"origin" missing from HTML spec

[jyasskin]

7bfd13c#diff-f329e3f05c7626cf44bf8ed92ba9f20df3c6a2bd7ef3f3f966f6583afa3508d2L55045 removed the "origin" term from HTML even though it's still defined at the same URL. This was the first run with reffy 10.3.0, and w3c/reffy#1120 looks maybe-related.

[tidoust]

Interesting. This is rather linked to w3c/reffy#1110, which introduced logic to eliminate duplicate definitions (feature released in the same version 10.3.0).

The term "origin" has more than one generic (i.e. not namespaced to any other construct) definition in HTML:

• The one you mention in Supporting concepts
• A second one in Processing [Link] headers

Our code just keeps the former one, and more or less silently discards the latter one (it outputs a warning in the logs but that's typically something we fail to look at).

Our assumption was that specs would never define the exact same term twice. There are a number (about 145) of such occurrences in HTML though ("origin", "ltr", "rtl", "geo", "captions", "descriptions", etc.). Looking at these terms, I see that all of them have only one exported definition, all other definitions being internal to the spec. That seems good.

So the valid assumption should rather be that there is always only one exported definition of a given term. If there are multiple internal definitions of the same term, that does not seem fantastic to me, but that's the spec's problem, and more importantly it won't crash ReSpec since it only considers exported terms. I'll update the logic in Reffy accordingly.

FWIW, there is actually one term that is exported twice in HTML: the span IDL attribute for HTMLTableColElement:
https://html.spec.whatwg.org/multipage/tables.html#dom-colgroup-span
https://html.spec.whatwg.org/multipage/tables.html#dom-col-span
... but I suspect the latter is a leftover from a copy and paste, and needs to be removed.

[johannhof]

Thank you both, this looks fixed to me now!

[tidoust]

Yes, fixed in Reffy and definition is now back in Webref and ReSpec's xref database. I also reported the duplicate exported span IDL attribute in HTML. Thanks for the report!