view roadmap.txt @ 6:3198742f7a86

notes to self
author k0s <k0scist@gmail.com>
date Mon, 05 Oct 2009 09:20:02 -0400
parents e3823be6a423
children
line wrap: on
line source

bitsyblog, middleware, and the roadmap to the future

the nice thing with working like software like bitsyblog is that the
intent of the software is clear.  bitsyblog aims to provide a solution
for users to keep a log of their activities on the web.  When thinking of
software in terms of the narrow intent of a component, it becomes much
easier to divine what sort of framework would look like around it and
how this framework might be applicable to other types of software.

Firstly, I noticed that the bitsyblog component app is not as small as
it could be.  bitsyblog provides a multi-user solution to (personal)
blogging.  This level of solution should exist, but thinking in terms
of components it is easy to see that users may want a personal blog
that is not multi-user.  lammy wants a blog on his desktop computer.
When at his computer, he probably wants to be logged in all the time
(always accept connections from 127.0.0.1).  He doesn't want other
people to be able to make a blog there.  When not at his desktop, he
wants to be able to login and post through the web and maybe used the
blogme.py script too. In bitsyblog terms, '/lammy'
becomes '/' and all urls ('preferences', 'post', 'css/') become
relative to this.  Doing this involves separating out the individual
blog components from the multi-user component and probably having
different .ini files from which to run paste.

I've also realized that the way I handle CSS is a solution to a
problem that has nothing to do with bitsyblog.  The other items in
preferences ('date', 'subject', but not 'friends' which has to do with
roles) are about formatting the blog posts.  The CSS is about themeing
pages which has nothing to do with blog posts.  So what is needed is a
CSS component, which is possibly linked to from the preferences page.
Lets put this at '/lammy/css'.  This pages may have the fields for
preferences, but also provides an interface for managing the CSS.  A
user may want to delete CSS or edit them, render the current page with
a selected stylesheet or what's in the editor, or render an arbitrary web page
This is like CSS zen garden, but more interactive, and
(optionally) allowing the user to save the CSS and apply it to their
pages.  I can also imagine, in a 2.0 version, an implementation that
allows permits transforming of CSS to suitable to theming a page with
a stylesheet intended for another page.  One has a stylesheet with a
bunch of rules.  Using javascript, one mouses over the DOM of the
other page.  The rules are modified in order to applied to the
appropriate container on the page.  This sounds much like what
deliverance is supposed to do, and maybe it could be utilized for this
intent, though perhaps the intents are different and its better not to
use it (to be investigated).

Both of the above components fall into a general pattern:  there is a
component which is the multi-user blog manager.  Lets call this
bitsyblog.  There are microcomponents, the single-user blog (lets call
this microblog) and the CSS zen garden component which lets you play
with stylesheets (I dub thee samadhi).  microblog and samadhi can live
on their own, in which case they live at '/', or they might be managed
by an external component -- microblog by bitsyblog, or samadhi
by...not really sure, either a microblog instance or bitsyblog.  In
the latter case, they need to know where they live.  I'm not sure what
provides the rewrite rule telling them where they live.  Ian?  Ethan?
There are a few configuration type details -- like auth, whether
samadhi needs to know about where to put these CSS files or how
bitsyusers can fetch them, but essentially if you give these the
appropriate path, they should be able to work relatively
independently.

In a different class of components are comments and tagging.  While it
is easy to imagine bitsyblog managing comments and tags, it is also
fairly easy to see that these services really don't need to know what
is being commented or tagged on.  For commenting, I imagine an app
similar to smokesignals, but with a different persistence model.  Lets
call it commentator.  Commentator is a javascript app with no
intrinsic URLs.  Given a set of rules (for instance, "allow commenting
on <div>s with an id attribute) a block of text is shown as the last child
of each <div> (or conditionally on mouseover).  This will initially
say something like "No comments | Add a comment" where each of these
do what you think ('N comments' displays the comments, 'Add a comment'
gives a textarea where you can say your mind).  When a comment is
added, it is put in a database (presumedly SQL, though really on a
super-simple db is necessary) with the URL and/or id
as its key.  When the page is loaded, the text of the page is examined
on response and the appropriate comments are displayed.  It is also
necessary to have some sort of mapping optionally between URL and id
to another URI, as blog posts both are displayed as part of the user's
entire blog and also have a unique URL.  Notice that nothing here is
particularly unique to blogs.  Commenting on <div>s or <p>s in a wiki
page is essentially the same (though in this case, it might be better
to comment on the entire page, as these may not have ids and the page
can be editted all willy-nilly like that.  Commenting on individual
<p>s and <div>s is, however, a really nice thing to have, so in a 2.0
type incarnation of commentator, this should be added regardless of
the difficulty).  If bitsyblog was part of a
site with a wiki, then commentator could deal with them both nicely.
Optionally, one might specify the rule that if an item had comments,
it either could not be deleted or that an alert would pop up before
deletion with [Confirm] and [Cancel] buttons.  This is easy to do if a
browser could do a DELETE request and things behaved all RESTful.
Otherwise, this would also be a non-trivial problem.

Tagging is very similar to commentator, though it might have a few
intrinsic URLs too.  You specify a similar set of rules concerning
what can be tagged ("all <div>s with ids in /%(user)s/ and
/%(user)s/200*") and a method pointing the ids to URLs.  Then you have
a database, which again is pretty simple and again in its simplest
form could be a dictionary.  Though here, the tags are the keys and
the URLs are the values.  You could also record number of tags per
item, specify whether non-authenticated members can tag, etc.  As the
first or last child of each (say) <div>, you have what it has been
tagged with, maybe weighted by how many times it has been tagged.  You
have a URL for searching by tags, maybe on a per-path basis ("I want
to see all the tags under /k0s that have been tagged with 'cats' and
'cute'").  You could also have a URL for the tag cloud.  Again, what
is being tagged is pretty divorced from the tagging component itself
(lets call it tagit, unless that name is taken).

I have presented samadhi as being managed by bitsyblog or microblog
and commentator and tagit living outside of bitsyblog in the
middleware stack.  I've done this because this "feels right" to me
right now, though I reseve my right to change my mind about it.  The
thing I want emphasize is the difference in the two approaches:

* When a WSGI app is "managed" by another component app, the manager
  has to know about the managed app and dispatch the appropriate
  information (request, etc) to it, though in doing so it has the
  opportunity to have more control over the URL structure and what the
  component app does.  It is also easier to return HTML this way.

* When a WSGI "app" is used as middleware and needs to markup the
  returned page in HTML, this is conducive to using javascript in
  order to insert nodes into the DOM dynamically, presumedly based on
  some rule set.  Alternatively, the middleware may decompose the page
  with lxml and insert HTML where it needs to go that way.  In this
  approach, the middleware need not have a URL (this may be
  preferable).

It is not necessarily clear to me at this time which approach is
better for different apps.  I have put commenting and tagging in the
latter category because commenting and tagging on web pages seems
divorced (save by a rule set) from what is being commented on or
tagged.  I have put microblog in the first category because it is
clearly in the same layer of software as bitsyblog, and some
management tasks will be necessary even for microblog.  samadhi I put
in the first category as well (though I hesistate right now) because
otherwise the rewriting of URLs makes me nervous.  Perhaps there is
need of a WSGI dispatcher that fetches apps based on some sort of rule
concerning the URL?  Also, it seems
necessary to put the CSS on user data.  This could also be
accomplished the other way:  1. abstract out members and roles
(teamroler?) and store the data (CSS, etc) on them here; 2. have
samadhi write a defined set of urls (how defined?  not sure) including
the stylesheets that it knows about, presumedly putting the 'foo'
stylesheet at '/lammy/css/foo.css'.  This still involves samadhi
knowing something about the url structure, and now it is dependent on
the user/role thing, though this could be a configuration option.

This is the middleware story.  Other things could also be done
internally to bitsyblog and its supporting software to make the
experience more seamless.  

I am bothered by the fact that the logic supporting the form and user
settings is somewhat divorced.  Ideally, rendering the form should do
all the validation and return the settings in a sensible way to use.
I don't think this is hard -- the settings.py should just be
incorporated into markup.form.  This goes some ways to reinventing
things like formencode and zope.schema, so if we can get these to do
what we want, maybe we shouldn't write our own.  But at cursory
glance, formencode won't and zope.schema isn't easily usable outside
of zope.  So maybe rolling our own isn't bad.

There's more middleware, but its hidden and is a bad coupling (that
is, the app will depend on the middleware unless its done cleverly).
The site nav does not appear on '/login' or '/join' (or any url not
specified internally to bitsyblog, for that matter).  It would be
conceivable to have a SiteNav middleware that (either through
javascript or lxml decomposition) writes the site nav at the top of
the screen.  bistyblog and bitsyauth would have to know about this
though, so that they could add the relevent links into environ.  Maybe
this isn't horrible, but it seems so.  Likewise, the site css doesn't
show up here.  Should it?  Is this another piece of middleware or part
of samadhi?

An important piece of middleware should involve members and roles.  I
really don't want to write this one, but I would like to consume it.
It should live just inside of bitsyauth and make the users.py file
unnecessary.  It might be better to sit on this one until last.

So should be bitsyblog 2.0.
There are a few other streamlining things (see the TODO.txt), but if
all of this was implemented, bitsyblog would be a rich web experience
that was highly componentized with each of the components doing one
thing well.  Moreso, since coupling is avoided, save through
configuration, auth, and other loose links, each of these components
could be used independently of bitsyblog or turned off without
undesirable consequences.

At least as much as being a good blogging app, bitsyblog is about
elegant design and component architecture.  This roadmap shows what is
possible in terms of having a fully functional blogging application
with middleware that can be used outside of just bitsyblog.  To take
the other end of the spectrum, wordpress says "Hey, write plugins that
will make me work better."  As a programmer, I really hate this
approach.  If I write a commenting plugin for wordpress, then I have
just that.  If I want to use it to comment on a wiki page, I'm out of
luck.  If I write commenting middleware, then it can be used for any
WSGI app with minimal configuration.  As a programmer, I generally
want the freedom to assemble a webapp however I want.  With this
freedom comes some work, but for me its a worthy sacrifice.  As a web
consumer, I understand that site administrators want out of the box
solutions.  bitsyblog will offer this too.  The default deployment of
bitsyblog will pull down commentator, samadhi, and tagit and have an
.ini file pointing to the appropriate factory to create the chain of
middleware to have all of these pieces in place.  There is no contradiction.