Wish list for component systems in python
Posted by Michael on 13 Jan 2008 at 01:32
Interesting wishlist for kamaelia (and others, but I'm interested from a kamaelia perspective) here: http://darkness.codefu.org/wordpress/2007/12/26/295
I'm not sure I buy all the criticisms, and feel they're more a wish-list in terms of improvements specific points mentioned about kamaelia as (IMO) potential wishlist items:
- "the implementation
needs to be updated to support the new features of generators in
Python 2.5...." - that I don't buy particularly. The new features of generators are : you can now do try..finally (which was a wart, and is just a matter of using that in new code) Another is that you can send values directly into a generator. If you do that, you effectively make your generator synchronous (or at best isochronous) with the caller. I'm not convinced by a long shot that's a good thing and it feels like a mistake in the language. However by chaining with a pub/sub generator you may be able to gain somethng a bit erlang-ish (to do the matching). I'm not convinced though. The other major thing is that you can send exceptions in to kill a generator. Now that is useful, but is something we'll only do when we decide to kill off completely the kamaelia on python series 60 - which is something I'm not sure I want to do yet. The other point is "how do you model that correctly in other languages" - which is unclear.
- "....as the current syntax strikes me as rather ugly." - is the other half of the sentence. That can be true at times, though I think the use of WaitComplete and the new STM code makes this somewhat nicer - cf the greylisting server.
- "In fact,
it looks like Kamaelia needs a recent release, period: the last one I
saw was from 2006." - ouch! I hadn't realised it had been quite as long as that. That's actually quite embarassing! It was on my todo list, but I'll take that on board.
- "Kamaelia also fails to offer multi-process operation" - yep. This is something I don't make any bones about - we wanted to make sure that the current API was useful and used it in a variety of scenarios before moving this forward. The recent experiments I've made here suggest this should be relatively simple to achieve. (multiwindow pygame, first multiprocess test)
- The idea of a networked pipe seems desirable as well given this: "it
needs a way to generalize its “wires” to support communication with,
e.g., remote components. You could actually combine the marshalling
component with the framing component and a TCP client/server
components and make this work" - this is something I've played with in /Sketches - but that's nowhere near the level I'd want to use as yet - partly because we hadn't at that point really got a feel for how this could be useful. (That's changing) However, this isn't something I've had a real chance to move forward at all.
- The code should be able to migrate from microthread to thread to process as desired. This comes back to the idea that ThreadedComponent should be a chassis (ala FirstProcessBasedComponent) rather than a component you instantiate. (that would allow the threaded component to accept any component into running in its thread). This is something I realised early on could be a real problem when it comes to the issue of shared data. However the recent STM code designed to replace the dictionary in the CAT could be a real boost there. Definitely worth reconsidering. This opens up the general idea of proxy components further.
- Components in multi-language- this seems at odds with the request to use more features of python generators. My guess is actually the core would be a core set of constructs,and then a predefined inter-language comms style. I've been leaning towards JSON more and more lately, and this would certainly hit the multi-language compatibility & ease of working with issues.
Beyond that however there's also an implication of a desire for lower level configuration as well - along the lines of Shards as experimented with by Tara last summer. All of this is doable, but with limited time, its debatable how much kamaelia can satisfy all these things, especially when I'm the only person working on it, with little feedback. As a result I'll focus on the areas I'm getting feedback on :-)
All in all though - useful wish list & food for thought.
Reply to this post
Comments
Back to front page
