Don Williamson

This certainly is a cross-platform minefield. https://daniel.haxx.se/docs/poll-vs-select.html

I mean, this is a pretty crazy old API limitation, almost funny. I'm not sure I want to introduce `poll` as I'd still need a `select` fallback for the systems...

Just looked at the implementation of `fd_set` and realised it's not actually a "bit array" but a `SOCKET` array (in `winsock2.h`). It looks like you can't rely on any implementation...

I think that would help and needs adding, with obvious error reporting to the user. But it's not the solution. The irony in all this is that it's fine on...

It's in wide use as a replacement and was introduced in 1986 vs 1983 but POSIX-like systems sometimes muck it up (e.g. see https://daniel.haxx.se/blog/2016/10/11/poll-on-mac-10-12-is-broken/). Remotery is used on so many...

Good idea. Maybe worth doing it the other way round: use `select` on Windows and `poll` elsewhere. If that fails to compile for somebody can introduce a user define that...

Thank you. There is scope in the future for use of `std::atomic` in C++ only code. It's unfortunate that the spec chooses to encode this in library primitives but we...

This looks somewhat safe, although I'm struggling to find a Mac OS primitive in C to express this: if the load of `write_pos` is stale it just means allocations can...

In `LoadAcquire`, can you try inserting a call to `OSMemoryBarrier()` after the call to `CompilerReadFence()`? This will require including ``. This all does need a very thorough cleanup on all...

I guess TSAN doesn't understand pre-C11 code bases. It also doesn't appear to support explicit fences, so I'll have to bring in some C11 primitives. I mean, that's no loss...