sql: support array-flatten subqueries within UDFs #98879
Conversation
|
This change is |
There was a problem hiding this comment.
Reviewable status:
complete! 1 of 0 LGTMs obtained (waiting on @DrewKimball)
There was a problem hiding this comment.
Reviewed 9 of 9 files at r1, all commit messages.
Reviewable status:
pkg/sql/opt/norm/rules/scalar.opt line 389 at r1 (raw file):
# and correlated ones only over the types that array_agg supports. # # Note that optbuilder should have already verified that if the input is
Since you're at it, I think we can remove this bit since #35710 is fixed. Same for the check in optbuilder/scalar.go line 156.
pkg/sql/opt/exec/execbuilder/testdata/subquery line 566 at r1 (raw file):
spans: FULL SCAN # Case where the a correlated subquery contains an uncorrelated array-flatten
[nit] the a -> a
mgartner
force-pushed
the
98738-array-flatten-udfs
branch
from
a6d69ad to
a8a3b75
Compare
There was a problem hiding this comment.
Reviewable status:
pkg/sql/opt/norm/rules/scalar.opt line 389 at r1 (raw file):
Previously, DrewKimball (Drew Kimball) wrote…
Since you're at it, I think we can remove this bit since #35710 is fixed. Same for the check in
optbuilder/scalar.goline 156.
Done.
pkg/sql/opt/exec/execbuilder/testdata/subquery line 566 at r1 (raw file):
Previously, DrewKimball (Drew Kimball) wrote…
[nit]
the a->a
Done.
|
bors r+ |
|
Build failed (retrying...): |
|
Build failed (retrying...): |
|
Build failed: |
Array-flatten subqueries (e.g., `ARRAY(SELECT a FROM t)`) are now supported within UDFs. They are now converted to a normal subquery with a ScalarGroupBy if they exist within a UDF, even if they are uncorrelated. This allows them to be executed without any changes to the execbuilder or the evaluation logic of `tree.Routine`. Fixes cockroachdb#98738 Release note: None
mgartner
force-pushed
the
98738-array-flatten-udfs
branch
from
a8a3b75 to
99169e8
Compare
There was a problem hiding this comment.
Reviewable status:
pkg/sql/opt/norm/rules/scalar.opt line 389 at r1 (raw file):
Previously, mgartner (Marcus Gartner) wrote…
Done.
This caused some test failures so I've reverted. I believe the check is still necessary.
Code quote:
# Note that optbuilder should have already verified that if the input is
# correlated, then we can array_agg over the input type. Also note that the
# Max1Row operator we introduce is guaranteed to be eliminated as
# MakeArrayAggForFlatten will return a ScalarGroupBy.
|
bors r+ |
|
Build failed (retrying...): |
|
Build failed (retrying...): |
|
Build succeeded: |
Array-flatten subqueries (e.g.,
ARRAY(SELECT a FROM t)) are nowsupported within UDFs. They are now converted to a normal subquery with
a ScalarGroupBy if they exist within a UDF, even if they are
uncorrelated. This allows them to be executed without any changes to the
execbuilder or the evaluation logic of
tree.Routine.Fixes #98738
Release note: None