[tei-council] @key to be deprecated?

[James Cummings]

On 19/09/11 16:12, Martin Holmes wrote:
> Hi James,

Hi Martin,

>> I'll play devil's advocate on this. I think using URIs (in the
>> form of URNs), even just locally constructed ones like
>> foo:blort:1234 is a much better system than just bare keys which
>> are just as much magic.
>
> Isn't foo:blort:1234 just magic too? If I've understood the proposal
> correctly, foo: and blort: don't resolve to anything meaningful; isn't
> foo:blort:1234 just a magic key that happens to have colons in it?

Yes, that is what I meant be 'just as much magic'. They are both 
magic. However, the URN-style magic key is a faceted one. (I know 
you could just make your @key value do this as well.)

> Backward-compatibility is the obvious one.

Yes, agreed.

> In that case, we're going to have an escalating tension between the
> Birnbaum doctrine and the need to clean up problems in P5 (like this
> one, perhaps, if you see it as a problem).

Perhaps, but the Birnbaum Doctrine doesn't say that we're not 
allow to break backwards compatibility, just that we should have 
a deprecation structure to do so.

> You actually caught me doing that (inadvertently) in an early version of
> the Image Markup Tool, IIRC.

Oh yes, I remember that, naughty Martin. :-P (He says quickly 
hiding any of his code where he certainly sins in greater orders 
of magnitude.)

> And I agree that's completely wrong when
> using @ref; but I would argue that's why @key is helpful. When you're
> still working out the structure of your repository and the relative
> locations of files and subcollections, not having to be precise about
> the path to a particular @xml:id is very handy.

Surely since you *can* do this ref="foo" then people will just do 
that while they are still working out their repository structure 
or system of magic keys?

> And if you get rid of
> @key, people are just going to use @n for the same job, I bet.

What people abuse @n for something that is not a potentially 
non-unique number or other label, but instead some magic token to 
identify specific classes of elements?  Never... no one would 
ever do that!  I mean that would be like using @rend to refer to 
_output_ rendition not source rendition. *grin*  Erm, yeah, ok, I 
see your point. People would certainly do that, yes.

I still think using a URN-like URI on @ref is better because it 
forces you to consider _some_ form of classification or 
documentation principle. But yours are all good arguments for 
maintaining @key.

-James

-- 
Dr James Cummings, InfoDev,
Computing Services, University of Oxford