[CALCITE-3203] When matching materializations, match Project with child of Aggregate

[jinxing64]

In current code, SubstitutionVisitor & MaterializedViewSubstitutionVisitor fail to support below matching:

   query:   Project(projects: [$0, *(2, $1)])
                  Aggregate(groupSet: {0}, groupSets: [{0}], calls: [SUM($1)])
                     Scan(table: [hr, emps])</li>
   target:  Project(projects: [$0, *(2, $1), *(2, $2)])
                  Aggregate(groupSet: {0}, groupSets: [{0}], calls: [SUM($1), COUNT()])
                     Scan(table: [hr, emps])</li>

And below test fails

// MaterializationTest.java
  @Test public void testAggregate() {
    checkMaterialize(
      "select \"deptno\", count(1), 2 * sum(\"empid\") from "
        + "(select * from \"emps\" union all select * from \"emps\")"
        + "group by \"deptno\"",
      "select \"deptno\", 2 * sum(\"empid\") from "
        + "(select * from \"emps\" union all select * from \"emps\")"
        + "group by \"deptno\"");
  }

The reason is that Project&Aggregate are not taken into consideration at the same time in current matching rules.
It might make sense to create a rule of ProjectOnAggregateToProjectOnAggregateUnifyRule to handle such case.

[hsyuan]

Maybe you can rename the rule name to ProjectAggregateToProjectAggregateUnifyRule

[jinxing64]

THX for look into this @hsyuan
I will update soon.

[hsyuan]

Can you rebase on master?

[jinxing64]

@hsyuan
Thanks for your kind reminder.

Actually I think this case can be covered by #1384
The matching pattern proposed in this PR is like below:

   query:   Project(projects: [$0, *(2, $1)])
                   Aggregate(groupSet: {0}, groupSets: [{0}], calls: [SUM($1)])
                     Scan(table: [hr, emps])
    target:  Project(projects: [$0, *(2, $1), *(2, $2)])
                   Aggregate(groupSet: {0}, groupSets: [{0}], calls: [SUM($1), COUNT()])
                     Scan(table: [hr, emps])

The matching pattern proposed in #1384 is like below:

 query:   Project(projects: [$0, +($1, 2)])
                 Project(projects: [$1, $3, $4])
                    Rel-A
 target:  Project(projects: [$1, +($3, 2)])
                 Rel-A

I think the second pattern is more general and I added the test case in this PR to #1384
How do you think ? Which one is preferred ?

If #1384 makes sense, it's great if you can shepherd and take a review when you have time. Then I will close this PR.

Thanks ~

Jin

[jinxing64]

This change can be covered by #1451
So I'm going to close it now.