default config should include host from the current rustc

[dwrensha]

Miri's run-pass tests use a compiletest default config without setting the host field. My expectation is that this would result in host and target being detected from the current version of rustc. Instead, host gets set to "(none)", which causes a test that has ignore-cross-compile to get unexpectedly ignored.

[dwrensha]

(cc @oli-obk @eddyb)

[oli-obk]

I've been wondering about that.

[laumann]

Would the change in #66 have any effect elsewhere?

[dwrensha]

I suppose someone could try running the change (#66) on the crates that depend on compiletest: https://crates.io/crates/compiletest_rs/reverse_dependencies --- something like a mini crater run. (cc @shepmaster, who I think at one point was wondering whether automation for such testing might be useful.)

I think the main things we should worry about are causing tests to fail silently and causing tests to be unexpectedly ignored. It seems unlikely to me that this change would result in either. Meanwhile, compiletest's current behavior is causing tests to be unexpectedly ignored right now out in the wild, as the miri code linked above shows.

[dwrensha]

c.f. #13, where the target field default behavior was added.

Why would the default have the target set from rustc, but have a "(none)" value in host?

[laumann]

I see your point wrt #13 - I'll merge #66

[laumann]

https://crates.io/crates/compiletest_rs/0.2.7 is out!