Le 06/02/2012 00:32, Andrew Benton a écrit : > On Sun, 05 Feb 2012 23:01:30 +0100 > Pierre Labastie<[email protected]> wrote: > >> If I understand the xxx.la files, they are used by libtool to find >> libraries. I am certainly missing something, but I do not understand >> why changing tools to usr in libstdc++.la would give a better >> result. When gmp is built (ie before gcc in chapter 6), there is no >> libstdc++{.a,.so} in /usr/lib. > Yes there is, we: > ln -sv /tools/lib/libstdc++.so{,.6} /usr/lib > http://www.linuxfromscratch.org/lfs/view/development/chapter06/createfiles.html > > We could create a .la file at the same time: > sed 's/tools/usr/' /tools/lib/libstdc++.la> /usr/lib/libstdc++.la > > For me, that results in a /usr/lib/libgmpxx.la that contains: > # Libraries that this one depends upon. > dependency_libs=' /usr/lib/libgmp.la /usr/lib/libstdc++.la' > > Andy OK, sorry for the noise, you are totally right!
Then the discrepancy between two successive iterations of ICA comes from the fact that at iteration 2, it has: dependency_libs=' /usr/lib/libgmp.la /usr/lib/../lib64/libstdc++.la' This shows up in libgmpxx.so.4.2.3 too, because /usr/lib/../lib64/ is added as rpath in the .dynamic section and also in .dynstr section. (I figured out that using objdump, and learning it at the same time, so maybe not totally accurate). I think the case is closed. It would be cleaner if your sed were added to createfiles. Another ICA discovery coming soon... Regards, Pierre -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
