Category errors


At the telecon you said:

> <masinter> I think there's a 'category error', mixing specifications
> of languages and their semantics with recommendations about best
> practice for operational behavior, and making the semantics depend on
> the operational behavior actually following best practice

I agree but would be interested to know what you mean specifically.

The httpRange-14 rule seems to fit this pattern. There are clearly two
camps, one that says that GET/200 is required to respect
identification semantics (MUST) and another that any such meaning is
only a result of someone choosing to apply recommended practice
(SHOULD) [1]. Backers of httpRange-14 fall in both camps and they take
different attitudes toward situations where the rule isn't followed.
Tim and others in the former camp, I and others are in the latter, we
all like httpRange-14, and we've been fighting about this question
unproductively for years (like the fabled circular firing squad).

If you're talking about HTTP 1.1: I did a quick survey and it appears
all of the REST-ish semantics in RFC 2616 - that is, anything that
talks about the resource and therefore might be at the same level of
analysis as httpRange-14 - are under SHOULDs, not MUSTs.  The PUT
method description is a good example of "here's what we meant it for,
but who are we to say".

Link:, site-meta, and description resource discovery are all at the
protocol level - they take no hard position on what the protocols are
for or what any resource "is", leaving that up to their applications.
There may be intent but I don't yet see a category error in how
they're being prosecuted.


[1]  - just
as pertinent now as it was then!

Received on Tuesday, 10 February 2009 15:08:52 UTC