So far, I have figured out a workaround: --disable-bootstrap lets me get a working compiler. Yet I would like to check whether this recipe follows the normal "gcc inside a sysroot" setup. It's a Linux-to-Linux cross-compiler (well, not really a cross-compiler as "host" and "target" are almost the same). On Tue, Mar 9, 2021 at 8:24 AM Oleg Smolsky <osmolsky@xxxxxxxxxxxx> wrote: > Hello, I've built glibc 2.33 with a stand-alone version of GCC10 this way: > > PATH=$gcc/bin:$PATH ../configure \ > --prefix=/usr --libdir=/usr/lib \ > --host=x86_64-unknown-linux --enable-kernel=4.15.1 \ > --with-headers=$sysroot/usr/include > > make -k install-headers cross_compiling=yes install_root=$sysroot > make > make install_root=$sysroot install > > Now I'm trying to build GCC10 against that glibc/sysroot: > > linker="-Wl,--dynamic-linker,$sysroot/lib64/ld-linux-x86-64.so.2" > rpath=",--rpath,$sysroot/lib64" > > LDFLAGS="$linker$rpath" ../configure \ > --prefix=$destination \ > --enable-languages=c,c++ \ > --enable-gold=default \ > --disable-multilib \ > --with-sysroot=$sysroot > > Things progress for a while... until I hit an issue with > `build/genmatch`. I see two executables in the tree: the first has the > right glibc baked in with rpath, and the second does not: > > $ ldd ./gcc-build/prev-gcc/build/genmatch > linux-vdso.so.1 => (0x00007fffe51d1000) > libc.so.6 => /opt/sysroot/lib64/libc.so.6 (0x00007f88bfa10000) > /opt/sysroot/lib64/ld-linux-x86-64.so.2 => > /lib64/ld-linux-x86-64.so.2 (0x00007f88bfdd3000) > > $ ldd ./gcc-build/gcc/build/genmatch > ./gcc-build/gcc/build/genmatch: /lib/x86_64-linux-gnu/libc.so.6: version > `GLIBC_2.33' not found (required by ./gcc-build/gcc/build/genmatch) > linux-vdso.so.1 => (0x00007ffdc8bf1000) > libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007f222fa33000) > libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f222f669000) > /lib64/ld-linux-x86-64.so.2 (0x00007f222fd3c000) > > The second one breaks the build. I can only guess that the LDFLAGS value > I'm trying to squeeze through the entire build process got lost > somewhere... > > Could someone give me a pointer as to how this kind of sysroot > arrangement is supposed to function please? This recipe worked with > older versions of GCC and GLibc... but that does not help much. > > Thanks in advance, > Oleg. > >
