On 04/06/10 22:33, Tom Lane wrote:
> Heikki Linnakangas<heikki.linnakangas@enterprisedb.com> writes:
>> On 04/06/10 17:33, Tom Lane wrote:
>>> Maybe the entire idea is unworkable. I certainly don't find any comfort
>>> in your proposal in the above-referenced message to trust index
>>> operators; where is it written that those don't throw errors?
>
>> Let's consider b-tree operators for an index on the secure table, for
>> starters. Surely a b-tree index comparison operator can't throw an error
>> on any value that's in the table already, you would've gotten an error
>> trying to insert that.
>
> Man, are *you* trusting.
>
> A counterexample: suppose we had a form of type "text" that carried a
> collation specifier internally, and the comparison routine threw an
> error if asked to compare values with incompatible specifiers. An index
> built on a column of all the same collation would work fine. A query
> that tried to compare against a constant of a different collation would
> throw an error.
I can't take that example seriously. First of all, tacking a collation
specifier to text values would be an awful hack. Secondly, it would be a
bad idea to define the b-tree comparison operators to throw an error; it
would be a lot more useful to impose an arbitrary order on the
collations, so that all values with collation A are considered smaller
than values with collation B. We do that for types like box; smaller or
greater than don't make much sense for boxes, but we implement them in a
pretty arbitrary way anyway to make it possible to build a b-tree index
on them, and for the planner to use merge joins on them, and implement
DISTINCT using sort etc.
>> I'm not sure. But indexable
>> operations are what we care about the most; the order of executing those
>> determines if you can use an index scan or not.
>
> Personally, I care just as much about hash and merge join operators...
Hash seems safe too. Don't merge joins just use the default b-tree operator?
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com
