On Fri, Dec 4, 2009 at 12:49 PM, Jeremy Williams <[email protected]> wrote: > I am seeing some QA errors when building the gcc target package. The two QA > errors are .so symlink in non -dev package (for the gcc package) and GNU_HASH > in > objc libraries. > > Looking into the first issue reveals something puzzling to me. If you look at > gcc-package-target.inc, the do_install operation creates a .so symlink to > libgcc. Looking at the FILES definition for the gcc package (non -dev), it > appears that this symlink is being sucked into the package by the entry > "${gcclibdir}/${TARGET_SYS}/${BINV}/lib*". Am I confused here?
Seems plausible > > I'm not sure what is causing the GNU_HASH issue with the objc libraries... LDFLAGS are not honored. May be you could try to force the --hash-style option into builds of target parts of gcc > > I do have one other question. What is the thought behind the rule that first > looks for the soft-float (nof) version of libgcc and copies that into > base_libdir instead of the hard-float version? This version of libgcc ends up > in > the package libgcc1. When building an sdk with meta-toolchain, libgcc1 ends up > getting installed in the target folder and overwrites the hard-float library > that the gcc-cross-sdk package installs. This causes issues with libstdc++ > which > is built to expect the hard-float version of libgcc when linking qmul and > friends (hard float is version GCC_4.1, soft float is version GCC_4.2). Due to multilib and nof happens to be first. I think we should patch gcc to not generate nof version if not asked for. > > thanks! > > -jeremy > > > _______________________________________________ > Openembedded-devel mailing list > [email protected] > http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel > _______________________________________________ Openembedded-devel mailing list [email protected] http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
