On 26 November 2014 at 23:19, Steve Ellcey wrote:
> I have a question about building multiarch GCC toolchains and was hoping
> someone could help me understand how toolchains that are currently using
> multiarch setups are being built. This may involve glibc as much or more
> then gcc but I thought I would start here.
>
> Using gcc/config/i386/t-linux64 as an example, it has:
>
> MULTILIB_OSDIRNAMES = m64=../lib64$(call if_multiarch,:x86_64-linux-gnu)
> MULTILIB_OSDIRNAMES+= m32=$(if $(wildcard $(shell echo $(SYSTEM_HEADER_DIR))/../../usr/lib32),../lib32,../lib)$(call if_multiarch,:i386-linux-gnu)
> MULTILIB_OSDIRNAMES+= mx32=../libx32$(call if_multiarch,:x86_64-linux-gnux32)
>
> So if we don't build with multiarch enabled GCC is going to search:
>
> ../lib64 (when compiling with -m64)
> ../lib32 or ../lib (when compiling with -m32)
> ../libx32 (when compiling with -mx32)
>
> If we do build with multiarch enabled then I am guessing that we would
> search:
>
> ../lib64/x86_64-linux-gnu
> ../lib32/i386-linux-gnu (or ../lib/i386-linux-gnu)
> ../libx32/x86_64-linux-gnux32
>
> So my first question is: Is that correct? My second question is:
> Where did the '/' between 'lib64' and 'x86_64-linux-gnu' come from?
> It looks to me like the 64 bit directory name would get created as
> '../lib64x86_64-linux-gnu'.
The makefile adds a colon, not a slash, which is documented in
gcc/doc/fragments.texi:
For configurations that support both multilib and multiarch,
@code{MULTILIB_OSDIRNAMES} also encodes the multiarch name, thus
subsuming @code{MULTIARCH_DIRNAME}. The multiarch name is appended to
each directory name, separated by a colon (e.g.
@samp{../lib32:i386-linux-gnu}).
Presumably the gcc driver then interprets the colon at run-time to
produce the actual search dirs, but I'm guessing now.
