So, the point seems to be whether the attribute identifiers should be opaque or not.
I'm arguing that it's pretty basic data modeling that all identifiers should be treated as opaque rather than as semantic keys.
I am not so sure where the example like OpenSocial or OpenGraph or Poco are using opaque identifier.
Web 2.0 and Computer Science 101 don't seem to be very closely associated, I've noticed, but your example seems to show parameters influencing the request, not overloaded identifiers (though the identifiers I see there don't seem to be suitable to a distributed environment).
can be trivially converted to
GET /rest/people/34KJDCSKJN2HHF0DW20394/@self?fields=name%23ja_Hani_JP,gender&fo rmat=xml HTTP/1.1 HOST api.example.org Authorization: hh5s93j4hdidpola for Japanese Kanji name instead of (probably) English name.
One problem seems to be with using a flat string or name/value set to make structured data requests. If all you have is a string, then everything has to get stuffed inside the string.
Same kind of procedures can be applied to OpenGraph, OpenID AX, etc.
Are you suggesting that we need to fix those protocols instead?
I'm simply pointing out that it's a bad approach and that it will bite people eventually.
Well, true. It may be a quick hack, but it seems to be somewhat more realistic than fixing all those protocols.
Hack away. -- Scott