Re: URI schemes - is widget: OK, but xri: not?

On Aug 27, 2008, at 4:02 AM, timeless wrote:
> On Aug 7, 2008, at 5:28 PM, Roy T. Fielding wrote:
>> packaging
>> multiple representations into a single archive cannot rely on the
>> ability to rewrite references within the content of individual
>> parts because the individual parts may be cryptographically
>> signed before the package is created.
> There's a misconception here.


> A widget is a representation of an application existing on some system
> with no correlation to other resources.
> In a sense, a widget is more similar to for instance a Debian package
> than to a web page.
> Web technologies can be used to author these applications.
> There is no http or other network resource to which the resources in a
> widget correspond.
> Creating a widget does not involve "packaging a web page as a widget"
> it involves *writing* a widget.

Which is just a package of other things, some of which are independent
and may be signed.  It is exactly the same problem -- the only  
is in what you expect to be within the parts, which is orthogonal from
identifiers and referencing.

>> If a part is
>> created without its own URI, the cid scheme is the recommended
>> choice for minting new URIs within a package.
> webapps have looked at cid
> <>
> Content-ID and Message-ID Uniform Resource Locators
> We have found cid: unsuitable for the specification because it's based
> on single mime resources and the widgets specification is based on
> zip.

No, it is just defined in terms of MIME content parts.  There is nothing
to stop you from defining your own "parts", giving those "parts" some
sort of reference, and sticking it in a cid URI for cross-widget refs.
However, a hierarchical packaging format like zip already has a
catalog and built-in relative referencing, so you don't even need cid.
I would just make the widget a collection resource and use relative
references within that collection.


Received on Wednesday, 27 August 2008 22:38:40 UTC