On Fri, 06 Dec 2013 21:05:51 +0000
lf...@cruziero.com (akhiezer) wrote:
> > Just to check: did you use underscores there - i.e. in the above
> > --with_gmp_include=$(pwd)/gmp and --with_gmp_lib=$(pwd)/gmp/.libs
> > ; or did you use hyphens/dashes thus:
> > --with-gmp-include=$(pwd)/gmp --with-gmp-lib=$(pwd)/gmp/.libs
> > ?
> Hazel, did you have any info on the above query re underscores?
I did use hyphens. I based the syntax closely on the --with-mpfr
options.
>
>
> >
> > lfs-6.7 and lfs-6.8 both use './configure ...' options:
> > --with-gmp-include=$(pwd)/gmp --with-gmp-lib=$(pwd)/gmp/.libs
> > ; and they don't use any mpfr/mpc stuff.
> >
> > And lfs-7.0 thru lfs-7.4 uses (I've replaced the gcc version
> > numbers with the literal string '${GCC_VERNUM}'):
> > --with-mpfr-include=$(pwd)/../gcc-${GCC_VERNUM}/mpfr/src \
> > --with-mpfr-lib=$(pwd)/mpfr/src/.libs
> > ; and they don't use any gmp/mpc stuff.
> >
> > Had a brief look at trac to see why the switchover at 6.8 -> 7.0 ,
> > and why wanting/needing to specify those options to
> > './configure ...' explicitly; but nothing obvious in the time spent.
>
>
> rgds,
>
> akh
>
I have now established that gmp was never installed on my host system.
I had the binaries because they come with the elflibs package, but not
the ancillary files. So I installed the library explicitly and tried
again. This time the gcc build went to completion but when I looked in
the .la files afterwards, this is what I found:
In /gcc-build/mpfr/src/libmpfr.la
dependency_libs=' /usr/lib64/libgmp.la'
In /gcc-build/mpc/src/libmpc.la
dependency_libs=' /usr/lib64/libmpfr.la /usr/lib64/libgmp.la -lm'
Now this, to my mind, is the real error. My unorthodox setup showed it
up by crashing the build, but what is the software doing looking
in /usr/lib64 in the first place? I thought the whole point of giving
the compiler a target was to trick it into thinking it was compiling
for a different computer so that it wouldn't look for stuff on the
host system. And it can't be anything to do with an LD_LIBRARY_PATH
variable because the lfs user hasn't got one.
On a full Slackware install, no one would notice this. Who is going to
root around in archive files looking for bad dependency paths when
everything has gone smoothly?
I then tried a second build with freshly unpacked source, using the two
--with-gmp options. Again the build went to completion, but with the
same bad dependency paths. So using these options makes no difference
at all.
I'm just curious to know what is going on here. I hate loose ends.
--
http://linuxfromscratch.org/mailman/listinfo/lfs-support
FAQ: http://www.linuxfromscratch.org/lfs/faq.html
Unsubscribe: See the above information page
