(Or: how to write an unreasonable and ergonomic test suite for meta tags for Django.)

Buttondown’s codebase is littered with things that are clever and bonkers. I am proud of these things, even as I recognize that they are probably unsafe and terrible ideas. Take expected_meta_tags.yaml, which is a file that looks like this (truncated version):

"/{user.username}/archive":
    "title": "{newsletter.name}"
    "meta[name='description']": "{newsletter.description}"
"/{user.username}/archive/{email.slug}/":
    "title": "{email.subject}"

You can probably suss out what’s going on here. It’s a mapping of routes to CSS selectors and expected values for those selectors. Here’s how it gets parsed and used:

    def test_expected_meta_tags(self):
        # Used in f-strings.
        user = FakeData.user()
        email = FakeData.email(user=user)
        newsletter = user.newsletters.first()

        cases = yaml.load(open("emails/tests/cases/expected_meta_tags.yaml"))
        for (route, selector_to_value) in cases.items():
            for (selector, expected_value) in selector_to_value.items():
                with self.subTest(route=route, selector=selector):
                    response = self.client.get(eval(f'f"""{route}"""'))
                    html = fromstring(response.content)
                    matching_elements = CSSSelector(selector)(html)
                    self.assertTrue(len(matching_elements) > 0)
                    [matching_element] = matching_elements
                    value = (
                        matching_element.get("content")
                        if matching_element.tag == "meta"
                        else matching_element.text
                    )
                    self.assertEqual(value, eval(f'f"""{expected_value}"""'))

This isn’t the cleanest code in the world, but you can kind of see what’s going on, right?

• I generate some fake data.
• I parse the YAML file, and interpolate the URL and expected values based on that fake data. ¹
• I create a request at that URL, parse the resultant URL, and verify that everything looks sane.

This is something that, if I ran into in a random codebase, I would probably be terrified. Using eval is an extremely bad idea. But I stumbled upon a really nice DSL for testing meta tags, which is an onerous and annoying source of regressions for me due to Buttondown’s weird mix of server-side and client-side rendering; future regressions can now be tested just by adding an entry to a YAML file rather than mucking around with the cssselect package.

And it’s overly clever, for sure. But writing clever code can be fun, and there are afternoons where the only way you’re going to get a task done is if you make it as fun for yourself as possible.

1. The pattern of eval(f’f”””{string}”””’) is probably the funniest and most terrifying thing I’ve ever stumbled upon. ↩

Liked this post? Follow me!

Twitter • RSS • Email