From: Vincent Manis <vmanis@...>
Date: 2011-04-19T11:51:40+09:00
Subject: Re: Tail Call Optimization (Tail Recursion)

On 2011-04-18, at 18:42, Steve Klabnik wrote:
> Guido has said that any interpreter that implements TCO is not Python.

I recall that quote, and it totally mystifies me. Of course, he IS BDFL, and has the right to say what is and is not Python. But it seems like a very peculiar comment, and, based on some of the message traffic I saw at the time, he seems to misunderstand what this optimization is. There are good reasons (relating to the JVM, which I mentioned earlier) why one might decide NOT to make it mandatory, but his absolute refusal seems somewhat doctrinaire to me. 

If every environment in which Ruby (or Python) runs supported TCO, then there would be little reason to ban it. If that isn't practical, then a way of discerning whether it's supported in a given implementation seems appropriate.

That said, very few tail recursive methods would likely be written in Ruby, given the many other fine ways of looping. So my reason for wanting it has more to do with a small set of cases where it's more expressive than other techniques, rather than something that substantially changes the way we write programs. 

In reality, the more useful benefit of TCO comes when it's not a recursive procedure, but simply invoking another method, and thus using less stack. That leads, for example, to a very nice and clean way of writing state machines, for example. 

-- vincent