[
https://issues.apache.org/jira/browse/CALCITE-4331?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vladimir Sitnikov updated CALCITE-4331:
---------------------------------------
Description:
Calcite ignores data types completely if one of them is {{SqlTypeName.ANY}}.
That enables to transform rowtype from {{name ANY NON NULL}} to {{name ANY
NULLABLE}} in a planner rule, even though, the planner does not allow to loosen
field nullability.
What I got was silent wrong results {{count(name)}} was transformed to
{{count(\*)}} since Calcite believed the field was non-nullable, and it did not
care that I provided nullable rex for the implementation. Frankly speaking, I
assumed "no asserts => the rowtypes in call.transformTo(..) are ok"
Current code:
{code:java}
for (Pair<RelDataTypeField, RelDataTypeField> pair : Pair.zip(f1, f2)) {
final RelDataType type1 = pair.left.getType();
final RelDataType type2 = pair.right.getType();
// If one of the types is ANY comparison should succeed
if (type1.getSqlTypeName() == SqlTypeName.ANY
|| type2.getSqlTypeName() == SqlTypeName.ANY) {
continue;
}
if (!type1.equals(type2)) {
return false;
}
{code}
I'm inclined add a check that both types have the same {{isNullable()}} value.
ANY checks were added in
https://github.com/apache/calcite/commit/5562fc3952f61d943f109796c3d62b7390a8ac48
[~jacques], [~julianhyde], do you know if it is OK to add the following?
{code:java}
if (type1.isNullable() != type2.isNullable()) {
return false;
}
{code}
was:
Calcite ignores data types completely if one of them is {{SqlTypeName.ANY}}.
That enables to transform rowtype from {{name ANY NON NULL}} to {{name ANY
NULLABLE}} in a planner rule, even though, the planner does not allow to loosen
field nullability.
What I got was silent wrong results {{count(name)}} was transformed to
{{count(*)}} since Calcite believed the field was non-nullable, and it did not
care that I provided nullable rex for the implementation. Frankly speaking, I
assumed "no asserts => the rowtypes in call.transformTo(..) are ok"
Current code:
{code:java}
for (Pair<RelDataTypeField, RelDataTypeField> pair : Pair.zip(f1, f2)) {
final RelDataType type1 = pair.left.getType();
final RelDataType type2 = pair.right.getType();
// If one of the types is ANY comparison should succeed
if (type1.getSqlTypeName() == SqlTypeName.ANY
|| type2.getSqlTypeName() == SqlTypeName.ANY) {
continue;
}
if (!type1.equals(type2)) {
return false;
}
{code}
I'm inclined add a check that both types have the same {{isNullable()}} value.
ANY checks were added in
https://github.com/apache/calcite/commit/5562fc3952f61d943f109796c3d62b7390a8ac48
[~jacques], [~julianhyde], do you know if it is OK to add {{if
(type1.isNullable() != type2.isNullable()) { return false; } }} ?
> RelOptUtil#areRowTypesEqual should compare type nullability even for ANY types
> ------------------------------------------------------------------------------
>
> Key: CALCITE-4331
> URL: https://issues.apache.org/jira/browse/CALCITE-4331
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.26.0
> Reporter: Vladimir Sitnikov
> Priority: Major
>
> Calcite ignores data types completely if one of them is {{SqlTypeName.ANY}}.
> That enables to transform rowtype from {{name ANY NON NULL}} to {{name ANY
> NULLABLE}} in a planner rule, even though, the planner does not allow to
> loosen field nullability.
> What I got was silent wrong results {{count(name)}} was transformed to
> {{count(\*)}} since Calcite believed the field was non-nullable, and it did
> not care that I provided nullable rex for the implementation. Frankly
> speaking, I assumed "no asserts => the rowtypes in call.transformTo(..) are
> ok"
> Current code:
> {code:java}
> for (Pair<RelDataTypeField, RelDataTypeField> pair : Pair.zip(f1, f2)) {
> final RelDataType type1 = pair.left.getType();
> final RelDataType type2 = pair.right.getType();
> // If one of the types is ANY comparison should succeed
> if (type1.getSqlTypeName() == SqlTypeName.ANY
> || type2.getSqlTypeName() == SqlTypeName.ANY) {
> continue;
> }
> if (!type1.equals(type2)) {
> return false;
> }
> {code}
> I'm inclined add a check that both types have the same {{isNullable()}} value.
> ANY checks were added in
> https://github.com/apache/calcite/commit/5562fc3952f61d943f109796c3d62b7390a8ac48
> [~jacques], [~julianhyde], do you know if it is OK to add the following?
> {code:java}
> if (type1.isNullable() != type2.isNullable()) {
> return false;
> }
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
