Hi!

At my workplace we use libmongoc in a single-threaded mode atop of stackful-coroutines-based asynchronicity: we have a small amount of threads that manage lots of coroutines, and instead of suspending a thread for I/O or sleeping we suspend a coroutine, while a thread is free to process other bits of workload, if any.

It works wonders with custom implementation of mongoc_stream_t, however because of blocking usleep call usage  in topology rescan we had to patch libmongoc to allow users to provide their own implementation of _mongoc_usleep (coroutine-friendly nonblocking sleep in our case).

Would you be interested in reviewing/accepting such a patch, which adds a possibility for users to provide their own sleep implementation?