In the sixth installment of my series on Scheme macros, we’re going to examine how to implement keyword argument functions in a single macro.
Keyword arguments are a very prominent feature of many programming languages, including Ruby, Python, Clojure, Julia, and OCaml (where they are called labeled arguments). If you don’t know any of these, think of keyword arguments as a way to name arguments on the call site, and optionally provide defaults should the caller not specify them. Let me illustrate this using Python:
def my_kw_fun(a, b, c=0): return a + b * c # a=10, b=3, and c defaults to 0 my_kw_fun(10, 3) # => 10 # a=3, b=10, and c=4 my_kw_fun(3, 10, c=4) # => 43
This is a very concise way to offer default values for some arguments––which I often find very convenient––and can result in much cleaner APIs. Of course it can also be abused, as any powerful language feature can, but it more often than not improves readability and clarity.
Let’s implement it in Scheme. A complete implementation can be found in zepto’s standard library.
We already used the macro in a previous blog post in this series, Deconstructing Classes. As a quick refresher, let’s look at how its API looks in zepto:
(defkeywords (my-kw-fun a b) (c :default 0) (+ a (* b c))) (my-kw-fun 10 3) ; => 10 (my-kw-fun 3 10 :c 4) ; => 43
The code in Figure 2 above is basically equivalent to the Python version in Figure 1. It is a little less concise, but functionally equivalent. This makes it simple to test and easy for our users to figure out!
Let’s try and figure how to implement this!
Today we’ll try to write a single, big macro. Most of this section—except a brief aside for a helper function—will be about
defkeywords, so buckle up! Don’t forget to take breaks and think through all of the steps; don’t be afraid to go back at any point if you lose track of anything.
As always, we’ll start with a simple macro skeleton.
(define-syntax defkeywords (syntax-rules () ((_ nargs kwargs body) ; definition goes here )))
As we saw in Figure 2 above, the macro will take three arguments,
body. But what do we do with them?
First let’s capture the environment so that we can add bindings to it. We need to do this because we will define a function that we build dynamically from the info we got. As we’ve seen before, in zepto we do this using
with-environment and passing the constructed value into
(define-syntax defkeywords (syntax-rules () ((_ nargs kwargs body) (with-environment env ; definition goes here ))))
We’ll now do two things: we’ll need to generate the function from the information we’re given, and somehow weave the keyword handling into it. Let’s try and build a function first, and see what extra work is required to get to the keyword arguments:
(define-syntax defkeywords (syntax-rules () ((_ nargs kwargs body) (with-environment env (eval (macro-expand `(define ,(reduce (flip cons) (cons 'args) (reverse 'nargs)) (begin ; what now? ,body ) ) ) env) ))))
Okay, so now we have an empty function in our macro. We need to
macro-expand it, and quasi-quote the body to use it as template text. At the end of that function we’ll call the actual function body, so we can add that already. But what in the world is that
Well, what we need to have is a dotted list with rest arguments. In plain English, this means that if our function signature looked like
(my-kw-fun a b) before, we need to rewrite it to
(my-kw-fun a b . args) to catch all of the extra arguments that the keyword function caller might pass into our function. To this end, we need to rewrite the list into a dotted list. In zepto, you construct dotted lists with one-valued
cons, in this case with
(cons 'args), which will give us the
( . args) we’re after. Then, we’ll prepend the arguments in reverse order until we arrive at our original list, but with a new element and a new type. This is all just a very fancy way of saying there are multiple weird list types in zepto, and it’s sometimes a little awkward to transform one list into another.
After we have that out of the way, let’s look at what we need to do. We now know where we’ll get those arguments, namely in a list called
args. They’re weird and untreated though, so we’ll have to transform them a little bit. Let’s try our hands at that, shall we? For the purposes of the next few figures, I’ll pretend that our templated function is a top-level thing, so you can ignore all of the cruft we already wrote:
`(define ,(reduce (flip cons) (cons 'args) (reverse 'nargs)) (begin (with-environment env (let ((dict (apply make-hash args))) ; ... what do we do here? )) ,body ) )
Fig. 6: More environment captures and a dictionary.
The next change is fairly straightforward as well: we’re capturing the environment again—a different environment this time, inside of the function—, anticipating the need to dynamically define something once more. Then we use a neat little trick to create a dictionary from the pairs of names and values that we’ve been passed. Remember, the rest arguments will be given to us in this format:
:key value. This means that we can take all of them and pass them into
make-hash as is, which will create a dictionary from the values it’s been given, grouping them up into pairs of two. Our keys will be the variable names, and the values will be, well, the values.
That’s great! Now we just have to define them:
`(define ,(reduce (flip cons) (cons 'args) (reverse 'nargs)) (begin (with-environment env (let ((dict (apply make-hash args))) (map ($ (let* ((k (car %)) (s (atom->symbol k))) (eval `(define ,s ,(if (in? dict k) (dict k) (eval (get-from % 1)))) env))) (quote ,(treat-keywords 'kwargs))))) ,body ) )
By Mimir’s beard! What did we do there?
To be fair, it has been a while since I wrote that macro and when I looked at that again, I wanted to kick my old self in the butt for not writing any comments—one of my nastiest code habits. I was eventually able to decypher it, however, and I will help you decypher it too!
map over the list of keyword arguments we’ve been given—except we’re using a little helper called
treat-keywords to make the keyword argument format uniform—we’ll get to that function in a second. These will be pairs of names and their default values (which default to
For each of these arguments, we’ll get the key
k and transform it into a symbol
s. We’ll then
define that symbol to be either whatever we find in the dictionary, which will be the values we’ve been passed, or the default value we’ve defined in the header.
This is actually all we need to do. At this point we’ll just let the body of the function run, and everything will be in place as we need it to be. Awesome!
For completeness’s sake, let’s look at
(defne (treat-keywords args) (case (length args) ((0) ) ((1) `((,(car args) nil))) (else (let ((key (car args)) (meta (cadr args))) (if (eq? :default meta) (++ `((,key ,(caddr args))) (treat-keywords (cdddr args))) (++ `((,key ,nil)) (treat-keywords (cdr args))))))))
treat-keywords does is split a list into a list of pairs of names and their defaults, or
nil if none was given. The first two cases in the
case form are base cases, the last one is a recursive case that checks whether the
:default key is there and either uses the next argument or
nil as the default. No black magic!
While there is a whole lot to take in here, we’ve just defined keyword arguments in just under 40 lines of code. Isn’t that pretty awesome? Yay us!
As always in this series, this implementation of keyword arguments is not exactly complete. We didn’t add any sort of error handling, and the API is fairly limited.
If you want to extend the functionality of this code, the best way to start would probably be to add some good, simple error messages so that your users actually understand what’s going on. To write good error messages, you’ll have to play around with the code a lot and figure out what messages would help you most if you miss something. What happens if the list of keyword arguments is malformed? What happens if the call to the keyword function tries to pass non-existant arguments? How about adding argument type checking1?
There’s a lot you could potentially add, but I think a simple clean abstraction with a good API is worth more than a giant macro that does it all—YMMV. Just go wild!
In this blog post, we implemented a fairly concise macro that adds keyword argument handling to your functions, similar in its API to many other languages, but not with much code at all!
I hope you enjoyed this installation of my little series! See you next time!
1. We’ll be writing a contract system in the next blog post of this series and it will do pretty similar things, so stay tuned!