On Fri, 4 May 2007, Francois Ochsenbein wrote:
> I could try to talk about the metadata requirements --
> but the information_schema being something existing almost only in
> SQL server, I don't feel it would be wise to insist on this solution.
>
> Cheers, francois
Probably the implementation of information_schema is pretty mixed, although I got the impression that MySQL and PostgreSQL have a pretty good implementation as well, and possibly DB2 and INFORMIX (but not Oracle evidently). And of course SQL Server.
I suggest we consider the issue of an information_schema like approach, and the issue of whether to use the information_schema specification directly, as two separate issues. The key question is what metadata is required to compose non-trivial queries, as opposed to discovery/planning via the registry, where only very limited information is required, and whether these are the same or distinct. I just would hate to see TAP crippled by an approach which can't easily grow to support more sophisticated applications, and efficient execution of complex queries against large tables.