Per

Typically, the special elements have the start and end tag tokens handled specifically, while ordinary elements' tokens fall into "any other start tag" and "any other end tag" clauses, and some parts of the tree builder check if a particular element in the stack of open elements is in the special category. However, some elements (e.g., the option element) have their start or end tag tokens handled specifically, but are still not in the special category, so that they get the ordinary handling elsewhere.

it seems keygen not being included there might be intentional.

Hmm yeah OK. It'd be ideal if someone like @RReverser or @inikulin or @zcorpan who knows the parser stuff inside out could confirm for us, and then we could add a code comment if it's intentionally not there

Relevant commits:
3b0abe6
2dbce39

I think this was an oversight.

Special is cross-referenced (normatively) in:

https://html.spec.whatwg.org/#parsing-main-inbody:special
https://html.spec.whatwg.org/#parsing-main-inbody:special-2
https://html.spec.whatwg.org/#parsing-main-inbody:special-4

These do not apply to void elements since they can't be on the stack here.

https://html.spec.whatwg.org/#parsing-main-inbody:special-3

Hmmmm. This ... algorithm has bugs, I think! Filed #2226. But this algorithm shouldn't have a difference if keygen is special or not (but making it special means you don't need to walk the stack to look for it).

So all void elements should be in the special list, as far as I can tell.

https://github.com/html5lib/html5lib-tests/search?utf8=✓&q=keygen - we could add some more tests for </keygen> in different insertion modes (but at least in-body).

I've never added html5lib tests before---could someone help with that? Maybe @gsnedders?

@domenic Is there anything much you need help with? Essentially, you have some input under #data, and some serialised expected tree under #document, and I think it's pretty obvious what that format is (there's a README that sorta defines it!). Pretty much ignore #errors for now.

I guess contributing to html5lib is not that hard, but I'm really unsure which scenarios need more testing here. @zcorpan, would appreciate your help if possible.