Infer integration tests under tests/

[Xanewok]

By default Cargo treats tests/*.rs as integration tests, so there's no need to explicitly specify them in Cargo.toml.

---

This change is 

[Xanewok]

Tests are indeed detected, but typedarray randomly failed under nightly-x86_64-pc-windows-msvc:

     Running target\debug\deps\typedarray-47ec600ef76b3e7a.exe
running 2 tests
test typedarray ... FAILED
test typedarray_update_panic ... ok
failures:
---- typedarray stdout ----
	thread 'typedarray' panicked at 'called `Result::unwrap()` on an `Err` value: ()', src\libcore\result.rs:906:4
failures:
    typedarray
test result: FAILED. 1 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out
error: test failed, to rerun pass '--test typedarray'

https://ci.appveyor.com/project/Ms2ger/rust-mozjs/build/1.0.485/job/vytvpct6tyxx96xa#L242

[jdm]

My suspicion is that when the tests are not declared in this way, they are run using multiple threads.

[fitzgen]

Could wrap each test in a thread::spawn(|| { ... }).

[jdm]

How would that help?

[fitzgen]

So that we know that we only ever create a single JSRuntime on any given thread. If teh default is spreading #[test]s across some thread pool, it could end up creating multiple JSRuntimes on the same physical thread which is not permitted.

[Xanewok]

@jdm explicitly stating test targets in Cargo.toml doesn't change the way they are built. The way Cargo works here is that it builds each integration test as a separate binary and runs #[test]s inside in parallel, across multiple threads (unless overriden). This means that tests like typedarray.rs containing both tests do execute in the same process in multiple threads here, regardless if the targets are specified in Cargo.toml file.

[Xanewok]

Since this tests bindings to an external C library that has to guarantee that single JSRuntime per physical thread, i think this is one of the rare cases were testing with explicit RUST_TEST_THREADS=1 (seems to be replaced with cargo test -- --test-threads=1 nowadays?) makes sense.

Seems that executing with --test-threads=1 is within margin of error on my laptop and test suite execution is ~0.5s with and without this option.

[fitzgen]

Have we verified that forcing one JSRuntime per thread actually does fix the aforementioned issues? Otherwise we may be getting ahead of ourselves :)

[Xanewok]

This is intermittent and I couldn't reproduce it locally myself. I might leave it to run longer and see if the parallel version doesn't encounter failures, but I'm not really experienced with SM and JSAPI, so I'm not really sure what is the expected environment in which SM is guaranteed to run 😢

[bors-servo]

☔ The latest upstream changes (presumably #382) made this pull request unmergeable. Please resolve the merge conflicts.

[bors-servo]

☔ The latest upstream changes (presumably #430) made this pull request unmergeable. Please resolve the merge conflicts.

[bytesnail]

@Xanewok @jdm

When I executed the 'cargo test' command, I got an unexpected result. The tests in rooting.rs were skipped. Can this PR help me solve this problem?

     Running target/debug/deps/rooting-2ed6793bd8bb11f6

running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

[jdm]

No idea. I've never seen that before.

[bytesnail]

Maybe this is a bug of cargo.

[Xanewok]

Doesn't

rust-mozjs/tests/rooting.rs

Line 5 in 2afcfa4

#![cfg(feature = "debugmozjs")]

effectively prune any code in this module unless you test with debugmozjs feature? I assume Cargo still collects that file and compiles it but there may be nothing for rustc's libtest to test/run anymore at that point.

[jdm]

Ah, yes. That is correct.

[bytesnail]

Got it, thank you!