Fun and Profit with Liferea Conversion Filters

Sun 13 September 2009

Liferea is an RSS feed reader, with some hidden powers. I use it to automatically download enclosures from the TED feed, and aggregate unread comic strips into a single scroll page in a custom view folder. But today, I'd like to present conversion filters.

Conversion filters are scripts that function according to UNIX pipes semantics to process feeds before display. Liferea performs the task of retrieving the XML, starts the conversion filter, and passes the XML over stdin. The filter prints a converted XML file back to Liferea over stdout, and which happily displays the new RSS feed. For the graphic minded:

liferea conversion filter diagram

A small repository of conversion filters has been started, mainly driven by a different, command line driven RSS reader, snownews. Most published conversion filters are what you'd call a source in dataflow, rather than a filter. Even more are site specific, making this more of a greasemonkey for RSS feeds.

I'll share an example I concocted today. I like the Freakonomics blog, but lately they've added some unrelated content called Quotes Uncovered. I have no idea how identifying the historical source of quotations relates to microeconomics, but I do have an idea of how to use a conversion filter to remove it. But first, we need to talk for a moment about XPath.

XML is tree structured, like a filesystem. I've seen at least two people take this analogy to it's conclusion, and implement an XMLfs via FUSE. XPath also uses this analogy, for a different purpose, describing paths in documents. Paths in UNIX filesystems can be singular, like /etc/apt/sources.list, or plural, like /etc/apt/sources.list.d/ (technically your shell expands this, but play along). Concepts like *../ and *** are supported by XPath. Instead of directories though, we're going to use XPath to navigate nodes in XML.

Unfortunately, I'm interested in a specific node, based not on it's name but it's contents. So I need some predicate to narrow this. In UNIX shell, find is commonly used for this purpose, but XPath integrates this concept and diverges from UNIX paths. In the case of my Freakonomics cleaner, I want to delete the item node that has Quotes Uncovered in a child title node:

//*/item/title[contains(text(),'Quotes Uncovered')]/..

Liferea doesn't understand XPath though, so we'll need to use the conversion filter to handle. Conversion filters I've seen thus far end up scripting parsing the XML, processing the structure and printing it. Only the middle part is actually going to be unique to conversion filters. Fortunately, there's a program that adheres to UNIX philosophy and handles this: XMLStarlet (in Universe). XMLstarlet reads XML from stdin, and writes XML from stdout, just as our conversion filters must. On a technical level, it converts the command line to an XSLT and applies it to the input. In this case I just need to tell it to delete the section matching that XPath from the feed. That's accomplished with the ed -d option.

So now I just create a new subscription, and put the command in the box:

liferea subscription properties dialog

The result is a one-liner that puts the Quotes Uncovered series to pasture. It's tempting to UNIX pipes to build more complicated filters, but I'm told Liferea isn't currently coded for chained pipes. So if you get more complicated than the basic 3 step Fetch-Filter-Display pipeline, you'll need to write a short shell script for that, or write some really crazy XPath.


The Snownews extension repo is interesting, but doesn't have a way to link feeds with scripts. Earlier I compared conversion filters to Greasemonkey. Well Greasemonkey has a partner extension called Greasefire and a website backend for script discovery. It'd be handy to have conversion filter discovery for RSS!

Comments !