On 14/06/21 11:49 am, Andres Freund wrote:
Hi,
On 2021-06-13 15:22:12 -0700, Noah Misch wrote:
On Sun, Jun 13, 2021 at 06:09:20PM -0400, Tom Lane wrote:
We might be able to get rid of the stuff about concurrent step
completion in isolationtester.c if we required the spec files
to use annotations to force a deterministic step completion
order in all such cases.
Yeah. If we're willing to task spec authors with that, the test program can't
then guess wrong under unusual timing.
I think it'd make it *easier* for spec authors. Right now one needs to
find some way to get a consistent ordering, which is often hard and
complicates tests way more than specifying an explicit ordering
would. And it's often unreliable, as evidenced here and in plenty other
tests.
Greetings,
Andres Freund
How about adding a keyword like 'DETERMINISTIC' to the top level SELECT,
the idea being the output would be deterministic (given the same table
values after filtering etc), but not necessarily in any particular
order? So pg could decide the optimum way to achieve that which may not
necessarily need a sort.
Cheers,
Gavin
