Subj : Re: win32 threads broken (was Win32 condition variables redux)
To   : comp.programming.threads
From : d|dq
Date : Wed Aug 03 2005 09:36 am

Joe,

Thanks for jumping in. I appreciate your insight.

You seem to say (even) "CV redux" can't be programmed easily using
Win32 primitives.

Implementations I have seen rely on:
- one or more critical sections
- one or more semaphores
- up to half a dozen counters
- a data structure such as a queue or a doubly-linked linked list
- a few Interlocked* this or Interlocked* that
- up to a 1,000 lines of C code

That's a lot of code to digest. In the end, it's very difficult to
determine if problems are hidden in such complex implementations. (And
I can't believe so much code and "logic" improves performance.)

Example: The Apache Portable Runtime (APR)
A Win32 implementation of condition variables has been part of APR for
several years. People thought it worked. a few days ago, 2 patches had
to be applied because it didn't work at all, even with 2 threads.

That's just one example. There are many others. None of them give our
industry a good name.

But, going back to "CV redux"...

Could it be done with:
- 1 auto-reset event object
(using WaitForSingleObject to wait and SetEvent to communicate a
broadcast to all waiting threads by repeatedly marking the event
signaled)
- 1 "broadcast" flag
- 1 "waiting" counter

?

PS: According to several Microsoft employees, PulseEvent should be
avoided like the plague. I even read that it remains in the Win32 API
for backward compatibility reasons only.

.