Re: Catalog circularity

[Norman Walsh <ndw@nwalsh.com>]

/ Paul Grosso <pgrosso@arbortext.com> was heard to say:
| I think we're safer with some weasel words about "implementations
| may be smart, or they may let careless users shoot themselves in
| their feet" rather than "must detect the error."

That's the other side of the coin. I wasn't thinking about web
resources, I guess. The only "equality" test that we could employ
would be "the same URI" and that doesn't necessarily return the same
data each time, so it's impossible to test accurately. (You can get
both false positives and false negatives.) Thanks, Paul.

I propose that we address this situation by changing Section 5 to
read:

<section id="s.catfiles"><title>Catalog Entry Files</title>

<para>Applications conforming to this &standard; must provide some
(implementation dependent) mechanism to establish the initial list of
catalog entry files. This may be a preferences dialog, an environment
variable, an application properties file, or any other appropriate
mechanism.</para>

<section><title>The <filename>xcatalog</filename> Catalog</title>

<para>Entity and URI resolution generally occurs in the context of
some document. If the user has not specified an initial list of catalog entry
files and the context document is known, and the base URI of the
context document is available, applications should establish a default
catalog entry file by loading a document named
<filename>xcatalog</filename>.</para>

<para>For the purpose of retrieval, <filename>xcatalog</filename> should
be treated as a relative URI reference against the base URI of the
context document.</para>

<para>It is not an error if this file does not exist, cannot be
retrieved, or is not an XML Catalog as defined by this &standard;.</para>

<para>The entries in the <filename>xcatalog</filename> catalog entry
file apply only to the context document against which it was
located.</para>

</section>
<section><title>Catalog Circularities</title>

<para>It is possible for catalog references to be circular. Given the
dynamic nature of resources on the internet, it is not possible for
implementations to detect without the possibility of error whether or
not a circular reference actually occurs.</para>

<para>Implementations should detect circularity where possible.
If a circularity is detected, it must be treated as an error.
Applications may recover from this error by
indicating to the calling application that
no match was found.</para>
</section>
</section>

                                        Be seeing you,
                                          norm

-- 
Norman.Walsh@Sun.COM   | The First Amendment is often inconvenient. But
XML Standards Engineer | that is besides the point. Inconvenience does
Technology Dev. Group  | not absolve the government of its obligation
Sun Microsystems, Inc. | to tolerate speech.--Justice Anthony Kennedy,
                       | in 91-155