SqlServer Equal/Not Equal/In predicate support for case sensitive varchar columns

[vlad-lyutenko]

Description

Enable Equal/Not Equal and IN predicate pushdown for varchar, nvarchar columns.
It requires, that column should have case sensitive collation.

Additional context and related issues

Release notes

( ) This is not user-visible or docs only and no release notes are required.
( ) Release notes are required, please propose a release note for me.
(x) Release notes are required, with the following suggested text:

Trino is able to perform predicate pushdown only for case sensitive varchar/nvarchar columns.
Supported operators:

• equal
• not equal
• IN
• NOT IN
  User can manually change collation for specific column in SqlServer to enable such pushdowns.
  Collation should be case sensitive (like _CS or _BIN) .

# Section
* Fix some things. ({issue}`issuenumber`)

[ssheikin]

nice! lgtm

[kokosing]

Looks nice

[kokosing]

why this one is disabled?

[vlad-lyutenko]

SUPPORTS_JOIN_PUSHDOWN_WITH_VARCHAR_EQUALITY(fallback -> fallback.test(SUPPORTS_JOIN_PUSHDOWN) && fallback.test(SUPPORTS_PREDICATE_PUSHDOWN_WITH_VARCHAR_EQUALITY)),

It was disabled before, because SUPPORTS_PREDICATE_PUSHDOWN_WITH_VARCHAR_EQUALITY was disabled.
Now we enabled it, but SUPPORTS_JOIN_PUSHDOWN_WITH_VARCHAR_EQUALITY should be disabled, because we don't support join on varchar in SQLServer:

    @Override
    protected boolean isSupportedJoinCondition(ConnectorSession session, JdbcJoinCondition joinCondition)
    {
        if (joinCondition.getOperator() == JoinCondition.Operator.IS_DISTINCT_FROM) {
            // Not supported in SQL Server
            return false;
        }

        // Remote database can be case insensitive.
        return Stream.of(joinCondition.getLeftColumn(), joinCondition.getRightColumn())
                .map(JdbcColumnHandle::getColumnType)
                .noneMatch(type -> type instanceof CharType || type instanceof VarcharType);
    }

[vlad-lyutenko]

But actually maybe now we can enable it on case sensitive columns?

[hashhar]

LGTM % comments.

[hashhar]

@Praveen2112 Can you please take a look if the Domain handling appears correct?

[Praveen2112]

Shouldn't we run getDomainCompactionThreshold check first and then check if the complement of given ValueSet is discrete ? What if we complement of a ValueSet is discrete but it has too many values so - we might compact it and then domain.getValues().complement().isDiscreteSet might not be valid.

It would be nice if we have a test coverage for this edge case also.

[vlad-lyutenko]

Thx, looks like I disable compaction threshold functionality by performing this check.
Refactored and test case added.

[Praveen2112]

Can we have additional test with NOT IN

[vlad-lyutenko]

Test case added

[hashhar]

@vlad-lyutenko reminder to update commit message with something like below, right now it just talks about equality.

Pushdown equality/inequality predicates on text columns in SQL Server

When the column uses a case-sensitive collation in SQL Server we can
pushdown the equality/inequality predicates on text columns without
affecting correctness.

Range predicates on text columns are never pushed down and
equality/inequality predicates on text columns are not pushed down if
the column uses a case-insensitive collation.

LGTM otherwise.