[Twisted-Python] Dropping support for old cred

[Glyph Lefkowitz]

Bruce Mitchener wrote:
> Jp Calderone wrote:

> Our codebase at work still uses old cred.  I've never had the time or 
> customer requirement to do the upgrade.  Unless I'd get a performance 
> benefit out of it, it'd be hard to justify the time spent.

old->new cred would have zero performance benefit, I'd guess.  Twisted 
1.3->2.1 might have some but I don't believe anyone's measured it 
comprehensively yet.  1.3->2.1 with deprecations would almost certainly 
be a slowdown.

> On the other hand, perhaps I do not need Twisted 2.1 either.  What do I 
> stand to gain by making the switch?  What benefit can I use to sell 
> this?  I assume that I could do this fairly incrementally as we have 
> about 10 different processes that communicate within a server cluster 
> over PB.  I still use the old PB APIs too (so no PBClientFactory for me!).

Given what I know about your application, upgrading to 2.1 wouldn't be a 
big win.  At some point - if we are anywhere close to the "schedule" 
that I imagine, this would be around 2.2 - there will be several new 
protocols availble and you might want to upgrade to make use of those. 
The main advantage of new cred is that it gives a consistent view of 
login and session management across all protocols that your application 
supports.  Right now the featureset is rather bare-bones, but I have 
plans to build upon the existing API and add features to support things 
like cross-protocol session management and authentication (e.g. 
automatic support for things like SMTP-after-POP authentication, but 
using any two protocols).  Something like that would be totally 
impossible with old cred, since it barely even provides an abstraction 
for the same sorts of things that new cred does.

At some point NewPB will arrive, too - and when that does, I believe 
that the upgrade in terms of performance and security will be well worth 
it.  Talk to warner about when that might happen though, it's still a 
mystery to me :).