Fix and clarify more around CDATASection

[foolip]

• Make appendChild(cdata) and similar not throw.
• Let nodeName be #cdata-section.
• Include CDATASection as a node in an exhaustive enumeration.
• Collapse cases of Text+PI+Comment to CharacterData where possible
  so that it's more clear that subtypes are included.
• Say "depending on" instead of "switched" on.

Part of #300.

[foolip]

The normalize() stuff from #300 is not addressed here.

[annevk]

The reason CDATASection is not mentioned here is because ShadowRoot isn't either. We cannot have it both ways.

[foolip]

Hmm, so I suppose that this is then enumerating the least-derived interfaces that can be instantiated? Otherwise it'd collapse into just saying "Node" :)

[foolip]

OK, I guess it's "So the new policy is to list concrete classes (DocumentFragment) unless they’re a subclass of a concrete class (ShadowRoot) and not list abstract classes (CharacterData)."

[annevk]

We could do something else (e.g., also use abstract classes). But without whatwg/webidl#97 I'm not sure this can be improved much.

[annevk]

The only thing that seems like a real issue here is "<code>#cdata-section</code>".

[foolip]

I've split out nodeName into #302 but think some of what's left would be worthwhile, will rebase and comment inline.

[foolip]

OK, so after looking at what remained again, I'm left with just eb9252f

Feel free to merge or reject and move on :)

[annevk]

I prefer the term "switch" from "switch case".

[foolip]

I don't understand, "switch case" doesn't appear in the spec. Do you prefer "depending on the context object" or "switching on the context object"? Both are now used and I picked the one that I thought was less likely to be blindly translated to switch-case in C++.

[annevk]

I prefer the latter, but perhaps you're right that the current setup doesn't work. In particular branching on type seems wrong.

[foolip]

New issue, or do you have a plan?

[annevk]

Branding in IDL is still the only thing that resolves this I think. New issue to track that on DOM side seems fine.

[foolip]

I see some discussion of branding in whatwg/webidl#125, is there a larger tracking issue for whatever is involved?

[annevk]

No, that's it. I think we could clarify DOM meanwhile by putting CDATASection checks before Text. Since Text checks are true for both, but CDATASection will only be true for CDATASection.

[annevk]

(Are you okay with deleting this branch btw?)

[foolip]

Branch deleted, all that's needed to reproduce is to change either "depending on the context object" or "switching on the context object" to the other for consistency.

[foolip]

One relevant difference between CDATASection and ShadowRoot is that CDATASection has its own nodeType. So switching on nodeType actually works for ShadowRoot, but not for CDATASection. Since we'll presumably never add a new nodeType, actually spelling out whether CDATASection is included anywhere that Text is mentioned would remove the risk of confusion.