Re: PATCH: index-only scans with partial indexes

Hello,

At Thu, 01 Oct 2015 01:36:51 +0200, Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> wrote in <560C7213(dot)3010203(at)2ndquadrant(dot)com>
> > Good point. I think we may simply point indexrinfos to the existing
> > list
> > of restrictions in that case - we don't need to copy it. So no
> > additional memory / CPU consumption, and it should make the code
> > working
> > with it a bit simpler.
>
> Attached is v5 of the patch improving the comments (rephrasing, moving
> before function etc.).

Looks good (for me).

> I also attempted to fix the issue with empty list for non-partial
> indexes by simply setting it to rel->baserestrictinfo in
> match_restriction_clauses_to_index (for non-partial indexes),

Although it is not the cause of these errors (or any errors on
regress), build_paths_for_OR (which doesn't seem to be called in
regression tests) doesn't set indexrinfos
properly. match_clauses_to_index can commonize(?) the process in
return for additional list copy and concat. The attached diff is
doing in the latter method. Avoiding the copies will make the
code a bit complicated.

But anyway regtests doesn't say whether it is sane or not:( (TODO)

> but that
> results in a bunch of
>
> ERROR: variable not found in subplan target list
>
> during "make installcheck". I can't quite wrap my head around why.

During considerartion on parameterized joins in
get_join_index_paths, caluseset fed to get_index_paths is
generated from given join (j|e)clausesets. So completing the
clauseset with necessary restrictinfo from rcaluseset fixes the
problem.

The attached patch applies on the latest v5 patch and will
address above issues. (And modifies expected files, which are the
manifestation of this improovement).

regards,

--
Kyotaro Horiguchi
NTT Open Source Software Center