Re: Order By

At 07:11 PM 6/8/00 -0700, Steve Haflich wrote:

>You still need to split that hair, methinks.  _If_ it is indeed
>possible for an implementation to return both zero length and null
>strings (is it possible???) then the sort order must specify which
>sorts first.

it is certainly possible in WebDAV for an implementation to return both
zero length and null strings, so I agree with Steve.  We do have to make
the distinction.

I would argue for option 3, (sort order MUST be either first, or last, but
which one is left to the implementation.)  Looking just at the plausible
underlying implementations (Oracle, Postgree, DMA, and generic SQL), we see
that option 3 is the most specific common definition possible.  

The meta-principle I am applying here is that WebDAV should be inexpensive
to implement.  If we mandated either choice 1 or 2, it would make
implementation expensive on systems that sorted the other way.  A client
that really cares about the sort order of null can  accomplish this cheaply
at the client side, since the values (or lack of values) are in the reply.
Given a choice between making the server side implementation expensive, and
adding a small penalty to the client side, I say the client should pay,
because it spreads the work out more evenly.

Is there any objection to choice 3

Received on Friday, 9 June 2000 14:40:24 UTC