Make job queue channel bounded

[ishitatsuyuki]

Reference: #6197

[rust-highfive]

r? @alexcrichton

(rust_highfive has picked a reviewer for you, use r? to override)

[alexcrichton]

Thanks for the PR! I would be wary to merge this though as I'm not sure it wouldn't lead to deadlock and/or unpredictable behavior, and in reality I don't think we'll ever run into the problem where child threads outpace the main thread that much to make this that necessary

[ishitatsuyuki]

I'd like to note that it definitely can outpace, like when you're using a pipe-based tool that doesn't aggressively drain the output (I remember less did that?), which makes the main thread blocked. Thus backpressure is important here.

Regarding deadlocks, I somewhat understand your concern, but I'm not sure if there's a concrete case. In particular, do we communicate back to the child thread? In that case, we may want to visit that to ensure deadlocks don't happen.

[alexcrichton]

I could go either way on the "whether this is necessary" part, but to weed out potential deadlock issues I think it'd be best to test this out with a capacity of zero or one and then test this one some large projects (like Servo) with a lot happening in parallel and perhaps with a variety of -v flags as well.

These sorts of deceptively small changes have all classically created a lot of headache down the road as bugs are tracked down in one way or another.

[ishitatsuyuki]

So yeah, it actually deadlocked with capacity=0 during tests. I'll try to figure out what dataflow is causing this.

[ishitatsuyuki]

Closing this: I don't have much time to refactor the code to remove all cases of deadlock.