Add debug logging output to Sender

[larsbergstrom]

It would be great if we could get logging output from channels so that send would do something like:

sending message MSG from task TASK

And any call to recv would do

received message MSG in task TASK

In Servo, we frequently need to debug race conditions and deadlocks or just explain what happens when e.g. we get a mouse click in the window. Having this debugging info would greatly help (and let us build visualization tools on top of the output).

[alexcrichton]

Note that debugging a message would probably involve using {:?}, which I'm hesistant to commit to in the usage of channels. I'd be OK with by default, these messages would be logged

• task {name} - sending message
• task {name} - blocking for a message
• task {name} - received a message

If you built libstd with --cfg debug (or something like that), then we could include the {:?} format of the message. I figure that if we put the task name up front it's easier to scan over.

[larsbergstrom]

That strategy sounds perfect.

Additionally, with the blocking for a message output, if we also have a way of logging task exit, it would help with another problem that we have in servo right now, which is that when you do pool.shutdown() and have a green thread blocked in recv, it doesn't show up in threads list (LLDB), but the fact that it was blocking would show up in the output.

[steveklabnik]

Triage: no change that I'm aware of.

@larsbergstrom do you still feel this pain?

[larsbergstrom]

So we feel the pain a bit less after the removal of green threads, because there's always a stack backtrace above the thread.

That said, we have these concurrency problems ALL the time, leading to huge numbers of intermittents in Servo. I suspect, though, that the right solution is for us to be using https://github.com/mozilla/rr to debug these issues in Servo, though, rather than broken printf-debugging. If we want that, we can always (at least on linux) use some cheesy gdb-automation breakpoints to spit out messages similar to the ones @alexcrichton is suggesting, without bloating the stdlib.

[apasel422]

We could take the approach @alexcrichton described, but use specialization to log the actual values when T: Debug.

[steveklabnik]

Triage: with no comments in years, I don't think this is likely to happen. I'm going to give it a close. If anyone strongly desires this, please let me know, and we can re-open.