Blogging with Org-Mode

A few weeks ago Matthew Pugh pointed me to Testing Grounds, a blog he started writing to help motivate him to learn more Python for doing statistical problems. The math is beyond me, but seeing the blog got me thinking about my old websites and how I haven't done anything with them for a lone time. I used to write a personal blog and put my publications on CraigUlmer.com, but I stopped updating both of them about the same time as when we had kids. A few years back I started writing things on Google+. As much as people like to hate on G+, it's been a good outlet for me. It's easy to add new things, you can write short or long posts, and it's a great way to hear stuff from all my friends that left the lab and went to Google.

Still, there are plenty of annoyances with G+. The main complaint I have is that it's effectively a "0-day content" site, where anything older than a day or so doesn't matter to anyone (well, except Google's data mining machines). It's easy for good posts to get buried in your stream if you've got a crazy uncle that likes to post about guns, cats, and Obama. When I first started posting, I often found myself wondering if my post had actually been dropped. It turned out they hadn't- it's just that other friends were posting more stuff and that had just pushed mine down on the list. At some level, you either get caught up in the popularity contest, or you write things for yourself and stop caring if other read it. I'm the latter. The crazy people are also a problem on G+. Comments aren't as bad as YouTube, but there are a lot of people on G+ that have nothing better to do than shout you down for mildly mentioning a view on a touch subject. The last big concern is what Google will do with the content in the future. I don't think G+ will be going away anytime soon, but then again I didn't think Google Reader would go away either. I'd just assume have a copy of what I write somewhere else.

Getting the Blog Back Together

So, I started thinking about ways I could pull some of my previous writing out of G+ and host it myself. Getting the content was easy- I just walked through my page and started cutting-and-pasting posts that were technical in nature. I use Emacs's org-mode a great deal at home and work, so I just stuck the content in a flat file and used org's markup to group and tag posts. The images got placed in a local directory, but those were easy enough to link back into the org file. The more stuff I put into the org doc, the more I felt like it was the native content format I wanted to use to host a blog. The next step was looking for something that could render it into a webpage.

People have written tons of tools that convert org-mode data to other things. Org-mode has a built-in exporter to html or latex. While this would be the easiest to get running, I decided against it because static-site generators drive me crazy. I wrote a static generator for my first blog, and it was a pain to maintain. Plus I don't like all the annotation you have to do to the org file to make it render right. Next, I saw lots of people do wiki-like blogs using org-mode. It seemed like most of these were more wiki-like, and used org's doc linking to stitch everything together. I wanted everything to get plugged into the same org file. Plus, the styles never seemed like what I wanted.

Next I started looking at client-side rendering. I figured I'd push org chunks out from the server and then do something in the client to render them. I'd been meaning to check out Dart, so I picked up the dev kit and started writing simple examples. Dart is pretty cool, especially since it looks like it'd let me skip having to learn JavaScript. However, I ran into a few roadblocks. First, I realized the parsing I'd need to do was way beyond what I would be able to figure out without a lot more work. Second, I realized that it would be dumb to send the entire org file to the client, and that what I really needed was something on the server side to chop up entries and serve them up. That meant parsing on both the client and the server, which sounded like a lot of work. During this time, I discovered org-js, a Javascript interpreter for org-mode that could do the rendering. I got it to work, but I had trouble shoehorning it into the web page I liked because the Javascript is complex enough that I wouldn't want to mess with it.

Back to the 90's

Since I needed some server side of code to chop things up anyways, I decided to resort back to my 1990's view of the web and just write some CGI. I looked into using Go, as there's an easy-to-use CGI lib available. However, my provider doesn't have Go support, and the iterative development process for web work makes compiled languages a pain. So, sadly, after all my hopes of doing something new, I wound up just using Perl again.

Ok, to be fair, parsing and web page generating is the kind of thing Perl does very well. It was satisfying to know that I could just throw in a couple of lines of regex and be able to get cool useful features. The bulk of the work was coming up with a simple tokenizer that could drive a simple state machine and begin/close sections properly. Fortunately, org's main keywords are usually easy to pick off, as they usually start at the beginning of the line. Bulletized lists were a pain because:

• They require surrounding stuff
• You have to know when you're still in a list
• You still have to parse for other markup
• They can jump around a lot

I took advantage of Google's Code Prettify to allow me to put code stubs in the stream. It's a bit of Javascript that sets the colors of code, which meant I didn't have to do anything to display code (besides handle lt/gt kinds of symbols that html has problems with). I need to tweak the colors, but good enough.

#!/bin/bash
echo "Hello, suckers!"

The one thing left I probably should do is put in tables. Org-mode has an awesome table entry system that makes it easy to create tables and put new things into it. I think that should just be a simple fix, but I'll get to that when I need it.

I probably spent just as much time writing the parser as I did trying to figure out how to make html do what I want to. I know it's not popular, but I decided to make this page a side scroller. Vertical scrollers are nice, but I get sick of dragging down to get to the next thing. Everyone has widescreen monitors these days- I don't know why more people do stuff horizontally. I decided not to absolute the top heading bar at the last minute. I like the idea of having navigation fields in a constant location, but I don't like the idea of losing screen space. If you have a better compromise, let me know. For the record, yeah, I know this looks pretty 90-ish. The 90s were still good times though.

It's Up

If you're reading this, then it looks like I've gotten the whole thing up and running. My intention is to post longer writeups on technical things to craigulmer.com and crosspost to G+, so people can send me comments there. I probably won't post personal blog stuff here, as I intend this to be the place where people can see what kind of work I do (plus, the world has gotten a lot creepier since when I first started running a personal blog, and thus I'm a lot more worried about what I say these days). Here's to hoping this isn't the last post that's made on this blog.