are where the code was actually "wrong" in some way and the cast was not
intentional.

If the compiler thinks you need to be warned about casting a pointer to an
int of a different size and that's exactly what we mean to do, then we
*should* be getting a warning.

Also, prevent a write of zero bytes from clearing the empty event.

your whole day)

Does Win32 hate this?

To prevent mutex leaks, lock the mutex before destroying it...
To prevent sem leaks, if the return value is -1 and the sem is not invalid,
wait, post, retry.

Unfortunately, POSIX doesn't define the behaviour or sem_destroy() if called
while there are threads blocked on sem... luckily, we rarely use POSIX sems.

semaphore from being prematurely reset.

which caused lock-ups in RingBufWrite().

utilization). We really need a cross-platform manually resetable boolean
semaphore (like a Windows "event").

numbers for wrapped buffers (off by one). Whoops.
Created RingBuf.empty_sem for tasks that need to wait for a buffer to be empty.

block size average).

each write), must define RINGBUF_SEM.

old 486 hardware!

Experimenting with this in sbbs to see if there are any race conditions between
threads that manipulate the I/O buffers (e.g. Ctrl-C in input_thread clears
output buffer).

(for example), you end up with 10K of usable buffer (need extra byte to detect
wrap). Was previously limiting usable buffer to size-1.