swears

a fixture generator inspired by vows

background

I found myself porting vows tests to tap. vows was the test framework that we used at nodejitsu, and all of the old code from then is tested with it. the truth is that we all hated it - we thought it was clunky and baroque as compared to tap (or heaven forbid mocha). but I also thought the core abstraction was being lost on us. perhaps that was having faith in our architects ;) but being in those tests today, I think I've identified the part of vows that's, if not good, then at least ok, sometimes.

vows structures its tests as batches of assertion suites executed against topics. these topics are effectively fixture generators, and the batches are effectively a group of tests that operate on the test's fixture. this is a structure common with behavior-driven development - "given ${TOPIC}, ${X} and ${Y}".

vows is extremely opinionated about this test structure, and if your mental model for your tests doesn't map to BDD very well then you'll have a bad time. this is really the core to why people got grouchy about vows. truth be told, the this-oriented DSL didn't age well either [1].

but suppose you are in a situation where you want to stand up a fixture or two and run a suite of tests against instances of that fixture. you could create and tear down the fixture using before/after hooks in your test framework, but I find that sort of thing is a little clunky when used to manage fixtures.

this is the thing that vows really shines at. it's able to specify how to create a test fixture, and the test harness manages creating and tearing down that fixture for each batch of tests. the problem, as it were, is that vows couples fixture creation with test suites and test harnesses.

what I did here was write a very lightweight abstraction which implements a 2022 spin of the idea of vows' topics, without the test harness. instead, you may use your test framework of choice - such as tap. this gets me the benefits of the vows setup/teardown DSL, while using my favored test framework, and in what counts as idiomatic javascript today.

enough blabbing, what does this look like?

ok - here's the first vows example ported to use swears:

const fs = require('fs');
const { open } = require('fs/promises');
const { promisify } = require('util');
const close = promisify(fs.close);
const write = promisify(fs.write);

const { test } = require('tap');
const { discuss } = require('../');

test('my first swears test', async (assert) => {
  assert.test('when we open a file', async (assert) => {
    const fileTopic = discuss(async () => {
      const file = await open('/tmp/fakefile', 'w');
      return file.fd;
    }, async (file) => {
      await close(file);
    });

    assert.test('it works', async (assert) => {
      await fileTopic.swear(async (file) => {
        assert.ok(file);
      });
    });

    assert.test('and we write to the file', async (assert) => {
      const writeTopic = fileTopic.discuss(async (file) => {
        await write(file, "My dog has fleas\n");
        return file
      });

      assert.test('it works', async (assert) => {
        await writeTopic.swear(async (file) => {
          assert.ok(file);
        });
      });
    });
  });
});

this example first creates a parent topic by opening a file (with surprisingly low-level APIs which must've been all they had when cloudhead was writing node), and the first test asserts that the file handle was successfully returned. then, a child topic writes to the file, and a child test asserts that it was successful. the fixture is created anew with each "swear" block, and the teardown hook is run after such blocks are run. this all happens without coupling the fixtures to the test harness, and without using a this-oriented DSL.

if you use typescript (like I do, on occasion), you might appreciate this example as well:

import { test } from 'tap';
import { discuss } from '../';

interface Discourse {
  spiceLevel: number;
}

test('when engaging with the discourse', async (assert) => {
  const topic = discuss<Discourse>(async () => {
    return { spiceLevel: 0 };
  });

  assert.test('and the discourse gets rowdy', async (assert) => {
    const spicierTopic = topic.discuss<Discourse>(async (discourse) => {
      discourse.spiceLevel++;
      return discourse;
    });

    assert.test('things get a little spicy', async (assert) => {
      await spicierTopic.swear(async (discourse) => {
        assert.equal(discourse.spiceLevel, 1);
      });
    });
  });
});

scenario aside, this is very similar to the file example. Note that discuss and Topic.discuss are parameterized by a type T, and that a child topic may return a topic of a different type U. in other words, swears should be type safe.

development

swears is written in typescript, but that should have minimal effect on this project, as run-scripts generally start by calling tsx. even so, be prepared for thinking in types!

install

nothing surprising here - npm i will do the trick!

test

npm test will run tsx and then execute test.ts and the examples with tap. you shouldn't ever accidentally run tests against old typescript, and the examples shouldn't break on accident.

build

most npm run-scripts will automatically call tsx as appropriate, but if you want to run it yourself, try npx tsx.

docs

I'm using exercise-bike to generate the README through a nunjucks template, which copies in the examples from the examples folder. basically: edit README.md.njk, then run npm run doc to update the README. note that this will automatically run before a publish.

license

swears is being released under an MIT license. for more info, check out the LICENSE file.

footnotes

[1] when vows was written, ruby-based DSLs were all the rage and everyone working in dynamic languages was trying to emulate them. it sure seemed like chainable APIs and binding this in callbacks was a similar trick to ruby's do syntax. vows, like almost every library from its time, discovered the hard way that this is a worse idea than it sounds. we live and learn!