We were sure we didn't want to use PHP. Facebook is stuck on that for legacy reasons, not because it's the best choice right now.[1] Our main takeaway from that experience is that programming language choice is very important and is extremely costly to change.

Python was a language that Charlie and I both knew reasonably well (though I know it a lot better now than I did when we started). We also briefly considered C#, Java, and Scala. The biggest issues with Python are speed and the lack of typechecking.

C# seemed pretty promising. As a programming language, it's great, but:

• We didn't want to be on the Microsoft stack. We were up for learning something new, and MS SQL Server actually seemed pretty good, but we knew we'd need to integrate with lots of open source code that has only second-class support for .NET, if it supports it at all. Also, most of the best engineers these days are used to open source stuff.
• We didn't want to take the risk of being on Mono (an open source implementation of C#/.NET). It's not clear how long funding will be around for that project, and I'd heard of various performance problems. Plus, it seemed like everything else in the C# ecosystem would assume we were on the Microsoft stack.

For a lot of little reasons, Java programs end up being longer and more painful to write than the equivalent Python programs. It's also harder to interoperate with non-Java stuff. Scala had a lot of the downsides of Java and the JVM, although it wasn't quite as bad. The language seemed a little too new and like it would bring some unnecessary risk (for example, who knows how good support will be in 10 years).

Two other languages we very briefly thought about were OCaml and Haskell (neither had big enough ecosystems or good enough standard libraries, and both were potentially too hard for some designers/data analysts/non-engineers who might need to write code).

We decided that Python was fast enough for most of what we need to do (since we push our performance-critical code to backend servers written in C++ whenever possible). As far as typechecking, we ended up writing very thorough unit tests which are worth writing anyway, and achieve most of the same goals. We also had a lot of confidence that Python would continue to evolve in a direction that would be good for the life of our codebase, having watched it evolve over the last 5 years.

So far, we've been pretty happy with the choice. There's a small selection bias, but all of the early employees who'd been working with other languages in the past were happy to transition to Python, especially those coming from PHP. Since starting the following things have happened:

• Python 2.6 got to the point where enough of the libraries we used were compatible with it, and we made a very easy transition to it.
• Tornado (web framework) was released as open source, and we moved our live updating web service to that.
• PyPy got to the point where it looks like it will eventually be usable and will give us a significant speedup.

All together, these give us confidence that the language and ecosystem is moving in a good direction.

[1] What are the horrors of PHP? and Do Facebook engineers enjoy programming in PHP? and Why hasn't Facebook migrated away from PHP? and What are some of the advantages of PHP over other programming languages? for more on that.