End Tags with Attributes

[Maltysen]

The tokenizer test case format doesn't have attributes for end tags, and it seems you're supposed to just ignore them and give a parse error (you can see an example in test1 with "<h></h a='b'>").

However, looking at the Tokenization section of the whatwg parsing spec, I don't see any step in the state machine saying to drop the attributes from end tags.

In fact I see a line saying "When an end tag token is emitted with attributes, that is an end-tag-with-attributes parse error." which seems to indicate that you *do* emit end tag tokens with attributes.

Similarly, the table of parse errors just says that the attributes in end tags are ignored and don't make it into the DOM, which again to me suggests that they exist in the token, but not in the DOM.

Is this a deficiency in the tests or am I missing/misunderstanding some part of the spec?

[annevk]

What exactly do you expect to see tested? They only impact tokenization so that is what I would expect to see tested.

[gsnedders]

Even if they're there in the token, it's not particularly meaningful — from the point of view of the whole parser, there's no difference between never putting them in the token and putting them there and then ignoring them.

As such, it doesn't really make sense to actually test behaviour of them.

[Maltysen]

Yeah that's fair from the pov of like parser behavior

My concern is that I (and probably others) use these tests partly to validate our spec interpretation. So, it's confusing when the tests themselves deviate from the spec.

Ignoring attributes on end tags was easy, but the real problem was distinguishing meaningful errors from these test-specific deviations (this wasn't the only one, e.g self-closing on end tags). It made me uncertain if my interpretation was wrong or if its just an efficiency trade-off ya'll made.

Maybe the best solution is listing all deviations from a literal spec reading in the .md file?

[gsnedders]

Maybe the best solution is listing all deviations from a literal spec reading in the .md file?

That would seem reasonable.