[patch] Don't produce stubs on darwin x86

[Rafael Espíndola]

> As you say, it's the linker version that matters (the very old tools require the stubs).
>
> However, I suspect that the reason the target system version is being used as a work-around is that "-target-linker-version" does not (yet) appear to be passed into the target creation process, so the information is not available at the point needed.
>
> [AFAICT -target-linker-version gets to compiler invocation and is then discarded (and I don't see it anywhere in llc)].

I think that is correct. If we support such an old linker at all, the
existing logic is not a too bad heuristic given the limitations you
mentioned.

> ---
>
> So the questions are
>  do we want to be able to:
> (a) self-host on earlier versions of OSX? (i.e. darwin8)
> (b) build objects that can be linked by the native toolset on darwin8?
>
> if either answer is "yes" then the stubs creation code would still be needed.
>
> Also, really, the linker version should be passed through to the target - it's the best way of making the correct decision here.  Not sure whether it would be best to parse it into a version number (which is easier to test) or just leave it as a string (which is more flexible).

I agree with the statement. I am not sure what the answer is regarding
darwin8. This patch is basically an alternative to

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20131125/197042.html

We should fix the stub creation or remove it :-)

Cheers,
Rafael