Functions in all our languages thus far take only a single parameter. What that may seem limiting, it’s actually not. Currying or curried functions is a way of constructing multi-parameter functions using only single parameter functions. If you’re a Haskell programmer, currying is something you do all the time. Look at the signature of an operation like plus3 that adds three Ints as an example:
plus3 :: Int -> Int -> Int -> Int
Usually we think about such a function as a mapping from three Int arguments to and Int result because of our upbrining in traditional languages. However, what it literally defines is a function from the single parameter Int to another function of type [Int] -> [Int] -> [Int]. Applying plus3 to a single argument results in a function:
(plus3 1) :: Int -> Int -> Int
Parenthesis added to the left side for emphasis. Applying another parameter results in:
((plus3 1) 2) :: Int -> Int
and another:
(((plus3 1) 2) 3) :: Int
and we now have a final result. But the function resulting from plus3 1 is just as much a result as 6. This is something we do all the time in Haskell to define and use partially instantiated functions:
plus2 = plus3 0
In FBAE and subsequent languages currying works the same way. We simply use application of a function to an argument. Given that we had defined plus3 in FBAE, the previous function application would look like this:
(((plus3 1) 2) 3)
It’s a bit messier, but it works just the same an emphasizes the currying that’s happening.
We can easily write plus3 using bind and nested lambdas. Let’s walk through it starting with the bind of plus3 to a lambda with one argument, ‘x’:
bind plus3 = lambda x in ?? in
...
Now we need to figure out what to replace ?? with. Literally, we want to add x to the result of adding two more arguments. Let’s write that much:
bind plus3 = (lambda x in x + ??) in
...
Now we need to get the next argument. The only way to do this is to use another nested lambda. Remember, that plus3 1 should return a function. What should immediately follow in should be a function and that function needs to know about x:
bind plus3 = (lambda x in (lambda y in x + y + ??)) in
...
See where we’re going? If we evaluate plus3 1 what we’ll get is lambda y in x + y + ?? waiting for the y parameter. Another lambda will pick up the third integer:
bind plus3 = (lambda x in (lambda y in (lambda z in x + y + z))) in
...
Just what we want. The bind now just needs a body where plus3 is used. Let’s copy what we did earlier:
bind plus3 to (lambda x in (lambda y in (lambda z in x + y + z))) in
(((plus3 1) 2) 3)
Now we can walk trough the evaluation of the bind body tracking the environment along the way:
(((plus3 1) 2) 3)
== ((((lambda x in (lambda y in (lambda z in x + y + z))) 1) 2) 3) []
== (((lambda y in (lambda z in x + y + z))) 2) 3) [(x,1)]
== ((lambda z in x + y + z))) 3) [(x,1),(y,2)]
== x + y + z [(x,1),(y,2),(z,3)]
...
== 6
We could add additional syntax to make both application and definition of n-ary functions simpler. We’ll leave a bit of that to an exercise.
Curried function semantics is used frequently to define multi-parameter function execution. Haskell certainly uses this approach as does ML. Scheme and Lisp on the other hand require explicit currying. We’ll not follow that path right now.