[
https://issues.apache.org/jira/browse/CALCITE-4085?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17150701#comment-17150701
]
Danny Chen commented on CALCITE-4085:
-------------------------------------
I think we should answer the question for a record type:
Row(A int not null, B varchar(20))
What the "NOT NULL" constraint means for the field A, does it mean
- 1. it is not null within the row
- 2. or does it mean it is not null when we use it(access it through field
access, dot or item .etc).
It the answer is 2, current factory does the right thing, if it is 1, we should
try to preserve the nullability when constructing it.
Since the SQL standard does not allow nullability constraint, i tried OracleL
here:
{code:sql}
CREATE TYPE address_t AS OBJECT (
street VARCHAR2(30),
city VARCHAR2(20),
state CHAR(2),
zip CHAR(5) );
CREATE TABLE customers (
custid NUMBER,
address address_t);
INSERT INTO customers VALUES (1, address_t('101 First', 'Redwood Shores', 'CA',
'94065'));
INSERT INTO customers VALUES (2, null);
INSERT INTO customers VALUES (3, address_t(null, null, null, null));
{code}
the query:
{code:sql}
select * from customers where address is null;
{code}
outputs:
{code:sql}
-- custid address
-- 2 null
{code}
Even though all the attributes are null for address, the object itself is not
null.
> Improve nullability support for fields of structured type
> ---------------------------------------------------------
>
> Key: CALCITE-4085
> URL: https://issues.apache.org/jira/browse/CALCITE-4085
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Reporter: Dawid Wysakowicz
> Assignee: Dawid Wysakowicz
> Priority: Major
> Time Spent: 20m
> Remaining Estimate: 0h
>
> As discussed in
> https://lists.apache.org/thread.html/r602ac95fff23dd1ef974fb396df7be061ab861384ec42f5c57ce0bc2%40%3Cdev.calcite.apache.org%3E
> I would like to change the way a type of a field of a record is derived at
> couple of locations. This helps frameworks such as Apache Flink to build
> support for nullable records with not null fields.
> I suggest to change:
> * SqlDotOperator#deriveType
> * SqlItemOperator#inferReturnType
> * AliasNamespace#validateImpl
> * RexBuilder#makeFieldAccessInternal
> * SqlValidatorImpl.DeriveTypeVisitor#visit(SqlIdentifier)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
