Stackless Platform Independence?

[Just van Rossum]

In article <mailman.1015320430.17383.python-list at python.org>,
 Christian Tismer <tismer at tismer.com> wrote:

> Paul Rubin wrote:
> 
> > Do you think it's feasible to modify Python to use non-blocking i/o
> > everywhere, and then simulate blocking in Stackless?  I.e. when a
> > microthread does an i/o operation, Python would start the operation
> > asynchronously, and the Stackless microthread scheduler would block
> > the microthread til the i/o operation actually completes (detected
> > using select or SIGIO).
> 
> 
> This is roughly how I understood Frederic's idea.

This is not at all how I understood Frederic's idea... I thought his 
idea was about using os threads for C calls (such as I/O), so you could 
simply use blocking sockets as they will be run in an OS thread. With 
non-blocking sockets you can indeed get a long way avoiding OS threads 
completely, while still presenting a blocking interface, which is what 
Paul means. It's not a complete solution, though, eg on windows file 
system files don't work with select and you haven't solved the problem 
of microthreads blocking when calling a C extension. And _that's_ the 
area where Frederic's idea would shine, not so much I/O.

Just