On 8/4/10 2:39 PM +0300, Dean Rasheed wrote:
> Does this sound like a useful feature? Is this a sane approach to
> implementing it? If not, has anyone else given any thought as to how
> it might be implemented?
I didn't look at the patch, but so far, I've identified three problems
with the existing view system:
1) You can't re-evaluate the UPDATE expression like an UPDATE on a table does. Consider for example UPDATE
fooSET a=a+1; If the tuples change before we get to them, we lose data because we simply can't re-evaluate
"a+1"in the trigger.
2) You can't set the number of affected rows.
3) You can't set the RETURNING results. You suggested that RETURNING for DELETE would return the OLD value,
butthat seems broken because that's not necessarily what was deleted. I didn't understand what you
suggestionfor UPDATE was; how does PG know that if the view doesn't have a primary key?
I think these are the main three problems that prevent people from
actually using views, and I think these should be focused on when adding
triggers on VIEWS. I would love to see the feature though.
Any thoughts?
Regards,
Marko Tiikkaja