Regression tests are tests that prevent unexpected behavior that occurred in a non-ideal environment (e.g. production) from occurring again.

Let's pretend we are working on an image scaling library which has a flaw in it:

function scaleImage(img, scale) {
  return {
    width: img.height * scale,
    height: img.height * scale
  };
}

and we have a test suite for it:

describe('An image when scaled by 2x', function () {
  before(function scaleImg () {
    this.result = scaleImage({
      width: 25,
      height: 25
    }, 2);
  });

  it('scales to doubles the height and width', function () {
    expect(this.result.width).to.equal(50);
    expect(this.result.height).to.equal(50);
  });
});

Unfortunately, our test suite is naive and misses the issue. We release our library and after a week, we get a bug report that some images are being cropped.

This type of bug is known as a regression, as in "an expected behavior has been undone".

We could fix the issue and release it, but we should want to prevent the issue from happening again. This is known as a regression test.

First, we will write our test first which should fail, verifying we have reproduced the bug.

describe('An image with uneven dimensions when scaled by 2x', function () {
  before(function scaleImg () {
    this.result = scaleImage({
      width: 25,
      height: 30
    }, 2);
  });

  it('scales to doubles the height and width', function () {
    expect(this.result.width).to.equal(50);
    expect(this.result.height).to.equal(60);
  });
});

Next, we will patch the issue:

function scaleImage(img, scale) {
  return {
    width: img.width * scale,
    height: img.height * scale
  };
}

Now, our tests are passing and we have prevented the issue from occuring again.

While this concept seems intuitive upon explanation, it was an "a-ha" moment of discovery for myself a couple years into becoming a developer.