Re: Request to register "identifier" relation type

[Herbert Van de Sompel <[e-mail]>]

On Aug 5, 2017, at 13:59, Ed Summers <[e-mail]> wrote:
> 
>> On Aug 5, 2017, at 3:38 AM, Herbert Van de Sompel <[e-mail]> wrote:
>> 
>> The answer is in the I-D and in a blog post that is referenced in the I-D. Please have a look. 
> 
> I think a crisp one or two sentence reply to Peter's question ought to be possible. He rightly points out that the shared use of 'preferred' by your I-D and canonical could cause some confusion for web publishers. It also begs the question of preferred *for what*.
> 

Good point. Here's my take on it:
- canonical: preferred for content indexing 
- identifier: preferred for referencing

> The core issue with canonical seems to be less a matter of semantics and more a practical matter of web publishers not wanting to assign a canonical or bookmark link to another domain (e.g. dx.doi.org) because of uncertainty about what this would mean for their Google juice. There is an SEO infrastructure built up around canonical, which has led to it being used quite a bit.
> 

Publishers indeed use canonical to point to pages on their own site.  This rel type is intended eg for reference manager applications: use the target URI not the context URI in a reference. Our research (referenced in the I-D) has shown that landing page URIs (instead of HTTP DOI) are frequently used for referencing. 

> I wonder if rather than adding another link relation to the mix if the HTML folks would be willing to update rel=bookmark to allow for usage with <link> which ought to make it amenable to use in an HTTP header as some other HTML relations are (e.g. alternate). The semantics of bookmark speak directly to the issue of persistence that your I-D seems to be addressing.
> 

Our I-D acknowledges that the semantics of "identifier" are similar to "bookmark". But "bookmark" is explicitly not allowed in <link> (and hence Link). One has to assume there were/are reasons for that choice because other rel types declared in HTML can be used in <link>.

> Also, while the name 'identifier' is elegant in some ways, I find it a bit hard to swallow since all link targets are technically identifiers.
> 

Indeed. Then again, this relation type indicates that the target URI should be used as identifier. We have done several presentations at conferences, workshops, etc over the past months in which "identifier" was floated. It's been on the Signposting web site (which has an associated mailing list) for almost a year, see http://signposting.org/identifier/ . We have not received negative reactions to the choice of term.  Some implementations are meanwhile already out there. 

Cheers

Herbert



> //Ed