[gentoo-dev] Re: toolchain.eclass and gcc 4.1 snapshots

[R Hill]

Simon Strandman wrote:
> It seems like toolchain.eclass does something wrong when configuring gcc
> 4.1 snapshots. I decided to try gcc 4.1 on my server so I created a
> gcc-4.1.1.20060324 ebuild and defined the SNAPSHOT variable in it
> (current cvs has a lot of bugfixes since the release). This is the way
> I've done it with gcc 4.0 and I never had any problems then. The ebuild
> emerges without problems but it installs files in both
> /usr/lib/gcc/i686-pc-linux-gnu/4.1.1-20060324 and
> /usr/lib/gcc/i686-pc-linux-gnu/4.1.1. So when I try to emerge anything
> it always fails with errors like this:
>
> configure:2239: i686-pc-linux-gnu-gcc -O2 -march=pentium3 -pipe
> -fomit-frame-pointer -D_FORTIFY_SOURCE=1   conftest.c  >&5
> /usr/lib/gcc/i686-pc-linux-gnu/4.1.1/../../../../i686-pc-linux-gnu/bin/ld:
> cannot find -lgcc_s
> collect2: ld returned 1 exit status
>
> Just copying over the files from one dir to the other and then
> symlinking it works around the problems. Any ideas?

This is caused by changes to the build system in 4.1 and GCC's BASE-VER not 
matching portage's ${PV} in snapshot builds.  Most of the system directories are 
set up by portage during configure using ${PV} as part of the dirname. (eg. 
includedir=/usr/<lib>/gcc/<chost>/${PV}/include).  However, libdir and 
libexecdir aren't set by portage (because they generate really strange paths w/ 
--enable-version-specific-runtime-libs in GCC 3.3/3.4) and default to 
/usr/<lib>/gcc/<chost>/BASE-VER/blah.  When ${PV} != BASE-VER, wackiness ensues.

Try this in your ebuild:

src_unpack() {
	toolchain_src_unpack

	echo ${PV/_/-} > "${S}"/gcc/BASE-VER
	echo "" > "${S}"/gcc/DATESTAMP
}

--de.

-- 
gentoo-dev@g.o mailing list