I would love to see a more stable version of the windows reactor. :-) Occasionally where the win32eventreactor starts spinning wildly out of control when I do too many things at once. I assume this is the 63 kernel events limitation or some such thing... -Justin On Sun, 14 Mar 2004 19:36:10 +0000, "Paul Moore" said:

Itamar Shtull-Trauring writes:

...

Fixes to win32event reactor or a finished IOCP reactor so we can abandon it are welcome ;)

I was interested to see this, as I've been aware of the limitations of the standard select reactor on Win32 (no process stuff, for example), but given the comments in the documentation for the win32event reactor, I wasn't sure that was the way to go either.

One neat feature in later versions of Win32 (Windows 2000+, I believe) is a kernel-level thread pool, which handles async I/O very cleanly, as well as a lot of other stuff, such as waiting on more than 63 kernel events, asynchronous function calls, etc, etc. Looking at the IOCP reactor, it doesn't seem to use this.

Some questions:

1. Would a reactor based on Win32 thread pools be worth having? 2. Is lack of NT support a problem (Win9x is already disallowed by using IO completion ports)? 3. Does anyone want to pick this up (the author of the IOCP reactor, or the existing win32event reactor, for example)? 4. If no-one else wants to pick it up, where would I find pointers on writing a reactor? Assuming there isn't a useful "what is a reactor supposed to do" document (I couldn't find one) what code would be a good starting point? [Note: If this is what ends up happening, expect completion in something like 2038, given my copious quantities of free time, and the likelihood of me managing to stick with a substantial programming project :-(]

Paul. -- This signature intentionally left blank

_______________________________________________ Twisted-Python mailing list Twisted-Python@twistedmatrix.com http://twistedmatrix.com/cgi-bin/mailman/listinfo/twisted-python