Cross-spec section links failing (after version 3.10.0)

[svgeesus]

Cross-spec section links, of the form [specname#id] are failing in Bikeshed and generating fatal errors which must be suppressed with the -f option. For example (on the IFT specification):

chris@SuperNomad:/mnt/c/Users/chris/Documents/GitHub/IFT$ bikeshed -f spec
LINE ~301: Couldn't find section '#name-get' in spec 'rfc9110':
[[rfc9110#name-get]]
LINE ~1093: Couldn't find section '#name-field-definitions' in spec 'rfc9111':
[[rfc9111#name-field-definitions]]
LINE ~1125: Couldn't find section '#name-accept-encoding' in spec 'rfc9110':
[[rfc9110#name-accept-encoding|Accept-Encoding]]
LINE ~1241: Couldn't find section '#name-field-definitions' in spec 'rfc9111':
[[rfc9111#name-field-definitions]]
LINE ~1259: Couldn't find section '#name-content-encoding' in spec 'rfc9110':
[[rfc9110#name-content-encoding|Content-Encoding]]
LINE ~1322: Couldn't find section '#name-freshness' in spec 'rfc9111':
[[RFC9111#name-freshness]]
LINE ~1460: Couldn't find section '#name-accept-encoding' in spec 'rfc9110':
[[rfc9110#name-accept-encoding|Accept-Encoding]]
LINE ~1602: Couldn't find section '#name-content-encoding' in spec 'rfc9110':
[[rfc9110#name-content-encoding|content encodings]]
 ✔  Successfully generated, but fatal errors were suppressed

Testing out the links manually, for example https://www.rfc-editor.org/rfc/rfc9110#name-get shows they are correct. (I don't know if this is relevant, but all the failing specs are the HTTP set, such as RFC 9110, HTTP Semantics).

Originally reported:

• Can't generate html w3c/IFT#143

where it was also reported that

something seems to have changed in bikeshed at the latest version. For now I've been able to get it to compile without errors using bikeshed 3.10.0

[tidoust]

This is probably due to the switch from Shepherd xref data to Webref xref data as a source. In Webref, RFC9110 gets crawled from https://httpwg.org/specs/rfc9110.html and not from https://www.rfc-editor.org/rfc/rfc9110

Unfortunately, the version on httpwg.org does not use the same IDs for anchors. For instance, #name-accept-encoding in the www.rfc-editor.org version becomes #field.accept-encoding in the httpwg.org version.

[svgeesus]

Unfortunately, the version on httpwg.org does not use the same IDs for anchors.

HTTP WG breaks url consistency. Lol.

BTW the Bikeshed-generated html uses the rfc-editor links (which work). Odd.

[tabatkins]

Yup, that's the issue. Bikeshed will silently assume you know what you're talking about if the spec is in biblio data but not in headings data, but if it is in headings data, it verifies you're using the known IDs.

Hm tho, ideally SpecRef and WebRef are using the same URL for RFCs. It looks like SpecRef uses a mix of both origins, unfortunately.

[tabatkins]

(Closing as this isn't a Bikeshed issue.)