Current conversion rules from CRIs to URIs don't peek into the first path segment -- but they should: [1, ["foo"]] can be converted to foo, but [1, ["foo:bar"]] needs to be expressed as ./foo:bar to avoid ambiguity against scheme "foo".
I think such a rule is missing (but I don't quite get all of the implications of the various "..., the conversion fails" statements, as they're expressed in terminology of the target syntax and thus very hard to apply to the object one has at hand when interpreting them).
Alternative is to fail conversion for such references altogether (but given it is a goal to convert URI references where possible, we should try to make it happen -- even though I'm not a big fan of that goal itself).
Current conversion rules from CRIs to URIs don't peek into the first path segment -- but they should:
[1, ["foo"]]can be converted tofoo, but[1, ["foo:bar"]]needs to be expressed as./foo:barto avoid ambiguity against scheme "foo".I think such a rule is missing (but I don't quite get all of the implications of the various "..., the conversion fails" statements, as they're expressed in terminology of the target syntax and thus very hard to apply to the object one has at hand when interpreting them).
Example is in the lastentry of https://github.com/core-wg/href/pull/47
Alternative is to fail conversion for such references altogether (but given it is a goal to convert URI references where possible, we should try to make it happen -- even though I'm not a big fan of that goal itself).