https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58638
--- Comment #8 from Daniel Richard G. <skunk at iskunk dot org> ---
(In reply to Jonathan Wakely from comment #7)
> Seems unlikely, but if -static-libgcc -static-libstdc++ gives you a runtime
> dependency on GCC then that should be reported as a bug.
I tried a 4.9.0 build without --disable-shared to jog my memory. The problem
was twofold:
First, GCC can't find its own runtime library when linking programs:
/usr/bin/ld: cannot find -lgcc_s
I can specify -L$(PREFIX)/lib/gcc/x86_64-unknown-linux-gnu/lib64 manually, and
that allows things to link. But then, unless I futz with
ld.so.conf/LD_LIBRARY_PATH, the resulting executable uses the wrong instance of
libgcc/libstdc++:
$ ldd swig
linux-vdso.so.1 => (0x00007fff76839000)
libdl.so.2 => /lib64/libdl.so.2 (0x000000302ee00000)
libstdc++.so.6 => /usr/lib64/libstdc++.so.6 (0x0000003039a00000)
libm.so.6 => /lib64/libm.so.6 (0x000000302e600000)
libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x0000003035a00000)
libc.so.6 => /lib64/libc.so.6 (0x000000302ea00000)
/lib64/ld-linux-x86-64.so.2 (0x000000302e200000)
(Note: It's probably relevant that I also build GCC with
--enable-version-specific-runtime-libs, to avoid a different way that the
runtimes can get mixed up.)
Not having to specify -static-libblah is a bonus, but these were the real
reasons that I was using --disable-shared. Should I file a new bug for this?