Re: Retire has_multiple_baserels()
On Wed, Oct 11, 2023 at 1:13 AM Tom Lane wrote: > I thought this test wasn't too complete, because has_multiple_baserels > isn't reached at all in many cases thanks to the way the calling if() > is coded. I tried testing like this instead: > > diff --git a/src/backend/optimizer/path/allpaths.c > b/src/backend/optimizer/path/allpaths.c > index eea49cca7b..3f6fc51fb4 100644 > --- a/src/backend/optimizer/path/allpaths.c > +++ b/src/backend/optimizer/path/allpaths.c > @@ -2649,6 +2649,8 @@ set_subquery_pathlist(PlannerInfo *root, RelOptInfo > *rel, > */ > remove_unused_subquery_outputs(subquery, rel, run_cond_attrs); > > +Assert(has_multiple_baserels(root) == > (bms_membership(root->all_baserels) == BMS_MULTIPLE)); > + > /* > * We can safely pass the outer tuple_fraction down to the subquery > if the > * outer level has no joining, aggregation, or sorting to do. > Otherwise > > and came to the same conclusion: check-world finds no cases where > the assertion fails. So it LGTM too. Pushed. Thanks for pushing! Thanks Richard
Re: Retire has_multiple_baserels()
Aleksander Alekseev writes: >> The function has_multiple_baserels() is used in set_subquery_pathlist() >> to check and see if there are more than 1 base rel, by looping through >> simple_rel_array[]. I think one simpler way to do that is to leverage >> root->all_baserels by >> bms_membership(root->all_baserels) == BMS_MULTIPLE > I used the following patch to double check that nothing was missed: > ... > It wasn't. The patch LGTM. I thought this test wasn't too complete, because has_multiple_baserels isn't reached at all in many cases thanks to the way the calling if() is coded. I tried testing like this instead: diff --git a/src/backend/optimizer/path/allpaths.c b/src/backend/optimizer/path/allpaths.c index eea49cca7b..3f6fc51fb4 100644 --- a/src/backend/optimizer/path/allpaths.c +++ b/src/backend/optimizer/path/allpaths.c @@ -2649,6 +2649,8 @@ set_subquery_pathlist(PlannerInfo *root, RelOptInfo *rel, */ remove_unused_subquery_outputs(subquery, rel, run_cond_attrs); +Assert(has_multiple_baserels(root) == (bms_membership(root->all_baserels) == BMS_MULTIPLE)); + /* * We can safely pass the outer tuple_fraction down to the subquery if the * outer level has no joining, aggregation, or sorting to do. Otherwise and came to the same conclusion: check-world finds no cases where the assertion fails. So it LGTM too. Pushed. regards, tom lane
Re: Retire has_multiple_baserels()
On Tue, Oct 10, 2023 at 5:43 PM Aleksander Alekseev < aleksan...@timescale.com> wrote: > I used the following patch to double check that nothing was missed: > > ``` > --- a/src/backend/optimizer/path/allpaths.c > +++ b/src/backend/optimizer/path/allpaths.c > @@ -2207,8 +2207,13 @@ has_multiple_baserels(PlannerInfo *root) > /* ignore RTEs that are "other rels" */ > if (brel->reloptkind == RELOPT_BASEREL) > if (++num_base_rels > 1) > + { > + > Assert(bms_membership(root->all_baserels) == BMS_MULTIPLE); > return true; > + } > } > + > + Assert(bms_membership(root->all_baserels) != BMS_MULTIPLE); > return false; > } > ``` > > It wasn't. The patch LGTM. Thanks for the verification. Thanks Richard
Re: Retire has_multiple_baserels()
Hi, > The function has_multiple_baserels() is used in set_subquery_pathlist() > to check and see if there are more than 1 base rel, by looping through > simple_rel_array[]. I think one simpler way to do that is to leverage > root->all_baserels by > > bms_membership(root->all_baserels) == BMS_MULTIPLE > > all_baserels is computed in deconstruct_jointree (v16) or in > make_one_rel (v15 and earlier), both are before we generate access paths > for subquery RTEs, and it contains all base rels (but not "other" rels). > So it should be a suitable replacement. I doubt that there would be any > measurable performance gains. So please consider it cosmetic. > > I've attached a patch to do that. Any thoughts? I used the following patch to double check that nothing was missed: ``` --- a/src/backend/optimizer/path/allpaths.c +++ b/src/backend/optimizer/path/allpaths.c @@ -2207,8 +2207,13 @@ has_multiple_baserels(PlannerInfo *root) /* ignore RTEs that are "other rels" */ if (brel->reloptkind == RELOPT_BASEREL) if (++num_base_rels > 1) + { + Assert(bms_membership(root->all_baserels) == BMS_MULTIPLE); return true; + } } + + Assert(bms_membership(root->all_baserels) != BMS_MULTIPLE); return false; } ``` It wasn't. The patch LGTM. -- Best regards, Aleksander Alekseev
Retire has_multiple_baserels()
The function has_multiple_baserels() is used in set_subquery_pathlist() to check and see if there are more than 1 base rel, by looping through simple_rel_array[]. I think one simpler way to do that is to leverage root->all_baserels by bms_membership(root->all_baserels) == BMS_MULTIPLE all_baserels is computed in deconstruct_jointree (v16) or in make_one_rel (v15 and earlier), both are before we generate access paths for subquery RTEs, and it contains all base rels (but not "other" rels). So it should be a suitable replacement. I doubt that there would be any measurable performance gains. So please consider it cosmetic. I've attached a patch to do that. Any thoughts? Thanks Richard v1-0001-Retire-has_multiple_baserels.patch Description: Binary data