Test output should be coloured even when run simultaneously

[chris-morgan]

At present, test output is done without colour¹ unless RUST_TEST_TASKS is 1. On supported operating systems, colour should always be done, regardless of the value of RUST_TEST_TASKS.

On a normal OS, you're just writing the likes of testing foo... ^[[4mok^[[0m\n, so I don't see what can go wrong with the formatting, assuming it's being done as one write call.

Windows might have race conditions here on account of its make-an-API-call approach; I don't know. (Is it even supported for colouring?) But that's no reason to hold back the Dazzling Glory of Useful Colours from the proletariat ².

---

_{¹ "color", for those of you who speak en_US.}

_{² Something tells me the proletariat wouldn't be running Rust tests. Too bad. That's his fault, not mine.}

[alexcrichton]

My understanding is that they're disabled for the reason you mentioned, color emission isn't one write call, but rather three (start, content, end). Because it's not one call to write, the output can get interleaved, causing "interesting" output in parallel tests.

Although I'm perfectly ok with string concatenation to have one call to write.

[Kimundi]

Couldn't the test runner simply be a single-task manager for running all the tests, receiving an message on success/failure and writing that to stdout sequentially?
The only thing that could corrupt the output then would be if the tests themselves write to stdout/stderr. Could those two be redirected per-task and piped to the test runner to be printed sequentially too?

[brson]

@Kimundi that's exactly right that the problem is that the tests themselves are outputing to stdout and that corrupts the colors printed by the test runner. Redirecting stdout per task would require the stdio types to be questionably sophisticated. A less-controversial approach might be to simply make logging redirectable per-task and not let tests use stdio directly at all.

Perhaps as @chris-morgan mentioned the term code is doing something silly and splitting the escape codes across writes, but that would be kind of suprprising.

[jdm]

See #782.

[gifnksm]

Can std::io::LineBufferedWriter guarantee that the write is called only once per line?
I try this idea, and I'll make PR if this goes well.

[treeman]

Dup of #782?

[alexcrichton]

Yes, thanks @treeman