[Twisted-Python] Re: [Zope3-dev] twisted zope

[Peter Hansen]

Glyph Lefkowitz wrote:
> 
> On Monday, July 28, 2003, at 01:20 PM, Gary Poster wrote:
> 
> > Oh yeah, forgot about that.  I've seen several discussions of the
> > licensing on the Twisted list... Maybe Phil Eby's implementation is
> > the only way out of that one.
> 
> The whole point of those discussions has generally been that the
> current structure is important to make relicensing a possibility :-).
> 
> While twisted as a whole is being very actively (one might even say
> "agressively") developed, some portions of the framework, like the core
> reactor, seem to be extremely stable at this point.
> 
> (Please, chime in if you disagree.)

<chime chime>

I mentioned in the past a concern that the default SelectReactor 
was being created with default variables that bind to globals for
the read and write lists that are passed to select.select().  Among
other things, it's ugly as sin, probably doesn't help performance 
much anyway, prevents creating multiple *separate* reactors (very
useful in, for example, proper unit and acceptance testing), and
has caused us at least some trouble with respect to stopping and
restarting reactors (because the globals aren't cleaned out 
properly at any point).

(Disclaimer: these comments are not based on the current Twisted,
but release 1.0.3 or 4, as I recall, but if this part has really
been "stable" then I doubt anything has changed much here.)

The last (first) time I suggested changes, the answer was a fairly
clear "we don't plan to change this ever", but if there's any
willingness to consider changes, then even the core reactor
might not be considered "extremely stable" at this point.

-Peter