[This response constitutes the official response to the comments by the W3C TAG to XRI Resolution 2.0 CD02]
Noah and other members of the TAG:
First of all, we appreciate your review of XRI Resolution 2.0 Committee
Draft 02  - it is clear from your questions that you've spent some time
reading this specification.
You asked about additional opportunities for feedback. The TC has just
incorporated feedback from the public review of Committee Draft 02 into
a revision on which it is currently conducting a vote to promote to Committee
Draft 03. We expect to commence another 15 day public review period on
Committee Draft 03 running from the middle to the end of March.
Following are answers to the questions you posed.
1) All access to resources identified by XRIs
require (at least) two round trips, the
first to retrieve metadata (XRDS, XRD or
uri list) and the second to retrieve
(a representation of) the resource itself?
First, we should clarify that XRI resolution defines two types of servers:
1) "authority servers", which serve metadata documents (XRDS documents)
describing resources directly, and 2) proxy resolvers, which provide a layer
of functionality that helps integrate XRI resolution infrastructure with the
WWW architecture. Proxy resolvers perform XRI-specific resolution logic in a
way which can hide XRI resolution semantics from the client of the proxy
resolver (e.g., a browser that is not XRI-aware).
The answer to your question depends on whether we are talking about
authority servers or proxy resolvers.
Authority servers play essentially the same role in XRI infrastructure that
nameservers play in DNS infrastructure. They serve up XRDS documents (which
can contain one or more XRD elements, each of which corresponds to an
authority subsegment, or a reference to one) in answer to resolution queries
from either a local resolver (one operating directly in an client's own
address space) or a proxy resolver.
In this case the set of resolution queries made to XRI authority servers to
resolve an XRI is directly analogous to the set of resolution queries made
to DNS nameservers to resolve a DNS name. There are as many of these
"roundtrips" as there are subsegments in an authority, minus one (e.g.
=gmw*foo*bar has four subsegments '=', 'gmw', '*foo', '*bar').
We are careful to suggest that caching semantics (both at the HTTP level and
*above* at the XRD level) be used aggressively to reduce these roundtrips,
just as caching is critical to good DNS performance. Also note that an
authority server can perform "lookahead" resolution and thus return results
for more than one XRI subsegment if it is asked (and the cached XRDs have
A proxy resolver, on the other hand, hides the set of interactions with
authority servers. It performs all these resolutions round trips on behalf
of a client (and hopefully builds up a substantial cache in doing so). A
client using a proxy server will use one round trip, typically, to obtain
metadata, and one trip to access a resource (in a RESTful manner, or
whatever manner is URI-identifiable). A common example of this pattern is
OpenID discovery, where a proxy resolver is used to request an XRDS document
from which the OpenID authentication service endpoint is selected in order
to redirect the user for OpenID authentication .
A proxy resolver may also be asked to redirect a client to a resource
directly (see question 2). While this is a separate roundtrip (a 302
response, typically), it is no different than any other HTTP redirect from a
client's perspective. The only difference is the manner in which a proxy
request URI is constructed.
Following are two examples of proxy resolver request URIs:
The "sep=false" media type parameter indicates that the proxy should
perform authority resolution on behalf of the client, but not perform any
service selection. The proxy will resolve "*example" against the authority
resolution server for "=", then resolve "*foo" against the authority
resolution server for "=example". It then returns the XRDS document
containing 2 XRD elements (one for "*example" and another for "*foo") to the
Request URI: http://xri.example.biz/=example*foo
This is a "Null Resolution Output Format" according to section 11.6 of the
specification. The proxy performs authority resolution as above. It then
performs service selection on the resolved metadata, and responds with a
HTTP 302 with the Location header value set to the highest priority URI from
the highest priority selected service. If service selection is unsuccessful,
it returns a non-302 HTTP response code. Since this is primarily a mechanism
for providing compatibility to applications with no native XRI support, we
recommend a human-readable error message to be returned in a text/* media
2) HTTP content negotiation can be used in
requests for XRIs to force either metadata
return or redirection to actual resource
Yes, local resolvers or proxy resolvers can do this on behalf of a client
(authority servers only handle XRDS documents).
As specified in section 11.7 of , a proxy resolver will return a redirect
if the request URI sent to the proxy resolver does not include a Resolution
Output Format paramater (_xrd_r) or the value is null. This allows XRI
authors to create a proxy resolver request URIs that can be used anywhere on
the web (e.g. on an HTML page) and will be resolved (via redirects) to a
resource representation just like any other non-XRI-related URL. Because of
constraints on POST/PUT, this redirect mechanism only works for GET/HEAD
For example, the following proxy resolver request URI could be used to
request a redirect to a server capable of returning a PDF representation of
The "sep=true" media type subparameter indicates the proxy resolver should
perform service endpoint selection using the media type requested; the
absence of an _xrd_r parameter means it must return a redirect as specified
in section 11.7.
3) Relative XRIs are of course allowed in the
normal way when a full-form XRI has been
established as the base URI. Are they also
allowed _without_ any full-form XRI as a
base URI? That is, for example, is "=henry"
intended to be recognize as an XRI in the
absence of any base URI? If so, what is
being done to ensure that both now and in
the future, the syntax of such abbreviated
XRIs is coordinated with (i.e. remains
disjoint from) the syntax of both absolute
and relative URIs that might be used in the
With regard to relative XRIs, the XRI Syntax 2.0 specification  followed
the lead of RFC 3986, in particular section 5.1:
5.1. Establishing a Base URI
The term "relative" implies that a "base URI" exists against which
the relative reference is applied. Aside from fragment-only
references (Section 4.4), relative references are only usable when a
base URI is known. A base URI must be established by the parser
prior to parsing URI references that might be relative. A base URI
must conform to the <absolute-URI> syntax rule (Section 4.3). If the
base URI is obtained from a URI reference, then that reference must
be converted to absolute form and stripped of any fragment component
prior to its use as a base URI.
So every relative XRI has by definition a base XRI. Section 2.4 of XRI
Syntax 2.0 defines how they are resolved as follows:
2.4.1 Reference Resolution
For XRI references in IRI-normal form or URI-normal form, resolving a
relative XRI reference into an absolute XRI reference is straightforward. If
the base XRI and the relative XRI reference are in IRI-normal form, section
6.5 of [IRI] applies. If the base XRI and the relative XRI reference are in
URI-normal form, section 5 of [URI] applies.
It is important that XRI references appear in a form appropriate to their
context (i.e., in URI-normal form in contexts that expect URI references and
in IRI-normal form in contexts that expect IRI references), since the
algorithms described in [IRI] and [URI] may produce incorrect results when
applied to XRI references in XRI-normal form, particularly when those XRI
references contain cross-references.
In contexts that allow a native XRI reference (i.e., an XRI reference in
XRI-normal form), it may be useful to perform relative reference resolution
without first converting to IRI- or URI-normal form. In fact, it may be
difficult or impossible to convert to IRI- or URI-normal form without first
resolving the relative XRI reference to an absolute XRI. The algorithms
described in section 5 of [URI] apply to XRI references in XRI-normal form
provided that the processor:
* treats the characters allowed in IRI references but not in URI references
the same as it treats unreserved characters in URI references (as required
by section 5 of [IRI]) and
* treats all characters within all cross-references the same as unreserved
characters in URI references (i.e., treats cross-references as opaque with
respect to relative reference resolution).
So to your question, "Is '=henry' intended to be recognize as an XRI in the
absence of any base URI?", the answer is no - it is only intended to be
recognized as a relative XRI in the presence of a base XRI.
It is true that if the same resource is identifiable with both one or more
base XRIs and one or more other base URIs of some other form, then a
relative reference in the context of that resource is relative to all of
these base identifiers. This appears to be no different for relative XRIs as
it would be for any other forms of URIs.
We hope these are answers are helpful.
Co-Chair, OASIS XRI TC
Gabe Wachob / email@example.com