As this blog ambles forward, albeit haltingly, I find that the process of generating the site has become a much more complicated proposition. I suppose that's the price of success, or at least the price of verbosity.

Here's the problem: I really cannot abide by dynamically generated publication systems: there are more things that can go wrong, they can be somewhat inflexible, they don't always scale very well, and it seems like horrible overkill for what I do. At the same time, I have a huge quantity of static content in this site, and it needs to be generated and managed in some way. It's an evolving problem, and perhaps one that isn't of great specific interest to the blog, but I've learned some things in the process, and I think it's worthwhile to do a little bit of rehashing and extrapolating.

The fundamental problem is that the takes a long time to rebuild. This is mostly a result of the time it takes to convert the Markdown text to HTML. It's a couple of minutes for the full build. There are a couple of solutions. The first would be to pass the build script some information about when files were modified and then have it only rebuild those files. This is effective but ends up being complicated: version control systems don't tend to version mtime and importantly there are pages in the site--like archives--which can become unstuck without some sort of metadata cache between builds. The second solution is to provide very limited automatically generated archives and only regenerate the last 100 or so posts, and supplement the limited archive with more manual archives. That's what I've chosen to do.

The problem is that even the last 100 or so entries takes a dozen seconds or more to regenerate. This might not seem like a lot to you, but the truth that at an interactive terminal, 10-20 seconds feels interminable. So while I've spent a lot of time recently trying to fix the underlying problem--the time that it took to regenerate the html--when I realized that the problem wasn't really that the rebuilds took forever, it was that I had to wait for them to finish. The solution: background the task and send messages to my IM client when the rebuild completed.

The lesson: don't optimize anything that you don't have to optimize, and if it annoys you, find a better way to ignore it.

At the same time I've purchased a new domain, and I would kind of like to be able to publish something more or less instantly, without hacking on it like crazy. But I'm an edge case. I wish there were a static site generator, like my beloved jekyll that provided great flexibility, and generated static content, in a smart and efficient manner. Most of these site compilers, however, are crude tools with very little logic for smart rebuilding: and really, given the profiles of most sites that they are used to build: this makes total sense.

I realize that this post comes off as pretty complaining, and even so, I'm firmly of the opinion that this way of producing content for the web is the most sane method that exists. I've been talking with a friend for a little while about developing a way to build websites and we've more or less come upon a similar model. Even my day job project uses a system that runs on the same premise.

Since I started writing this post, I've even taken this one step further. In the beginning I had to watch the process build. Then I basically kicked off the build process and sent it to the background and had it send me a message when it was done. Now, I have rebuilds scheduled in cron, so that the site does an automatic rebuild (the long process) a few times a day, and quick rebuilds a few times an hour.

Is this less efficient in the long run? Without a doubt. But processors cycles are cheap, and the builds are only long in the subjective sense. In the end I'd rather not even think that builds are going on, and let the software do all of the thinking and worrying.