← Back to team overview

launchpad-dev team mailing list archive

Re: Stop the line: Test failures in *stable*

 

On Fri, Oct 8, 2010 at 7:46 PM, Martin Pool <mbp@xxxxxxxxxxxxx> wrote:
> I wonder if it'd be possible to make Storm send up a different type of
> resultset for an unordered query, that would trap if you tried to do
> something on it that seemed to indicate you needed something ordered?
> For instance perhaps it would be reasonable to say that collections
> iterated by TAL should always be ordered?  If uses of unordered sets
> are really rare, perhaps they could stick out in code?

The problem here is that 'set(resultset)' and 'list(resultset)' aren't
distinguisable by the resultset (without -very- ugly code); slices /
listification is where we would care about 'must have an order'; but
even if that issue didn't exist, some order_by clauses are not
sufficient, its not just *having one* that matters, its that it is
sufficient to guarantee stability (where we care about that).

The previous approach - add a random order-by to all queries which
support order_by at all, worked well; I'm in favour of just bringing
it back.

-Rob



Follow ups

References