We're using this in rustup; no particular view on 2.0; 1.x is pretty nice.

Hey @dns2utf8, I appreciate you putting in the effort to write this proposal! Although I'm very limited on time nowadays and don't have the bandwidth to review this right now. Maybe someone else in the Rust community could help out with this?

Thank you all for the feedback :)
I will focus my first effort on cleaning up the API in the 2.0 branch.

@frewsxcv that is okay, I will find someone.

Hope you can enjoy the summer weather sometimes 🦀
And if you are at CCCamp we could meet there.

Cheers,
Stefan

Disclaimer: this is my first time to actually participate in open source community so some ideas might look naive and totally wrong.

It seems to me that the feature "Access to the pool handle from within the task" is not very much needed (I didn't find the corresponding issue tho). What would be the use case of it?

As for NUMA optimizations, will that involve pin threads to a particular core and box the passed-in function onto the memory adjacent to the particular core? If all threads are busy, how does the scheduler decide which function to send to which core? I am just curious about the implementation and would like to contribute to that part if possible, since coding it sounds very fascinating :)