Re: @xml:base with @rendition (and maybe other pointers)

> On May 3, 2017, at 5:53 PM, John P. McCaskey =
<[email protected]> wrote:
>=20
> So, bottom line: A standalone fragment identifier refers to the loaded =
document and all the xml:base values above it in the hierarchy are =
irrelevant. Encoders cannot use xml:base to direct a standalone =
#fragment value to a location outside the loaded document.
>=20
> Is that right?

Yes and no.  For some purposes, the ways in which the answer is 'no'
are pedantic and can be ignored; for others they seem important.
I did not respond to your summary yesterday, because objecting to=20
the wording you used seemed unnecessarily pedantic.  Also, I=20
overlooked the second sentence, which I think is the wrong conclusion
to draw.=20

Given the following fragment of resource http://example.org/eg.xml

  <div xml:base=3D"http://www.dictionary.com/a.html">
    <p>
      <ref target=3D"#apple">Apple</ref>
      <ref target=3D"a.html#avocado">Avocado</ref>
      <ref target=3D"http://www.dictionary.com/a.html#anise">Anise</ref>
    </p>
  </div>

we can consider two ways of interpreting the target attributes.

Note that the discussion below ignores some possibly salient facts:

- URIs can denote different resources at different moments.
    (The discussion assumes the URI-resource mapping is not changing.)
   =20
  - A given resource can have multiple representations.  (The
    discussion ignores any resulting complications.)
   =20
  - URIs whose path component ends in .xml and .html do not
     necessarily have particular MIME types, so there is no guarantee
     that a fragment identifier like #apple will have similar
     meanings.  (The discussion assumes that fragment identifiers
     point to elements assigned IDs by the HTML 'id' attribute
     and/or the xml:id attribute.)    =20

Interpretation 1 ('xml:base values are irrelevant').  Ignore xml:base
when resolving '#apple' [but not when resolving other relative
references].  The target attributes are interpreted as denoting

    (a) http://example.org/eg.xml#apple
    (b) http://www.dictionary.com/a.html#avocado
    (c) http://www.dictionary.com/a.html#anise

None of these have anything to do with any of:

    (d) http://www.dictionary.com/a.html#apple
    (e) http://example.org/eg.xml#avocado
    (f) http://example.org/eg.xml#anise   =20

Of these, (a) is a same-document reference and RFC 3986 says it
"should" be dereferenced without a new retrieval action.  If a new
retrieval action is nevertheless launched, the resource retrieved is
(a).

[It is not clear to me whether the XVAI ('xml:base values are
irrelevant') interpretation takes a position on whether any of these
other than (a) are same-document references.]

Interpretation 2 ('same-document references can be surprising', or
SRCBS).  Resolve all relative references against the base URI in the
usual way.  Dereference same-document references either by looking in
the same document (as recommended by RFC 3986) or by launching a new
retrieval operation.

The target attributes are resolved to the absolute forms

    (d) http://www.dictionary.com/a.html#apple
    (b) http://www.dictionary.com/a.html#avocado
    (c) http://www.dictionary.com/a.html#anise

All of these are same-document references, so according to RFC 3986,
they should be dereferenced without a new retrieval action.  If a
retrieval action is nevertheless launched, it will go to URIs (d),
(b), (c) respectively, not (a), (e), (f).  =46rom the fact that (d),
(b), and (c) can be dereferenced without a new retrieval, it follows
(as far as I can tell) that these three resources can also be denoted
by URIs (a), (e), (f).

The XVAI and SRCBS interpretations agree on the following proposition,
which has important relevance for operations on the data:

P1 The relative reference target=3D"#apple" can be dereferenced by
locating the element in the current document with xml:id=3D"apple", if
such an element exists.

For people whose main interest is the truth or falsity of that
proposition, then, the answer is "yes, that's right" -- the effect is
the same, and all else is just pilpul.

=46rom P1, it follows (I think) that

P2 The resource identified by target=3D"#apple" is identified by the
absolute URI http://example.org/eg.xml#apple ((a) above).

The two interpretations disagree, or seem to disagree, on a number of
other propositions, most obviously:

P3 The relative reference '#apple' does not identify the resource
identifed by http://www.dictionary.com/a.html#apple.

XVAI does not actually entail P2, but it is compatible with P3.  (To
reach P3 it is necessary to assume some rule like "No two URIs
identify the same thing" or "If we don't know that a URI identifies a
thing, then it does not identify that thing.")  SRCBS entails the
negation of P3.

They also prescribe different URIs for the case that software
determines to perform a fresh retrieval action for the relative
reference #apple: XVAI prescribes the absolute URI (a), SRCBS
prescribes URI (d).

I don=E2=80=99t believe anyone has seriously suggested XVAI as the
relevant rule of interpretation for examples like the one given;
what I have suggested (and I have understood Hugh Cayless=20
to be agreeing with) is SRCBS.  Operationally, they can have
similar results in some circumstances (specifically:  they can=20
both result in no new retrieval action being undertaken in order
to dereference =E2=80=98#apple=E2=80=99), but they differ in ways which =
can be
critical.


********************************************
C. M. Sperberg-McQueen
Black Mesa Technologies LLC
[email protected]
http://www.blackmesatech.com
********************************************