RFC6265bis: Use navigables concept

[sbingler]

Closes #2372

The HTML spec recommends avoiding using "browser contexts" and instead recommends "navigables".

This PR refactors rfc6265bis to use "navigables".

[sbingler]

@johannhof PTAL

@jakearchibald If you have any time. These concepts look like they should be drop in replacements but I'd appreciate your feedback just in case.

[jakearchibald]

Couple of nits, but LGTM

[jakearchibald]

Similar to above, I'd reword this to avoid assuming that a document is its navigable's active document. Or, if you're sure that this algorithm only runs when the chain of documents are active, add a line that asserts that.

[sbingler]

I'm pretty sure we want this only for an active document chain, and I've added a note at the bottom of the algorithm saying as much.

@mikewest does that seem correct to you?

[mikewest]

Are prerendered documents "active"? If not, we need to make sure we don't make it impossible to define their "site for cookies" by excluding non-active documents.

That said, I'm not sure we need to care about activity. The point of this algorithm was to walk up the ancestor chain from the document in which a given request was initiated and to do some origin comparisons. I think it should run on the set of ancestor documents, active or not. I'm assuming there's still a single path up the ancestor chain, and it's not clear to me that any ancestor can be inactive without the document we're talking about also being inactive. In that case, 🤷 .

That said, this definition should probably move to HTML at some point, and the RFC should depend upon the embedder's definition of terms. :)

[jakearchibald]

Are prerendered documents "active"?

Yes. I believe the goal is to spec them as a nested traversable, so it'll be the active entry of that traversable. I think @domenic has been thinking more about the spec here.

I think it should run on the set of ancestor documents, active or not. I'm assuming there's still a single path up the ancestor chain

Yes, you can get a document's container document https://html.spec.whatwg.org/multipage/document-sequences.html#doc-container-document

[domenic]

I think you want to use https://html.spec.whatwg.org/#inclusive-ancestor-navigables . See e.g. this call site for something that does exactly what you're doing: https://html.spec.whatwg.org/#the-autofocus-attribute:inclusive-ancestor-navigables

You kind of need to care about activeness because you need a non-ambiguous way of going from navigables (which are in a tree) to documents (which are not). Each navigable has a single active document and potentially many inactive documents; the latter are stored in bfcache. You want to pick out the active document.

E.g. given:

parent navigable: A --> [B] --> C
                         |
child navigable:      [child]

where [] denotes the currently-active document, you want the set to contain B, but not A or C.

[sbingler]

Thanks for all the feedback, I used the inclusive ancestor navigables concept. I also kept the note that all docs should be active, @mikewest do you have a preference to remove/keep it?

[mikewest]

I'm happy where this ended up. Will give it one more read-through before landing it.

[johannhof]

Thanks @sbingler, this looks conceptually right to me but I'll defer to @jakearchibald for detailed inspection :)

[sbingler]

@mikewest ping

[mikewest]

Yup. LGTM.

Thanks!

[mikewest]

If you resolve the conflict with the previous patch, I'll land it.

If you don't have time, I'll do it (since I made you wait so long!).

[sbingler]

Thanks Mike, conflicts should be resolved