> From lfs-dev-boun...@linuxfromscratch.org Wed Feb 26 20:04:08 2014
> Date: Wed, 26 Feb 2014 14:00:08 -0600
> From: Bruce Dubbs <bruce.du...@gmail.com>
> To: LFS Developers Mailinglist <lfs-dev@linuxfromscratch.org>
> Subject: Re: [lfs-dev] gcc pass 1/2 instructions re mpfr/gmp/mpc.
>
> >>>>>
> >>>>> Does gcc pass1 & pass2 instructions need '--with-gmp...' (possibly also
> >>>>> '--with-mpc...') configure options (re-)added, or have mpfr/gmp/mpc
> >>>>> listed
> >>>>> in the book's general host-requirements ('prologue/hostreqs.html') ?
> >>>>>
> >>>>>
> >>>>
> >>>> Both gcc pass1 and pass2 extract gmp, mpfr and mpc into the gcc source
> >>>> tarball before starting the build. They get built in the gcc build
> >>>> process that way.
> >>>>
> >>>
> >>>
> >>> Yeah, that was broadly folks' 'carried-along' understanding: but per
> >>> thread
> >>> details, it seems maybe not - and seems that gcc was looking into host-os;
> >>> and if mpc not present in host-os - or moved-aside as a debug-test of
> >>> the issue - then gcc pass build fails; & iirc similarly for at least
> >>> mpfr; & iiuic/iirc that's at least partly why configure gets told those
> >>> '--with-...' details.
> >>
.
.
> >
> > Init problem was caused by OP (Hazel Russman) doing a non-'full' install
> > and instead selecting packages, and omitted gmp. But then went back and
> > did full install and could reproduce the problem: ref Hazel's email (on
> > lfs-support) of Tues 17 Dec - the one where she gives the '2x2 matrix'
> > of tests+results. Full-install slackware 14.0 64-bit and slackware 14.1
> > 64-bit, satisfy host-sys-req for lfs-7.4 ; haven't checked for lfs-7.5-rc.
>
> I'm not sure I understand. Does that mean that slackware builds gcc
> 4.7.1 without one or more of mpc, gmp, and mpfr?
>
The Slackware installer has separate tarballs for each of gcc, mpc, mpfr,
& gmp; and you can install them independently.
> Reading the message on Dec 17, it seems that the host had mpfr but not
> gmp. I think I'd suggest just deleting or renaming libgmp.la on the
> host. Stupid libtool.
>
Yes, but the main point here is that the lfs instructions _seem_ to perhaps
be not quite rigourous enough, per following chain:
--
* The host-requirements don't say that gmp/mpc/mpfr must be installed
in host-os.
* And if e.g. gmp is not installed in host-os then - in the case investigated
- the gcc pass 1 build in lfs failed.
* And the cause of the failure was traced to gcc looking into host-os for
gmp stuff.
* Whereas if those gcc pass 1 instructions are augmented with
'--with-gmp...' lines just like the '--with-mpfr...' lines, then the
build works.
* Those '--with-gmp...'-type lines were in lfs-6.7/6.8 but then for lfs-7.0
were dropped and replaced with only the '--with-mpfr...' lines.
* Overall, if the instructions in book up to and including gcc pass 1 build
are followed, then build should succeed. Whereas that's not necessarily
true, per the case in question
--
Hope that makes sense.
rgds,
akh
--
--
http://linuxfromscratch.org/mailman/listinfo/lfs-dev
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
