- From: Nathan <nathan@webr3.org>
- Date: Mon, 25 Jun 2012 14:06:35 +0100
- To: Harry Halpin <hhalpin@ibiblio.org>
- CC: Melvin Carvalho <melvincarvalho@gmail.com>, Kingsley Idehen <kidehen@openlinksw.com>, www-tag@w3.org
Harry Halpin wrote: > On Mon, Jun 25, 2012 at 2:40 PM, Melvin Carvalho > <melvincarvalho@gmail.com>wrote: >> On 25 June 2012 14:28, Harry Halpin <hhalpin@ibiblio.org> wrote: >>> On Fri, Jun 22, 2012 at 10:35 PM, Kingsley Idehen <kidehen@openlinksw.com >>>> wrote: >>>> On 6/22/12 1:02 PM, Harry Halpin wrote: >>>> >>>>> It seems like it would be better to use just mailto: (i.e., the >>>>> original WebFinger design was around *email addresses*, not accounts per >>>>> se) or a http: URI rather than mint a whole new scheme. >>>>> >>>> Here's why mailto: won't work. Most user agents will invoke an email >>>> client app. Remember, this is about using the URI as a name rather than >>>> data locator or access mechanism. That's the reason my acct: is required. >>>> Basically, follow-your-nose linked data traversal (with alternative URI >>>> resolvers hooked in via host-metadata pattern usage) without unintended >>>> consequences i.e., de-reference the next chunk of data etc... >>> >>> Just to be clear, WebFinger is optimizing around Linked Data >>> "follow-your-nose" pattern? Somehow I find this unlikely at this stage in >>> the game, but you never know. >>> >>> I thought it was just using the email address to track down a domain to >>> some info about capabilities supported by the domain, and thus was never >>> actually dereferencing the email address in a web-browser. >>> >> The ideas and authors have changed slightly in 3 years since its original >> form. WebFinger started by addressing a gap that additional information >> about an email address was not easily obtainable. More recently it has >> become a general discovery app. >> >> It uses a combination of "follow your nose" and the "well known" pattern. >> >> It's possible to use webfinger with any URI, e.g. mailto: http: etc. but >> the new scheme acct: is recommended. acct: at this point (and probably >> going forward) will be playing catchup to those that use HTTP (eg >> facebook). And there will be those caught in the world where they now will >> have two identifiers both acct: and http: to represent the subject of an >> email address. But Web Arch handles this already with owl : sameAs. >> >> > Again, what behavior is acct: supposed to invoke on the client-side? For > general purpose discovery, again, please stick to HTTP and well-known. From > De Jong's earlier posts, it seems the sole reason for the existence of this > URI scheme is to avoid a "host@domain BNF OR URI" and just have "a URI" > without a prescribed transformation into the a HTTP URI. > > In general, it sounds like folks on this mailing list, consisting mostly of > Linked Data aficionados, are mushing up a simple (albeit likely > wrong-headed solution) for specifying "account@domain") URI scheme with > general ideas of follow-your-nose in HTTP with a browser. Go figure. > > The meta-point is that unless a *new behavior* is specified, there should > not be a new URI scheme. That is why URNs are a failure, as is replicating > URN-like URI schemes for domain-specific purposes. There seems to be no > prescribed behavior with the accnt other than its usage as a parameter on > the server side to discover capabilities, which are then returned as JSON > (SWD) or XRD (Webfinger). Exactly, so make it a variable: https://gmail.com/.well-known/host-meta?acct=joe@gmail.com as RFC6415 allows under 3.1.1.1 instead of creating a new scheme. It delivers the functionality, and avoids the need for that BNF or creating a new acct: scheme. Or, as Melvin pointed out, "Simple Web Discovery" would handle the needed functionality also. Best, Nathan
Received on Monday, 25 June 2012 13:07:51 UTC