ADQL simplified not - Column typing
mchill at dial.pipex.com
Wed Jan 28 10:35:05 PST 2004
I'm confused... easily done...
Certainly, if we're doing the type checking at the datacenter backend, where the
ADQL is being parsed for meaning (whether to convert to SQL to submit to an
RDBMS, or for other searches on other data types) then it's 'too late' to worry
about typing, as that will occur naturally at that stage.
I was thinking there might be a case (a long time down the line so this is a bit
pie in the sky just now) where type information on columns could be made
available to the ADQL validator. This would mean (presumably) some kind of
restricted ADQL schema based on a combination of the normal ADQL schema and the
registry information _on the particular datacenter the ADQL will be applied to_
. You could restrict not just types but also names of columns, tables, etc....
Hmmmm... So any such specific ADQL query would need to have a reference to the
datacenter it is being applied to - or rather, the restricted schema for that
datacenter. If there are mirrors or similar datacenters, I suppose one ought to
be able to lookup all datacenters that will take that schema....
As I say, this may be some time off! But I'm wondering now whether being able
to generate such a restricted, data-center specific ADQL schema for a datacenter
would help us build fancy GUI pick-from-list ADQL query-generating tools...
Wil O'Mullane wrote:
>>When you mention the query will " ... include a reference
>>to the datacenter/registry entry it is expected to run on ",
>>I'm assuming that since the ADQL will reference a particular
>>table (VOResource) then its description should be same regardless
>>at which data centre it as at (the same table might be
>>replicated at various data centres).
> To follow your SQL analogy you would assume that a table named
> X was the same in any SQL database.
> This is probably not a particularly fair assumption.
> ADQL allows for an alias representing the archive to appear in
> front of the table name.
> There is noting in the ADQL spec as of now to force to destination of the
> query to be in the query. I do not think that is a necessary or
> good idea. It seems however some people feel ADQL will linger and
> they will be unable to know what it was meant to do and what the
> types of the columns were without know where it was intended to go to.
>>So the column based type information need not be included
>>in the ADQL itself, but it can be determined during query
> Correct - each NODE publishes its type info and the assumption is
> the ADQL coming to it is in the correct types and UNITS!
> But since SQL is as typeless as ADQL I fail to see what extra processing you have to do here.
AstroGrid @ ROE
Tel: +44 7901 55 24 66
More information about the voql