The joys of functional programming with side-effects

→ 01 June 2009. Tags: programming, python, ruby

Quick quiz: what will the following pieces of ostensibly identical python and ruby programs print out?

If you answered “1 2 3 4” or something similar, you’re half right. That’s what you’ll get from ruby. Surprisingly, python will give you “4 4 4 4”. I'm not the first to discover this, but that doesn’t make it any less startling.

The “fix” in python is to replace the lambda line with q.append(lambda i=i: puts(i)), because default function paramaters are evaluated at definition-time, not evaluation-time (another difference from ruby, potentially even related?). It makes sense once it’s explained what’s going on, but it’s hardly obvious.

I don’t know enough about how closures are done to say that python is wrong, but it’s certainly less desirable and more surprising…

(for those keeping score, this makes ruby: 1, python: still heaps ;P)

P.S. I couldn’t resist mentioning the title of a proposed solution to this problem: For-loop variable scope: simultaneous possession and ingestion of cake

update: As matt just pointed out, the ruby port is different in that it uses i as an argument to a block. A more faithful translation would be:

for i in (0..9)
  q << (lambda {puts i})
end

Which has exactly the same outcome as python.

So there you go. It turns out closures are confusing. Who knew? ;P