Testing your Application: explain why interdependent tests are bad

In https://github.com/strongloop/loopback4-example-getting-started/pull/2, the proposed acceptance tests are relying on the first test to fill the database with data that's used by subsequent tests:

  it('creates a todo', async () => {
    item = (await client
      .post('/todo')
      .send(payload)
      .expect(
        Object.assign({}, payload, {
          id: 1
        })
      )).body;
  });

  //...

  it('successfully deletes todos', async () => {
    await client.del(`/todo/${item.id}`).send();
    await client
      .get(`/todo/${item.id}`)
      .send()
      .expect(404);
  });

(see https://github.com/strongloop/loopback4-example-getting-started/pull/2#discussion_r147389550)

This is an anti-pattern to avoid:

• if the first test fails, all subsequent tests fail too (because data was not filled in), but with an unhelpful error message is confusing.
• it is not possible to run individual tests on their own, e.g. via it.only() or mocha -g "test name".

We should extend Data Handling in Testing your Application to mention this pattern and explain why it's a bad thing to do.

Start with the Data Handling in Testing your Application section linked in the issue, then read the referenced example tests and pull request discussion for context. Update the documentation to explain why interdependent tests are problematic and how independent tests support isolated execution; done means both failure cascading and it.only()/mocha -g usage are covered.