Santosh,
While I have no objection to making changes to the OpenID spec to support Webfinger if that's what folks want to do, I don't think it's strictly necessary. We could just have a very brief document that explains how to publish an OpenID identity via Webfinger and how OpenID RPs might use Webfinger to get the OpenID identifier. These steps happen before the current OpenID auth procedures start, which means we could keep it separate and fairly clean. I'm somewhat in the same position as you. I've implemented OpenID (OP side) and I use it, but I'm not a part of the team writing or changing specs. I'd be happy to engage with folks to do that, but not sure who is leading the work. Paul From: Santosh Rajan [mailto:[email protected]] Sent: Friday, March 26, 2010 12:29 PM To: Paul E. Jones Cc: [email protected]; [email protected] Subject: Re: WebFinger at Google Paul, I can see where you are coming from, and most people will agree with your suggestion for the rel value (even me). However I think the problem is with the "ground realities" at the moment so to speak. There are two problems with two solutions we have to look at. 1) How to integrate webfinger with the current OpenID 2.0 spec. 2) How to integrate webfinger with OpenID v.next. I am not too worried about case (2). It will happen in the future and there are a lot of competent people around here to sort that out. My suggestion for looking at webfinger as "normalizing an email like identifier for OpenID" is to allow email like identifiers to work with the existing OpenID 2.0 spec. To make webfinger work with OpenID 2.0 all we need is an addendum to the "normalizing the user supplied identifier" section (7.2) of the 2.0 spec, as explained in my earlier post. Also we need to acknowledge that the identifier returned by webfinger is not the claimed id, because their may be redirects. Also I must acknowledge that I am not an expert on OpenID or webfinger (only a user), and all this is IMHO. Santosh On Fri, Mar 26, 2010 at 8:33 PM, Paul E. Jones <[email protected]> wrote: Santosh, The identifier returned via Webfinger may or may not be the "claimed ID" since it might be user-entered and not yet normalized. However, I think a rel value of http://openid.net/identity would still suffice for both normalized identifiers and those which are not yet normalized. The value really should be considered "user provided", in my view. Once the value is retrieved, I think it should be given treatment just like any other ID entered into the OpenID login box. Should a person be allowed to enter an email address form and then return a different email address form in the identity field, thus forcing another Webfinger lookup? I can see an opportunity for abuse there, so I'd prefer the disallow it, but I could go with whatever is decided. Paul From: Santosh Rajan [mailto:[email protected]] Sent: Thursday, March 25, 2010 6:28 AM To: Paul E. Jones Cc: [email protected]; [email protected] Subject: Re: WebFinger at Google >From the OpenID perspective we have to see webfinger as a part of "normalizing the user supplied identifier". So the OpenID normalization process would go something like this given a user supplied identifier. (I will ignore XRI for simplicity) 1) Check to see if the identifier starts with http or https. If yes proceed as per protocol. 2) If not check to see if the identifier has an "@" sign within the identifier. If yes use webfinger to get the normalized identifier and proceed. 3) If not add http to the identifier and proceed. So really what webfinger returns is the normalized identifier, it is NOT yet a "claimed id" nor is it a "Local id". So I am suggesting one of these two rels. "openid.normalizedID". "http://specs.openid.net/auth/2.0/normalizedID";. On Thu, Mar 25, 2010 at 11:02 AM, Paul E. Jones <[email protected]> wrote: Jared, > It seems weird to return the user's OpenID identifier, when ultimately > the OP Endpoint URL is what you need if you want to authenticate the > user. However, I think "http://specs.openid.net/auth/2.0/server"; > should have been used for the rel type, as it is actually defined by > OpenID Authentication 2.0 spec for that purpose. I don't think it's weird at all to use webfinger to return one's OpenID identifier. After all, Webfinger is intended to be a means of discovering information about a person. Once the identifier is learned, then the OP can be discovered based on that ID. Returning the OP URL without the user's identifier is not as useful, since the OP would not know who is being authenticated: it would then have to prompt the user for his identity. > What is really needed is an agreed upon URI for what was the "http:// > specs.openid.net/auth/2.0/signon" type (which carried the user's > OpenID URL in XRDS' LocalID element (which is gone from XRD)). If the rel value is "http://openid.net/identity"; and the href value represents the user's OpenID identifier, then the RP knows what to do with that. I really think that's what we should try to agree upon. This would minimize the additional effort an RP would have to make, just adding a Webfinger resolution step and making no changes to the OpenID spec. The RP might want to implement Webfinger, anyway, in order to discover information about the user, such as his name, picture, or other information he wants to share with the world. Paul _______________________________________________ specs mailing list [email protected] http://lists.openid.net/mailman/listinfo/openid-specs -- http://hi.im/santosh -- http://hi.im/santosh
_______________________________________________ specs mailing list [email protected] http://lists.openid.net/mailman/listinfo/openid-specs
