Move trans, back, driver, and back into a new crate, rustc_trans.

[nikomatsakis]

Move trans, back, driver, and back into a new crate, rustc_trans (which probably needs a better name). This is not a principled split per se, though it does make some sense, as the things which have been moved all fall into "back end" territory. The primary motivation is reducing memory usage. Measurements with massif suggest that peak memory usage after this change is 1.5GB. In comparison, measurements on master show a range from 1.8 to 2.5 GB of usage.

r? @brson
cc @alexcrichton
cc #18784

[rust-highfive]

Warning

• These commits modify unsafe code. Please review it carefully!

[ghost]

❤️

[eddyb]

Extraneous newline?

[pcwalton]

Might want to update README.txt in librustc now.

[pcwalton]

Also (this can be a followup of course) we might as well rename librustc to librustc_middle at this point. (Or even librustc_sema following clang? I always liked the name "sema" better than "middle", especially since trans was historically part of "middle"…)

[arielb1]

How much does this affect compile times (by reducing inlining)?

[alexcrichton]

This looks amazing @nikomatsakis! I think @pcwalton's right in that rustc doesn't make too too much sense any more now that it's sharded so much, but r=me anyway on this PR.

[nikomatsakis]

On Sun, Nov 16, 2014 at 12:22:59PM -0800, Alex Crichton wrote:

This looks amazing @nikomatsakis! I think @pcwalton's right in that rustc doesn't make too too much sense any more now that it's sharded so much, but r=me anyway on this PR.

Yes, I agree, the names are bad, and I liked his suggestions. I'm
inclined to leave that for a follow-up PR that just does the renaming.
I'm not sure what would be best, though @pcwalton's suggestions make
sense (though I'd probably go ahead and spell out "semantic" or
"analysis" rather than "sema" (I presume that is what it stands for))

[nikomatsakis]

On Sun, Nov 16, 2014 at 07:40:39AM -0800, arielb1 wrote:

How does this affect compile times (by reducing inlining)?

I haven't measured compilation time, though I would expect an
improvement because of increased ability to parallelize and because
there is less overall code to compile at a time.

[nikomatsakis]

Closing because bors seems confused by this PR.