Paul Boddie wrote: > Well, if the client is free not to bother signalling anything about > erroneous value types, one has to wonder why there's so much of a > specification.
If you read it, I think you'll notice that the committee has managed to produce a lot of text without spending too much ink on error handling. I'm not completely up to date with the latest standards, but last time I looked, a lot of things that are essential parts of all serious implementations were missing from the standards. I'm pretty certain it would overwhelm the SQL standards committee if they had to specify how error conditions are handled. It would certainly be useful though. E.g. is PostgreSQL SQL compliant in forcing a rollback as soon as an operation fails. E.g. in Oracle, you can have a transaction with a loop where you try to insert values, and fall back to updating instead if you get a duplicate key on insert error. In PostgreSQL you need to set a savepoint before the the insert and rollback to that. Does the standard say whether this is kosher? I think not. -- http://mail.python.org/mailman/listinfo/python-list
