run test suite also against libstd with full MIR

[RalfJung]

I have zero experience with Travis, so I will see whether my script makes any sense when Travis checks this PR. Also, I don't think this entire multi-platform-testing stuff will work out here and I am not sure what is the best way to fix that.

Someone should look at the situation with these two environment variables (RUST_SYSROOT vs. SYSROOT).

[RalfJung]

Well, that actually went better than expected. I had to make the stack overflow test less precise, because the exact point where the stack overflow happens depends on how many functions are called "between" the 'start' lang item and main, but other than that, things seem to Work (TM) now.

Maybe the compiled xargo should be cached somehow?

[oli-obk]

We could cache it, but the next nightly will break the cache

[oli-obk]

I think this only runs for the host now. Fine by me. Xargo can't generate libstd for other targets anyway.

[RalfJung]

Yeah, that's true. I tried using xargo build --target ..., but that failed. I assume it's because there is some C code being compiled here (in jemalloc)?

[oli-obk]

Nevermind. It runs the old Test suite for all targets and the Xargo one for full mir

[RalfJung]

Glad you're fine with that :)

[oli-obk]

Idk what the issue is. Just that the xargo docs state the fact that libstd can only be generated for the host

[RalfJung]

They do? Must have missed that part.^^

[oli-obk]

They did, but I got it the wrong way around. You couldn't build a sysroot for the host. But you can now