Re: PoC/WIP: Extended statistics on expressions

On 3/25/21 2:33 PM, Dean Rasheed wrote:
> On Thu, 25 Mar 2021 at 00:05, Tomas Vondra
> <tomas(dot)vondra(at)enterprisedb(dot)com> wrote:
>>
>> here's an updated patch. 0001
>
> The change to the way that CreateStatistics() records dependencies
> isn't quite right -- recordDependencyOnSingleRelExpr() will not create
> any dependencies if the expression uses only a whole-row Var. However,
> pull_varattnos() will include whole-row Vars, and so nattnums_exprs
> will be non-zero, and CreateStatistics() will not create a whole-table
> dependency when it should.
>
> I suppose that could be fixed up by inspecting the bitmapset returned
> by pull_varattnos() in more detail, but I think it's probably safer to
> revert to the previous code, which matched what index_create() did.
>

Ah, good catch. I haven't realized recordDependencyOnSingleRelExpr works
like that, so I've moved it after the whole-table dependency.

Attached is an updated patch series, with all the changes discussed
here. I've cleaned up the ndistinct stuff a bit more (essentially
reverting back from GroupExprInfo to GroupVarInfo name), and got rid of
the UpdateStatisticsForTypeChange.

I've also looked at speeding up the stats_ext regression tests. The 0002
patch reduces the size of a couple of test tables, and removes a bunch
of queries. I've initially mostly just copied the original tests, but we
don't really need that many queries I think. This cuts the runtime about
in half, so it's mostly in line with other tests. Some of these changes
are in existing tests, I'll consider moving that into a separate patch
applied before the main one.

regards

--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company