Checking the result of image generation was quite difficult using JavaScript until pixelmatch appeared.

To test d3-composite-projections, I used to copy the approach of the d3-geo-projection, which used imagemagik and now graphicsmagick.

I found that comparing if the expected image was like the libary stopped working if I changed the computer. Slightly different versions of the libraries create different outputs that are visually identically.

Pixelmatch

Pixelmatch describes itself as

The smallest, simplest and fastest JavaScript pixel-level image comparison library.

It’s got a command line command and can be used from JavaScript on node or the browser, which is what makes it cool for testing.

The function we need is really simple:

pixelmatch(img1, img2, output, width, height[, options])

Where

  • img1 and img2 are the images data to compare. We’ll see how to read them
  • output is the place to write the image data with the differences (or null, when it writes nothing)
  • width and height are the dimensions of the three images
  • Options have a threshold option that allows the images to be slightly different, which can be great in some cases.

The output is the number of pixels that are different, so we’ll check if it’s 0 for testing.

I’m using tape to make the tests. Because d3js uses it and becase is simple and great.

The function would be:

tape("Checks the actual image outputs", async function(test) {
    projections.forEach(async d => {
        await render(d.name, d.topojson, d.field);
        let img1 = PNG.sync.read(fs.readFileSync("test/output/" + d.name + ".png"));
        let img2 = PNG.sync.read(fs.readFileSync("test/samples/" + d.name + ".png"));
        let diff = pixelmatch(img1.data, img2.data, null, img1.width, img1.height, {threshold: 0.0});

        test.true(diff == 0, d.name + " matches the sample file");
        
       });
    test.end();
});

You can check the whole script here

  • Note that there are many files to test, so a forEach is used to check them
  • We’ll see the render function, but it does what it seems, rendering the image
  • Note that PNG.sync.read is the method to get the generated images data

To create the images using canvas, the canvas library is used, so:

npm install --save-dev canvas

and require it using

{createCanvas} = require("canvas")

Then, just get the context using

context = canvas.getContext("2d");

and use it as in any browser example. Easy!

This file has the whole function.

Since everything is in JavaScript, node can run all the tests, for functions or rendering from the same tape call:

tape 'test/*-test.js'

Using ImageMagik or GraphicsMagik

If you are curious about the old method, the basic part was using something like:

compare -metric rmse test/samples/projection.png test/output/projection.png

This uses ImageMagik to output if the two images are identical. To run the creation, a bash script is run. Of course, this makes it difficult to run it in Windows, for instance.

Newer d3 versions use [GraphicsMagik][graphicsmagik] to do the same. The good thing is that the output is similar to pixelmatch. The bad one is that is a bash script anyway.

gm compare -type TrueColor -highlight-style assign -highlight-color red -file test/output/$i-difference.png test/output/$i.png img/$i.png;