rustc: Pass -dead_strip on OSX

[alexcrichton]

This flag to the linker asks it to strip away all dead code during linking, as
well as dead data. This reduces the size of hello world from 1.7MB to 458K on my
system (70% reduction).

I have not seen this impact link times negatively, and I have seen this pass
'make check' successfully. I am slightly wary of adding this option, but the
benefits are so huge tha I think we should work hard to work around any issues
rather than avoid using the flag entirely.

[thestinger]

Can you add --gc-sections for the GNU-style linkers too? If these strip out metadata it indicates a bug in Rust (not marking it as used) which should be easy to fix.

[alexcrichton]

We gain no benefit from --gc-sections currently. I'm investigating what sort of improvement we can get from this.

We can work around stripping out metadata (or we could only run it on executables). I think the benefits will be worth the cost of working around the flag (if we get some big benefits)

[thestinger]

Passing --gc-sections strips out a bit of data for me, but it's not going to have the full effect without forcing all of the functions to be output in individual sections, which comes with a cost. The section per function part could be a command-line switch, but there's no reason to leave in the dead sections by default, at least when optimizing.

[alexcrichton]

I have separate analysis for -ffunction-sections in #13833. Adding that is less clear cut to me. This is clear cut in that I can't find a downside to running -dead_strip.

[thestinger]

I don't think we should enable -ffunction-sections by default because splitting into sections has a performance cost, but I do think --gc-sections should be there.

[alexcrichton]

I think you may have meant to comment on the other PR, this has nothing to do with -ffunction-sections or --gc-sections.