Re: Explain-lite (Explain discussion)

At 14:39 16-11-00 -0500, Robert Waldstein wrote:

>EXPLAIN is for clients.  My hope for EXPLAIN was that servers would be asked
>to populate fields that it had no idea why a client would want, and thus
>support client functionality beyond the common.

Always my hope as well. Problem with current Explain to me is that there is 
very little overlap between how different servers choose to describe 
themselves - supported attribute combos an TermLists in particular is an 
area where Explain is *way* too flexible for its own good. This translates 
into way too much complexity for Explain-aware clients that need to 
second-guess the server's thinking.

>But even without this hope
>I wonder how you defined "common" - I only use 7 or 8 fields, I hope all and
>only those are "commonly used".
>   Z39.50 does have this ongoing problem of appearing complex - but in
>EXPLAIN like Z39.50 you can ignore all but what you actually need or care
>about - the rest is noise, and easy in this case to ignore...

Absolutely... Bit Z39.50 has a more or less well-defined "core" of 
functionality that we all support. This focus  is lacking from Explain.

>And XML is going to be different why??  TO me that is no different than a
>change to the ASN.1 definiton.

XML is currently a much sexier buzzword than ASN.1 - therein lies the main 
difference. But other than that, I agree with ONE-2 on that one. Casting 
Explain in ASN.1 rather than GRS.1 in the old days was a huge mistake 
because it complicates subsequent extensions to the definition. Since 
Explain doesn't have to require any GRS.1-specific functionality 
(multilingual strings possible exception), I am totally happy doing it in 
XML to lower the bar for newcomers.

> > - our proposed use is to carry this as an XML external on the Init
> >   service means that none of the explain database or the explain querying
> >   has to be supported;
>
>I have alot of trouble with this.  So I have approx 300 databases, all with
>their own layout.  I send this everytime to every client? CHecked - it is
>12 Meg.  And note my "common" is probably smaller than yours - certainly
>BER/ASN.1 is smaller than XML.  And no caching - since doing it in init?
>   I always thought this was a clever part of the EXPLAIN model - clients
>ask for just what they need.

This has *got* to be a negotiated exchange - I assume the server doesn't 
spit all this stuff in the face of the client unless it receives an 
OtherInfo (or whatever) in the Init Request asking for it. Even better, the 
client should be able to say "only give me all that stuff if it has been 
modified since date XXXXXX" - to allow for caching. I don't know if ONE-2, 
mechanism does this, but it will obviously have to if it's ever going to 
become a viable solution.

The issue of dropping the current search/present based paradigm for Explain 
is the real heavyweight. It allows a lot of flexibility which will 
otherwise have to be built into the block that we load on top of our poor 
init exchange...  that's going to take some real heavy analysis - and I 
seriously hope the profile-meisters will allow time for that analysis (You 
Know Who You Are), before they run along with a solution which is 
superficially simpler but less thought-through.

--Sebastian
--
Sebastian Hammer        <quinn@indexdata.dk>            Index Data ApS
Ph.: +45 3341 0100    <http://www.indexdata.dk>    Fax: +45 3341 0101

Received on Thursday, 16 November 2000 16:15:48 UTC