Day 6: Declarative APIs, easy peasy with Raku

Raku APIs tend to be easy to read, with named arguments alleviating the need to remember argument order in method calls, for example.

But sometimes a library author goes above and beyond to produce extra nice, declarative APIs. One example is Cro, a framework for writing HTTP-based services, which allows you to write things like

my $application = route {
    get -> 'greet', $name {
        content 'text/plain', "Hello, $name!";

to declare your routes, that is, callbacks that Cro calls for you and which Cro calls for you when the user requests URLs that match the the pattern introduced by the routes, in this example /greet/fido or so.

Today, we’ll explore the mechanics that make such declarative APIs work, that is, APIs that read naturally and that use as little boilerplate as possible.

We’ll explore how you can enable similar interfaces for libraries you write.

Declarative API Fundamentals

The example above relies on a few major ideas:

  • bare words like route, get and content are just functions with the same name, and you can call them simply with their name followed by a space. The rest of the statement is interpreted as arguments to these functions.
  • In route { ... }, the { ... } is a Block, that is, a piece of code like a function.
  • Likewise, -> 'greet', $name { ... } is a block, this one with an explicit signature (the 'greet', $name part). The library code can introspect this signature, that is, find the name of the parameters ($name) and the value of the constant string 'greet'.
  • There is also an invisible mechanism that ties the get to the outer route { ... } block.

This last point needs some more explanation. In Cro you could have multiple independent route { } blocks, like so:

my $app1 = route {
    get -> 'meet' { content 'text/plain', 'Nice to see you' }
my $app2 = route {
    get -> 'greet' { content 'text/plain', 'Oh hai' }

How does Cro know that the meet callback belongs to $app1 and greet belongs to $app2? The route subroutine needs to call the block passed to it to find out what callbacks it declares, so it needs to inject some kind of context into the block. The way to do that is through a dynamically scoped variable.

In Raku, you can do that by declaring a variable with * after the sigil.

sub outer(&callback) {
    my @*DYNAMIC;
    return @*DYNAMIC.list;
sub inner() {
say outer(&inner);

Here sub outer declares a dynamic variable @*DYNAMIC. All called that is run until outer finishes can see that variable, including the inside of inner, which is bound to the parameter &callback. Thus, the code prints [42].

If you look at Cro::HTTP‘s definition of sub route, you can see that it uses basically the same trick, except that initializes the dynamic variable with an empty RouteSet instead of an empty array.

Getting Practical

Suppose you are writing a library that observes a directory tree, and you can configure it synchronize flies to another local, or to automatically delete them based on certain properties, or call your code when certain files change.

You want to provide an extra awesome, declarative API like this:

my $syncer = directory 'Documents', {
    watch name => /.*/, -> $file { say "File $file changed" }
    delete name => /\.swp/;
    delete name => /\.swo/;
    delete age_days => * > 5; 
    sync extension => 'txt';

To get this example to compile, you just need to declare the four functions directory, delete, sync and watch with appropriate signatures:

sub delete(*%conditions) {}
sub sync(*%conditions) {}
sub watch(&callback, *%conditions) {}

sub directory(Str $path, &callback) {}

Of course, you also need to capture the conditions and the callback in a data structure so that your hypothetical library can do something with it.

This could be an enum to store the action type, and a class for the condition and the optional callback:

enum Sync::Action <Delete Sync Watch>;
class ConditionalRule {
    has Sync::Action $.action is required;
    has %.conditions;
    has &.callback;

Plus a class that stores the directory and a list of ConditionalRule objects:

class Sync::Spec {
    has Str $.path;
    has ConditionalRule @.rules;
    method add(ConditionalRule $r) { @.rules.append: $r }

Finally, the four functions we started with need to be fleshed out. directory creates a Sync::Spec object and then calls its callback:

sub directory(Str $path, &callback) {
    my $*SYNC =$path);
    return $*SYNC;

The other three need to create new ConditionalRule objects, and add them to $*SYNC:

sub delete(*%conditions) {
sub sync(*%conditions) {
sub watch(&callback, *%conditions) {

This is annoying boilerplate, but it allows the user of the pretty interface to forego all the boilerplate.

Once you piece all of this together, directory returns a Sync::Spec object that holds all the knowledge necessary to fuel the hypothetical syncer library.

All that is left is actually implementing it. A task well outside the scope of this article — left as the proverbial exercise to the reader, should you chose those.

But wait, we aren’t quite done yet, because when somebody misuses our neat little API. If you just call delete outside of a directory block, you get the error Dynamic variable $*SYNC not found, which is not worthy of the awesomeness we aspire to.

Luckily, we can improve that easily:

sub delete(*%conditions) {
    die 'delete outside a directory { } block'
        unless defined $*SYNC;

… and analogously for the three other two actions. Again more boilerplate, inline with Raku’s motto of tormenting the implementer on behalf of the user.

3 thoughts on “Day 6: Declarative APIs, easy peasy with Raku

  1. Cool article. One question: why not use with in sub delete?

    with $*SYNC {
    else {
        die 'delete outside a directory { } block'

    Liked by 4 people

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: