Extract tests from libcore to a separate crate

[sfackler]

Libcore's test infrastructure is complicated by the fact that many lang
items are defined in the crate. The current approach (realcore/realstd
imports) is hacky and hard to work with (tests inside of core::cmp
haven't been run for months!).

Moving tests to a separate crate does mean that they can only test the
public API of libcore, but I don't feel that that is too much of an
issue. The only tests that I had to get rid of were some checking the
various numeric formatters, but those are also exercised through normal
format! calls in other tests.

[sfackler]

This is necessary to finish the implementation of RFC 28.

@alexcrichton would you be able to handle the makefile updates for this? I got lost in the depths of test.mk. Once libcore isn't built for tests anymore I can cut all of the #[cfg(test)] stuff.

[huonw]

The only tests that I had to get rid of were some checking the
various numeric formatters, but those are also exercised through normal
format! calls in other tests.

Why couldn't these be moved?

[sfackler]

The structs they were testing are private (core::fmt::num::Decimal etc)

[huonw]

Ah, yes, I see that most of the tests (all the "interesting" ones) in core::fmt::num could move, and it's just the few testing the private details that had to die (and as you say, they're exercised in the other tests there).

[alexcrichton]

I don't think it will be easy to have make check-stage2-core not actually run rustc src/libcore/lib.rs --test, but it should be easy-enough to get core at least building with --test even if it's running 0 tests.

Could these perhaps be moved to one giant run-pass test? Something like:

src/test/run-pass/coretest.rs
src/test/run-pass/coretest/any.rs
src/test/run-pass/coretest/int.rs
...

That would at least involve fewer changes to the makefile infrastructure!

Could you also add a comment to the top of libcore (not a doc comment) explaining why there are 0 unit tests in the library itself?

[sfackler]

There's a fair number of dubious #[cfg(test)] and #[cfg(not(test))] stuff in coretest that we'd ideally be able to get rid of. When I was poking around in the makefiles, it seemed like it might be possible to avoid building tests for libcore by filtering it out of TEST_TARGET_CRATES?

It seems like it'd be a bit unfortunate to move it to rpass since I don't think you'd be able to do TESTNAME filtering on it.

[alexcrichton]

Ok, that makes sense. Another possibility would be a src/coretest/lib.rs folder which is only part of TEST_TARGET_CRATES?

[sfackler]

@alexcrichton seem reasonable? I still have to strip all of the test cfg-ing from libcore and rebase.

[alexcrichton]

I'm ok if you filter out core from TEST_TARGET_CRATES to avoid all the cfg-madness

[sfackler]

Should be ready to go (modulo a rebase). Look good other than that?

[sfackler]

@alexcrichton rebased, r?

[alexcrichton]

I checked mk/tests.mk and src/libcore/lib.rs for a comment explaining the split of tests. Was it placed elsewhere?

r=me if there's a comment somewhere.

[sfackler]

Oh, whoops, updated with a comment in libcore/lib.rs