To borrow a phrase from our wise elders,

"There's a lot of people flailing around and growling like epileptic bulldogs over this new [Facebook thing]. I guess this is just because [PHP developers] are intrinsically retarded."

Frankly, I don't get the buzz at all, on any level. Here's how I see it.

• Fact: PHP is a terrible language viewed from any angle.

  After a year of JavaScript, I feel completely qualified to make this claim. In other words, I know a thing or two about terrible languages, and PHP is much, much worse than JS.

  It's important to state this upfront. The language is bizarre, the standard library is completely inconsistent, the community is terrible, and in general you can tell very easily that PHP is a template language that grew out of hand. On top of that, it's an evolutionary dead-end.

  Simple example: how many programming languages do you know of that are case-insensitive? Well, count PHP among them. Really! Both class names and variable names!

  Bonus example: Can you tell me what the difference is between the following PHP expressions? They're all allowed according to the syntax.

  • foo::bar
  • foo::bar()
  • foo::$bar
  • foo::$bar()
  • $foo::bar
  • $foo::bar()
  • $foo::$bar
  • $foo::$bar()
  • $foo->bar
  • $foo->bar()
  • $foo->$bar
  • $foo->$bar()
  • foo\bar // yes, really; N.B. this only becomes allowed in PHP 5.3

In short, if your application is written in PHP, you have bigger problems. Which raises the larger question: why the fuck is Facebook written in PHP? Surely they know better by now.

• Fact: PHP isn't blazing fast to begin with.

  The programming language shootout shows PHP is approximately 1-3 times slower than Python in most benchmarks. A factor of 2 performance gains given that context are not impressive.

• Fact: Facebook is not a company known for its open source efforts.

  Not to say they haven't tried. Their platform fbOpen was released couple years ago but largely nobody cares. Their NoSQL database, Cassandra, has been out for a while, and has been gaining traction now that some guys from Digg basically spent six months patching it. Interestingly, Facebook has forked Cassandra and has been developing their own version ever since.

• Fact: the technique is not new.

  Even at 10gen we were compiling JS to Java to get speed. Compiling a higher-level language (I use the term loosely) to a lower-level one is a well-known technique for improving performance. The Python interpreter "Jython" even uses this as its main implementation strategy, directly emitting Java bytecode and then executing that. Even emacslisp has its own bytecode format, meant to make execution faster. What's the big deal?

Taken together, what are you left with? An uninteresting, unreliable and pointless attempt to make a dying language faster. And yet, it really isn't a joke to say, as Marco Tabini did, that you'd have to be living under a pretty big rock to have not heard about this. The energy around this announcement is enormous. So why all the fuss?

Color me confused.

Ethan

Origins

I got the idea for lockers towards the end of 2007 when I was playing with ikiwiki. ikiwiki, for those of you who don't know, is a piece of software that translates a tree of files into HTML, thereby "compiling" a wiki from a version control system. It has some hooks to interrogate the version control system for history, and is capable of doing some querying and rendering of files as RSS feeds. The number one use case for ikiwiki seems to be documentation/bug tracking for open source software, but some nerds use it to build their own blogs. (I'm one of those; ikiwiki powers www.betacantrips.com.) You can usually identify these people by the fact that their technical writing occasionally says something like: "the gtkbuttonnew function creates a new gtk_button". This is because they are using Markdown, and.. yeah, about that.

As I saw it, ikiwiki was inappropriate in a few ways and didn't go far enough in others. To wit:

1. ikiwiki uses Markdown by default to translate plaintext into HTML. I don't like Markdown for a lot of reasons; I prefer ReST, and ikiwiki's ReST support wasn't great, and I couldn't be bothered to fix it.
2. ikiwiki was unapologetically written in Perl, which is a completely valid software choice but one I don't love (I prefer Python).
3. ikiwiki's function is to translate files into HTML; it has very little support for truly dynamic functionality. You can edit files via the web, and you can write-lock files, but you can't easily read-lock files, for example. If you want to do something much more complicated, like say build a picture gallery, you're basically going to hurt a lot.
4. "ikiwiki" is really hard to type in Dvorak!

Of course, these criticisms may not be valid any more. I haven't been keeping up with ikiwiki too much. But in 2007 they were enough to encourage me to start working on lockers, which I did first by exploring the performance characteristics of version control systems, with the hope of using one to power a web application in realtime.

Other related work: Sputnik, which is based on Lua and Markdown. Lua stores code and data both in the database, which seems kind of crazy to me, but there you are..

Goals

For a long time I've wanted to build a "home" on the Internet where I could hang out with my friends. Over the years this desire has gotten stronger when I've made use of sites like the Button Men Web Game or ktvoelker's The New Game Center. I really loved the concept of wiki when I first found out about it, but it's hard to "hang out" on a wiki -- and as a gamer, whatever site I built had to have game functionality too.

If I was going to build some kind of home/site, I wanted it to be everything: wiki, photo gallery, board game place, discussion board, event planner/calendar -- essentially "home base" -- and additionally I loved the concept of wiki so much that I wanted it to be version-controlled.

(I know about Google Wave. More about that later.)

Development

My first commit on lockers was 2008-01-13. At the time I was using hg but eventually converted to bzr. My first priority was to build a read/write lock using POSIX's fcntl. I probably "only" spent a couple days on it, but it already goes to illustrate a few points about my development style:

1. I seized on a fairly minor aspect of the codebase (locking) and immediately went tumbling down the rabbit hole building the best implementation of that aspect that I could imagine. Refining that aspect somehow became a priority, when moving on (at most, stubbing it out) should have been the priority.
2. Eventually this component became a liability -- after I rewrote it to get it to work (ha ha), I had problems with lock contention within the same process. When this happened, I continued to refine it beyond the point of diminishing returns. It turns out that by the time lockers went in to production, no locking was necessary because only one thread ever actually did any writing.

I got the first 0.1 release of lockers running in April 2009, a year and a half later. It supported user logins, but not user registration; photo uploads; and some primitive historical querying. The picture module -- the only thing that even approached finished -- does not support multiple users; it only looks at one hard-coded user's pictures. Nevertheless I successfully uploaded a few gigabytes of pictures. But, appalled at the site's speed and deplorable code quality, I decided I would spend some time refactoring it.

It's been six or seven months. I just realized I've gone down another rabbit hole, and the functions lockers was designed to have aren't really that important anyhow, so I think I'm going to stop working on it for a while.

Design

Analysis

The larger question of whether it's important to maintain history for "everyday" actions like chatting, planning events, etc. needs to be addressed or else lockers is still just a solution in search of a problem. Over the last couple of years I've come to the conclusion that this kind of history is not important. Two things, in particular, hammered this home for me:

1. I've lately started going through lots of old files I've had lying around on my computer. I have megabytes and megabytes of old chat logs, classwork, dot-rc files, etc., not all of which are completely well-organized. I've been trying to sort everything out and cull the crap I don't really need. It turns out that the vast majority of the files I come across are pretty much worthless, even to me. It makes it hard to form a compelling argument for archiving this stuff; if I can't care about it, why will anyone else?
2. Google Wave is the obvious comparison when you talk about things like social networks, wikis, and conversations. It's an interesting project, but watching it get used, I'm even less convinced of the utility of accessing history. Nobody does it. So far it's too clunky to use, and the concepts are still a little alien, but even when those things are addressed, the history of a given interaction is usually obvious just by looking at it. After all, if something changes, people usually make that clear using in-band data: "NOTE: Time changed! I have a doctor's appointment that morning," etc.

My other goal -- building a place to hang out online -- seems to have been obviated also; most of the people I care about "hang out" on Twitter (on which I have an account; identi.ca too). And if I want to play games online with my friends, I'd be better served by loading up BSW.

With all of these considerations, and considering the fact that hacking on lockers just hasn't been fun the last few months, I'm putting it down for a while.

Lessons learned?

1. I really really suck at coding "in the large". I can knock out 500-1000 line hacks -- even well-engineered, user-friendly ones -- in a couple of weekends, but there seems to be some kind of limit -- 7500 lines, or 18 months -- after which I just don't cope very well. I don't really know why this is, if it's something I can get over with practice, or what. It's a little depressing. This isn't the first project I've abandoned, and I'm sure it won't be the last. Historically speaking, I'm afraid I'm just very very bad at following through. Sorry.
2. I still haven't learned how to find out the needs of users and build from that. I don't think I'm about to learn, either; I'm still interested in making things "for me".
3. I need to cut my losses earlier -- both in terms of stopping work on an unimportant component, and dropping projects that aren't interesting.
4. Not Invented Here syndrome can sometimes be good because it saves you from going off into the weeds trying to adapt yourself to something that doesn't fit, but it also can be bad because it encourages you to go off into the weeds to build something that still won't be exactly what you want.
5. Worse is Better. Actually, Worse is Much, Much, Better. Also: Bad is Good, Baby, Down with Government.

I don't yet know what I'm going to work on next, or indeed if I'm going to work on anything next. I haven't really had time to hack much lately -- between work, reality, my family, and trying to hang out with my friends once in a while, I haven't really had the time to do everything I wanted. It seems like organization helps you make time, but perhaps at the cost of robbing your life of every joy.

I will say this: my current life goal is to do something notable. Secret projects seem the best way to go about doing that, if you find the right secret project.

Ethan