[workshop-lite] rantelope day 003: the channel hierarchy

[Michal Wallace]

rantelope: day 003
------------------

Yes, I'm a day behind schedule. But this is the
third development day, so it's still day 003. :)

Today I implemented Yahoo!-style hierarchical channels. 
Whether this is a good idea or not remains to be seen. :)

As usual, the new code and live demo are up on:

    http://www.rantelope.com/


==[ today's objective ]==

I promised last time to make rantelope a true
content management system, capable of managing an
entire website, rather than just a blog. It's
not there yet, but it's closer.

I wanted hierarchy. I wanted to be able to maintain
blogs, normal web pages, link directories, FAQs,
and other content all through the same interface. 

The trick was figuring out what that meant. :)
I toyed around with a couple possibilities, and
finally decided that my Channel object really
ought to be a node in a tree. 

So, the goal was to have let each Channel have
some number of subchannels as well as some number
of stories. That means a channel can take on any
of several forms:
 
  - blog: 1..* stories, 0 subchannels
  - about page: 1 story, 0 subchannels
  - link directory: 0..* stories, 0..* subchannels 

I liked this idea, so that's what I built.

==[ implementation ]==

I ran in to a wall almost immediately. My "sixthday" 
libraries handle relationships between classes 
transparently: you can have many stories in a channel
and the Clerk does all the magic of keeping the
relationships up to date in the database.

Unfortunately, I haven't gotten around to implementing
hierarchical structures like Node, which wants to look
like this:

class Node:
    parent = link(Node)
    kids = linkset(Node)
    # breadcrumb trail: top/node1/subnode/etc...
    crumbs = linkset(Node)

I can build objects like that just fine; the problem
is persistence. Clerk sees the recursive structure 
and freaks out. I'm going to fix that, but who knows
how long that will take? So I wrote a kludge.

One nice things about object-oriented languages is
that you can encapsulate your messes. I knew how I
wanted Node to look to the outside world, but it
wouldn't yet work right out of the box. So I hacked
together an implementation and then dressed it up
to look nice, even though it wasn't.

The end result is a file called Node.py, which you 
can see in today's CVS if you're interested:

  http://cvs.sabren.com/sixthdev/cvsweb.cgi/rantelope/Node.py?rev=1.1

It's ugly and ought to be replaced, so I won't go into it. 
But, once Node worked, I could make Channel a Node subclass,
so now Channels have "kids" and "parents" and "crumbs".

Adding the interface was more or less trivial. When you
view a channel, you now see the breadcrumb trail at the
top, a list of subchannels in the middle, and a new
"add subchannel" link. 

I did have to introduce a few lines of cruft in the 
RantelApp.show_channel() command. It's ugly, but I'll
show it because it shows the kind of thing that's going
on behind those generic_show() calls:

    # note: strongbox.BoxView is a proxy class that gives
    # Strongboxen a zebra-template-friendly dict interface:
    def show_channel(self):
        chan = self.clerk.fetch(Channel, long(self.input["ID"])) 
        chan.clerk = self.clerk 
        model = {"errors":[]} 
        model.update(BoxView(chan)) 
        model["kids"]= [BoxView(k) for k in chan.kids] 
        model["crumbs"]= [BoxView(k) for k in chan.crumbs] 
        print >> self, zebra.fetch("sho_channel", model) 


Zebra expects a data model built from dictionaries and
lists of dictionaries. To render a zebra template, you
just make a dictionary with all the data the template
needs, and call zebra.fetch()


The only other thing I did was fix a bug with the filenames.
Last time, I added code to force the RSS and html filenames 
to end in ".rss" and ".html". The looked lik this:

   rssfile = attr(str, okay=lambda x: "/" not in x and x.endswith(".rss"))
   htmlfile = attr(str, okay=lambda x: "/" not in x and x.endswith(".html"))

I hinted that I could use a regular expression instead, and
it turns out I had to. Otherwise an empty string wouldn't
validate, and I wanted to let people leave these fields blank.
It now says:

     rssfile = attr(str, okay="([^/]+.rss|^$)" )
     htmlfile = attr(str, okay="([^/]+.html|^$)" )

... Which I think is a lot nicer.


Anyway, that's day three in a nutshell. 

Now that we've got the potential for channels within 
channels, it makes sense to let those channels share 
templates, so at some point we'll break the XSLT off 
into its own class.

Tomorrow, though, I'm dusting off my python-powered
indexer, and giving rantelope a search engine.



Sincerely,

Michal J Wallace
Sabren Enterprises, Inc.
-------------------------------------
contact: michal@sabren.com
hosting: http://www.cornerhost.com/
my site: http://www.sabren.net/
--------------------------------------