parking_lot mutex is slower than standard library

[tsxiaofang]

I also found this issue during performance testing, and I think it's best to disable parking_lot feature by default.
Amanieu/parking_lot#338

[carllerche]

Thanks for the report. This is interesting. I think before making a decision on what to do with this, we need more data. How much does this impact Tokio in practice?

Assuming the issue does impact Tokio, we have a few options. Technically, we don't have to actually use the parking lot mutexes when the feature is enabled. It looks like the issue only impacts some platforms. One option would be to conditionally compile using the "best" mutex based on the platform. If some end users want strict control over which mutex implementation Tokio uses, we could provide cfg flags for the end application to use. E.g. --cfg tokio_force_parking_lot_mutex (or something).

This is just off the top of my head. I'm eager to learn what others think as well.

[tsxiaofang]

Yes, my current approach is to disable the default features and add the necessary features myself