Enzyme: Selector regression in v3 - bare child selector

[realityking]

Trying to migrate our tests to Enzyme 3 I seemed to have discovered a regression. A bare child selector, e.g. > a does not work anymore while working fine with v2.

I'm not sure what the preferred format for a test case would be, please let me know.

[aweary]

Thanks for the report @realityking! > a is not a valid selector, so if it was working it was working by mistake. Selectors can't start with combinators. @ljharb are you OK with accepting this as a bug fix? I think changing the grammar to support leading combinators would be tricky.

[ljharb]

I think matching the spec is better. In other words, if > a is invalid, then it should never have worked, and we should take the opportunity with the major bump of v3 to shed that baggage.

[aweary]

Sounds good. The formal grammar in the spec states that a selector must start with an actual selector, followed by an optional combinator. You can verify this selector is considered by the browser parser by running document.querySelector('> a')

Sorry for the churn @realityking, but you'll have to update your test logic to avoid selectors with leading combinators. Thanks!

[ljharb]

Seems like * > a would work though?

[aweary]

@ljharb that should definitely work, and we do parse it correctly. I just need to implement the behavior https://github.com/airbnb/enzyme/issues/1158

[realityking]

@aweary Thanks. I figured that would be the answer 😉 Maybe worthwhile to update the migration documentation?

@ljharb Unfortunately it's not exactly the same. Consider this example:

- div
-- ul
--- li
---- ul
----- li
----- li
--- li
---- ul
----- li
----- li

The selectors > ul > li and * > ul > li would return different results. We actually used a > ul > li selector in testing our menu component.

I had in the back of my mind that querySelectorAll had a similar problem and that I had seen drafts for a solution. Indeed there is :scope in Selectors Level 4 (see https://developer.mozilla.org/en-US/docs/Web/CSS/:scope). Maybe that'd be worth implementing 🙂