I am not satisfied with the existing per-object tls library, so I made this.
The following is quoted from README
Efficient per-object thread-local storage implementation.
thread_localcrate, it can release objects normally at the end of thread,
and will not move objects between multiple threads.
os-thread-localcrate, it only uses the
std::thread_local!abstraction of rust standard library,
and does not use any other platform-related APIs.
This means that its capacity is not limited by
And its performance is relatively good,
value access for less than 64 threads is completely lock-free, and has
But since we store thread id in
std::thread_local!, so we will be slightly slower than