Zig NEWS

Arya-Elfren
Arya-Elfren

Posted on

Methods

I have corrected some small inacuracies in the first post and uploaded the code to GitHub.

Let's make our binary tree structure also encode some behaviour.

Namespaced functions

Functions placed in a struct, or in a file (as we saw they're equivalent), are namespaced within that struct. For example, the std.testing.expect function we used in our tests last time. This is a convenient and logical place to put functions related to our data structures. So let's use this to implement some simple traversals for our binary tree.

So, in our Node.zig file let's create a function that writes the preorder representation of our tree to a buffer.

pub fn traversePreorderToBuffer(self: *Node, buf: []?u8) void {}
Enter fullscreen mode Exit fullscreen mode

We take a pointer to our current Node and a slice of nullable u8s. This is so that we can go through (without an index) and find where we should insert the next item.

For preorder we first append our current data to the buffer. We do so by looping over it and finding the first slot that isn't null. We set this to the data in our current node and break. Then we recurse on the left and right subtrees, if they're there.

pub fn traversePreorderToBuffer(self: *Node, buf: []?u8) void {
    for (buf) |*slot| if (slot.*) |_| {} else {
        slot.* = self.data;
        break;
    };
    if (self.left) |left| Node.traversePreorderToBuffer(left, buf);
    if (self.right) |right| Node.traversePreorderToBuffer(right, buf);
}
Enter fullscreen mode Exit fullscreen mode

Let's write a test to see how to use this:

const expectEqual = @import("std").testing.expectEqual;

test "traversals" {
    var left = Node{ .data = 0 };
    var right = Node{ .data = 2 };
    var root = Node{
        .data = 1,
        .left = &left,
        .right = &right,
    };

    { // isolate the preorder test so we can add other traversals later
        var out = [_]?u8{null} ** 3;
        Node.traversePreorderToBuffer(&root, out[0..]);
        try expectEqual(out[0], 1);
        try expectEqual(out[1], 0);
        try expectEqual(out[2], 2);
    }
}
Enter fullscreen mode Exit fullscreen mode

First we create the binary tree (1 (0) (2)) and a length 3 buffer to store the output. Then we call the traversePreorderToBuffer function from the Node namespace, pass a reference to our root node and a slice of our buffer. We then use the testing.expectEqual to check our result (instead of expect(a == b) from last time, we just do expectEqual(a, b)).

Method syntax sugar

As I mentioned at the start, this is the most convenient and logical place to put functions that use our struct. It turns out that we frequently want to implement functions that look like this:
Node.zig

pub fn traversePreorderToBuffer(self: *Node, buf: []?u8) void {}
pub fn traverseInorderToBuffer(node: Node, buf: []?u8) void {}
pub fn traversePostorderToBuffer(node: *const Node, buf: []?u8) void {}
Enter fullscreen mode Exit fullscreen mode

However, it's a pain to always write Type.function(instance, ...) or having to import and rename every function to shorten it. Especially since we know it will always be @TypeOf(instance).function(instance) (or @TypeOf(instance.*).function(instance)).

Zig lets you call these functions with dot syntax as if they were methods. Effectively, if the first argument to a function that is namespaced in a container is that container (or a pointer to it) you can use the instance as the namespace.

So, for example, we can call traversePostorderToBuffer like this (in the above test):

{
    var out = [_]?u8{null} ** 4;
    root.traversePostorderToBuffer(out[0..]);
    try expectEqual(out[0], 0);
    try expectEqual(out[1], 2);
    try expectEqual(out[2], 1);
    try expectEqual(out[3], null);
}
Enter fullscreen mode Exit fullscreen mode

Conclusion

This time we learned how to give our binary tree behaviour and how to easily access it through dot notation.

Latest comments (0)