Switch to automatic cross-references

[tidoust]

[This PR cannot be merged as-is. Some discussion needed on the treatment of the remaining errors]

Since the specification was written, authoring tools have made a lot of progress on cross-references support, and groups have become better at differentiating between definitions that should remain internal to a given specification (think "private" in object oriented languages) and those that the group commits to maintain over time and that can be referenced by other specifications (think "public").

This update uses the xref feature of Respec throughout the spec, and drops all indirect definitions that were in the Terminology section that can now be taken care of automatically by Respec. Some of them cannot be because the underlying specifications are not known to Respec.

Some of the definitions no longer existed (for instance the concept of user activation was modified in HTML), or specs are no longer the right ones (reference to Web Storage should now be a reference to HTML).

The exercise also reveals cases where the definitions exist but are not yet exported. Most of them are linked to the creation of the receiving browsing context algorithm. This is not surprising since this part cherry-picks algorithms in different specifications to define an "incognito mode".

In any case, these remaining errors signal the need to exchange with the groups responsible for these specs to ask them whether they can flag the definitions we need as "exported".

The definitions that the Presentation API references but that are not exported by the underlying specs are:

In HTML:

• parse a URL
• list of descendant browsing contexts
• creating a new browsing context
• session history
• allowed to navigate
• navigating to a fragment identifier

In IndexedDB:

• databases

Note this update also adds the IDL index and the index appendices that Respec generates automatically.

---

Preview | Diff

[markafoltz]

We could:

1. Ask other groups to export these definitions
2. Find out of there has been progress in defining an "incognito" browsing context in HTML we could reference, instead of defining our own
3. Remove most of these requirements for the receiving browsing context, leaving them up to the user agent
4. Leave things as-is

I vote for #4 for now, I think we need more input before deciding how to proceed.

[tidoust]

I would split the list into two.

First, the terms that the spec needs in other algorithms:

• parse a URL, used in Constructing a PresentationRequest
• list of descendant browsing contexts, used in Select a presentation display

For these, I think that we should reach out to the WHATWG and request that they get an export flag.

Second, the list of terms that the spec needs to define the "incognito" mode. Ideally, that definition would rather be in HTML and it does not make a lot of sense to export these terms from HTML. It may be worth checking whether there is interest in upstreaming the definition.

[tidoust]

I adjusted the PR to work around ReSpec's errors. The terminology section now contains:

1. terms that are defined in specifications that are not in the cross-reference database used by ReSpec, typically those defined in RFCs (e.g. Accept-Language)
2. terms that are not a direct match of the terms defined in external specification (e.g. trusted event)
3. terms that are "internal" to another spec. Ideally, that last list should disappear over time, since a spec should not have to reference internal concepts from other specs.

[markafoltz]

This looks good to me, thank you @tidoust for making these extensive cross reference fixes to the spec. It would have taken me a long time to track down all of them!