[
Lists Home |
Date Index |
Thread Index
]
Greetings,
On Thu, 25 Jul 2002, Tim Bray wrote:
> Mike Champion wrote:
>
> > So. I'm now asking if we should create a URI scheme with these
> > properties.
>
> In general, we should not create URI schemes unless the benefit of doing
> so is very high, because the cost is very (amazingly, remarkably) high.
> The number of instances of software in the world that are expected to
> be able to process URIs is very high. Correct processing of a URI is
> dependent on its scheme. Thus introduction of new schemes carries an
> insanely high cost.
There's a similar remark (though in the context of URLs) in RFC2718.
Can anyone expand on the assertion that creating new URI schemes is very
rarely justified.
The project I work with is mulling over at least one URI scheme for
describing the rather structured results of astronomical observations,
and the components of the heavily structured files that contain them.
Timescales are such that we're still at the diagram-on-a-beermat stage,
but the advantages of a URI scheme seem sufficiently Obvious to us
that I'm rather alarmed at Tim's remark, since it seems to suggest
that there's some terrible show-stopping problem or cost that we've
simply missed completely and which, alerted by remarks such as this,
we've subsequently been able to uncover.
The data we're working with has features:
* It potentially exists in several replicas, perhaps one on the
local disk, one in a national archive, and one at the telescope
sitting in the middle of the Pacific. These are all equivalent, so
we want something which names the resource rather than addresses it,
so we can leave that last URI->URL step to some other resolver.
* Locating the files, and the contents within them, is very
naturally hierarchical.
* We'd want it to be possible at least in principle for components
within a structure to refer to other components within the same
structure, so relative URIs are natural.
* You might want just one component of a data product, so components
of a product have to be addressable as well, and not as a fragment.
* Though our software would be the first to grok the URI scheme, we'd
hope that other packages would in time -- that is, it can't be just
a private convention, but at least potentially standardisable.
* ...but not everyone in the world is interested in multi-Gb spectra
of unnamed stars, so we wouldn't be expecting Netscape to support
the scheme anytime soon.
This all sounds like just what URIs were invented for (to exaggerate
only slightly). It sounds very much like what URNs are for, but since
they're non-hierarchical and non-relative, they're out.
We could do something odd with URLs which point to redirections (like
PURLs), but that would be artificial, and lose at least some of the
benefit of the replicas.
There is indeed a connection with Grids, but as far as I'm aware (he
says cautiously), there isn't a completely straightforward solution to
this there.
Thanks for any comments,
All the best,
Norman
--
---------------------------------------------------------------------------
Norman Gray http://www.astro.gla.ac.uk/users/norman/
Physics and Astronomy, University of Glasgow, UK norman@astro.gla.ac.uk
|
