Del.icio.us-style keyword blogging

I have completed working prototype plugin for keyword blogging. You will see, at the bottom of any blog article, a list of keywords similar to what you might see on del.icio.us. If you try clicking on one of the links, you should be brought to a page with all entries containing that keyword.

I hope that this will provide an easier method for browsing blog entries; instead of being rigidly categorized, I hope that they'll be more browseable and interconnected. There are interesting possibilities for graphing the connections, and keeping statistics on connections, but for right now I'm just happy that it's working.

Implementation

It took me just over a day to implement my keyword plugin, and there was only one major snafu that required me to write some, well, "interesting" code. Throughout the process of writing the plugin, I have become more and more impressed with the way that pyblosxom is coded. Its infinite configureability makes anything possible; more complicated things just mean that you have to rewrite more of the engine.

The first thing that I had to decide was the encoding of the keywords. Since pyblosxom blogs are simply text files uploaded to the server, it didn't make sense to enter the metadata separately. Since they were in the text file, and I needed them to not render with the HTML of a blog, I decided to put them in a comment within the blog. With my plugin, any line beginning with "<!--keywords:" and ending with "-->" is a tag line.

The first step in getting my plugin working was to get a story to display its own keywords. To do this, I simply overrode cb_story (the hook which is called right before a story is rendered), parsed the file for keyword lines, and set any keywords equal to the $keywords pyblosxom variable. They could then be displayed in the article simply by referencing the variable $keywords.

Because I needed each keyword to be a link to a keyword search for itself, and because the simple pyblosxom templating language can only display a single variable, I had two choices. I could either extend the templating language to include function calls or iteration, or I could simply make the whole sequence of links right there in the cb_story function. For now, I've done the latter, since it was simpler.

Once I had the keywords displaying, the next step was to make it possible to search for a keyword. No problem - I'll just override the cb_pathinfo hook, and if there's a "/keyword/" starting the URL, take control of the selection of files to be displayed.

Once I'd gotten searching for the "/keyword/" part of the URL down, I realized that I had a major problem. How could I find the keywords when they were buried in the blog files themselves? Searching through all the text of all the blogs didn't seem like a sensible solution, but there was no easy way to implement a new piece of metadata for each file. I was stuck.

After thinking about how to implement the metadata for a while, I decided that it was important enough that it deserved a general solution. After all, there are lots of possibly interesting bits of data about a blog entry - keywords, category, comments, trackbacks, and lots of other possibilities. So the meta module came into being.

The meta module, right now, simply mirrors each "*.txt" file with a "*.meta" file. Inside the meta file is a pickled dictionary representing keywords of the "*.txt" entry. If the meta file already exists, the module simply loads that and does no parsing of the blog entry. In the future, I would like it to be a more general solution to the problem of metadata. I intend to implement, at the least, a new comments module using it.

Since the meta module loads up its information (hopefully quickly) right at the cb_start hook, which is called on plugin initialization, it makes enough information about each blog entry available that keyword searching is easy. I now simply scan through the meta information dictionary of each file and see if the requested keyword is present. If it is, the file is included for rendering, and if not, it's excluded.

And that's what I spent the last day and a half doing. Hopefully somebody will find it worthwhile, but I know that I learned an awful lot doing it. I will release my code in a couple of days, when I know that it works and I've set it up a little more generally. If you're interested in it, please leave a comment or drop me an email with suggestions or flames.