I remember reading this one article that basically says: let's assume that an algorithm for a function basically means doing something repeatedly until you get a result. Then an implementation of a function $f : A \rightarrow B$ really ought to be a set $X$, together with an initializer $A \rightarrow X$ and an "iterator" $X \rightarrow X + B$ that, given $x \in X$, either produces a new element of $X$, or else returns an element of $B$. The article proceeds to study this idea as a very simple model of what it means to be an algorithm for $f$.
Question. I'm trying to track this article down, but haven't been able to find anything. I can't remember any of the buzzwords that appear in the article, nor can I remember the author. E.g. the word "iterator" probably doesn't appear. Does anyone know what this article is called?
Remark. The definition might have been: an implementation of $f : A \rightarrow B$ consists of a set $X$ with a function $A \rightarrow X$, a function $X \rightarrow X$, and a partial function $X \rightarrow B$. I can't remember the details.