using Dec 2 sources) build (I only
cut-n-paste the section related to your post):

[...]
/bin/sh ../libtool --tag CC --mode=compile /local0/scratch/ljrittle/gcc-build-re
f5-1212/gcc/xgcc -B/local0/scratch/ljrittle/gcc-build-ref5-1212/gcc/ -B/home/ljr
ittle/tmp-ref5/install-ref5/i386-unknown-freebsd5.0/bin/ -B/home/ljrittle/tmp-re
f5/install-ref5/i386-unknown-freebsd5.0/lib/ -isystem /home/ljrittle/tmp-ref5/in
stall-ref5/i386-unknown-freebsd5.0/include -DHAVE_CONFIG_H -I. -I/home/ljrittle/
tmp-ref5/outside-cvs-src/gcc/libstdc++-v3/libmath -I..     -O2 -g -O2 -c /home/l
jrittle/tmp-ref5/outside-cvs-src/gcc/libstdc++-v3/libmath/stubs.c
mkdir .libs
/local0/scratch/ljrittle/gcc-build-ref5-1212/gcc/xgcc -B/local0/scratch/ljrittle
/gcc-build-ref5-1212/gcc/ -B/home/ljrittle/tmp-ref5/install-ref5/i386-unknown-fr
eebsd5.0/bin/ -B/home/ljrittle/tmp-ref5/install-ref5/i386-unknown-freebsd5.0/lib
/ -isystem /home/ljrittle/tmp-ref5/install-ref5/i386-unknown-freebsd5.0/include 
-DHAVE_CONFIG_H -I. -I/home/ljrittle/tmp-ref5/outside-cvs-src/gcc/libstdc++-v3/l
ibmath -I.. -O2 -g -O2 -c /home/ljrittle/tmp-ref5/outside-cvs-src/gcc/libstdc++-
v3/libmath/stubs.c  -fPIC -DPIC -o .libs/stubs.o
/local0/scratch/ljrittle/gcc-build-ref5-1212/gcc/xgcc -B/local0/scratch/ljrittle
/gcc-build-ref5-1212/gcc/ -B/home/ljrittle/tmp-ref5/install-ref5/i386-unknown-fr
eebsd5.0/bin/ -B/home/ljrittle/tmp-ref5/install-ref5/i386-unknown-freebsd5.0/lib
/ -isystem /home/ljrittle/tmp-ref5/install-ref5/i386-unknown-freebsd5.0/include 
-DHAVE_CONFIG_H -I. -I/home/ljrittle/tmp-ref5/outside-cvs-src/gcc/libstdc++-v3/l
ibmath -I.. -O2 -g -O2 -c /home/ljrittle/tmp-ref5/outside-cvs-src/gcc/libstdc++-
v3/libmath/stubs.c -o stubs.o >/dev/null 2>&1
[...]

The only way I can see behavior close to your report, is to do the
following:

cd i386-unknown-freebsd5.0/libstdc++-v3/libmath
rm stubs.lo
gmake CFLAGS\='-fno-builtin -Wall -Werror'

(this yields an error for every error line of your report; but the
exact text is different.)

I think if you run 'gmake CFLAGS\=-v' in the directory with the
bootstrap failure, you might see something very interesting.  However,
until you try it, I don't know exactly what to expect.  Do you have a
rouge/old spec file in the target prefix path (which you will note is
mentioned in a -B flag --- it seems strictly wrong to be referring to
things outside the bootstrap tree at this point; but whatever)?  Is
some cc1 binary found other than the one just bootstrapped?

Regards,
Loren