Re: Declarative partitioning - another take

Sorry about the delay in replying.

On 2016/09/15 21:58, Ashutosh Bapat wrote:
> Hi Amit,
>
> It looks like there is some problem while creating paramterized paths
> for multi-level partitioned tables. Here's a longish testcase
>
> [ ... ]
>
> Please check if you are able to reproduce these errors in your
> repository. I made sure that I cleaned up all partition-wise join code
> before testing this, but ... .

Thanks for the test case. I can reproduce the same.

> I tried to debug the problem somewhat. In set_append_rel_pathlist(),
> it finds that at least one child has a parameterized path as the
> cheapest path, so it doesn't create an unparameterized path for append
> rel. At the same time there is a parameterization common to all the
> children, so it doesn't create any path. There seem to be two problems
> here
> 1. The children from second level onwards may not be getting
> parameterized for lateral references. That seems unlikely but
> possible.
> 2. Reparameterization should have corrected this, but
> reparameterize_path() does not support AppendPaths.

Hmm, 0005-Refactor-optimizer-s-inheritance-set-expansion-code-5.patch is
certainly to be blamed here; if I revert the patch, the problem goes away.

Based on 2 above, I attempted to add logic for AppendPath in
reparameterize_path() as in the attached. It fixes the reported problem
and does not break any regression tests. If it's sane to do things this
way, I will incorporate the attached into patch 0005 mentioned above.
Thoughts?

Thanks,
Amit