[tm-pubsubj-comment] Re: [topicmaps-comment] referring to a topic fromoutside a TM -- PURL

[Bernard Vatant <bernard.vatant@mondeca.com>]

[Thomas]

> "4.4.1 - Every PS Indicator in a PS DocSet shall be identified by, and
> retrievable through an unique canonical URI.
> This canonical URI is the corresponding PS Identifier, uniquely defined in
> the PS DocSet namespace."
> Note: ***Every*** PS Indicator, not only the whole set.

[Tom]

> I'm actually against making this mandatory, although it's good practice to
> do so when possible.  Here's why:
>
> 1) Right now, anything retrievable through such a URI must be read and
> understood by a person, there is no topic maps mechanism to let a processor
> understand the "meaning" of the PSI.

Of course it is that way - and that is exactly what PSIs are about and intended to be. The
processor is not supposed to "process" the PS Indicator, let alone "understanding" it, but
only to *identify* it.

> Being online does not seem to be an
> runtime requirement for processing, but rather a design time benefit for
> software and map creators. So any means by which a person can get the
> information should be ok.

It's not only a question of getting the information contained in the indicator, but to
make the correspondence one-to-one with the URI with no possible ambiguity. So "any means"
is maybe too open :))

> 2) It prevents people from creating useful PSIs if they don't have a stable website.

What's wrong with that? Of course if I don't have stable address I should not pretend to
create PSIs. And it pushes me towards a constraint. Am I serious about it? An important
thing about PSIs is trust and stability. Why should anyone trust my PSIs if I'm not even
serious enough to set stable URIs? Stability is critical.

> 3) It would seem to invalidate previously PSIs whose web site goes away for
> some unfortunate reason, even when the copies of the information are
> available on many other web sites.

Lars Marius has answered that, and I agree with his view.

> 4) If technology were to change and so the owner of a set of PSIs were to
> want to host it using some other protocol or technology, they would not be
> able to make the change, since that would break the existing PSIs.

If we want all that to work one day, maybe we should think about stability of protocols
too.
We made it for phone numbers, no?

> For these reasons, I strongly suggest that it be a matter of good practice
> rather than mandate to have the PSIs be retrievable from their URI.

I disagree. Making that would be simply changing the very definition of what a PSI is.

Bottom line: Making conformant PSIs will not be easy. It will make them all the more
valuable, because they will be untrusted with all the difficulty to set them. We are
speaking about serious stable binding points there. No fancy stuff allowed I'm afraid.

Regards

Bernard