multiarch proposal

[Goswin von Brederlow]

Mike Frysinger <vapier@gentoo.org> writes:

> On Saturday 03 May 2008, Daniel Jacobowitz wrote:
>> On Sat, May 03, 2008 at 06:40:47PM +0200, Goswin von Brederlow wrote:
>> > It is not a problem per se but it is a lot of work all together.
>> >
>> > Is not including a trivial patch worth breaking ld just to force
>> > people to use gcc to link? Why not go all out and remove the linker
>> > scripts alltogether so it won't work without gcc at all? :)
>> >
>> > I would rather have people use gcc to link because it is the right
>> > thing to do instead of forcing them because we will break ld with
>> > multiarch.
>>
>> Linking with ld has been broken for all those years on plenty of
>> platforms.  Just not on x86 Linux.
>
> and will continue to be broken.  i think the mutlilib example i pointed out 
> recently in a different thread is a good one ... an ld that supports a wide 
> range of bfd's cannot automatically select the right bfd based on the input 
> objects.  people who execute ld have to explicitly specify the emulation 
> with -m or it'll fail right off the bat.  that's worse than adding random 
> search paths i'd think ...
> -mike

But the paths are not random. That is the whole point of having the
paths in the ldscripts.

The -m option specifically selects one (group of) linker script from
/usr/lib/ldscripts/ and that has the specific multiarch directories for
that emulation included and only those.

-m elf_i386     will add /usr/lib/i386-linux-gnu
-m elf_x86_64   will add /usr/lib/x86_64-linux-gnu

MfG
        Goswin