[CALCITE-3214] Add UnionToUnionRule for materialization matching.

[jinxing64]

Below materialization matching fails now

  @Test public void testDEV() {
    String sql0 = "select * from \"emps\" where \"empid\" < 300";
    String sql1 = "select * from \"emps\" where \"empid\" > 200";
    checkMaterialize(sql0 + " union all " + sql1, sql1 + " union all " + sql0);
  }

This PR proposes to add a rule for Union matching

[hsyuan]

What if the inputs of union have different column aliases? The union's output column names are expected to be the same with its first child.

[jinxing64]

@hsyuan
Thanks for looking into this ~
After this change -- #1310 , there's no strict column name checking but only column count and column type checking (SubstitutionVisitor::rowTypesAreEquivalent).
Do you think there's strong need to keep the column name same during materialization matching?

[hsyuan]

How about using for loop (i .. n) here? I don't think removing elements in query inputs is necessary. Other than that, looks good to me.

[jinxing64]

hmm~ let me think. The process of removing elements might be useful/necessary.
Let's make an example. Say Query and MV are like below:

Query:
Union
+--ScanA
+--ScanA
+--ScanB

MV:
Union
+--ScanA
+--ScanB
+--ScanB

Obviously the matching should be failed, but if I implement like below:

if (queryInputs.size() == targetInputs.size()) {
  for (MutableRel rel： queryInputs) {
    if (targetInputs.indexOf(queryInputs.get(0)) == -1) {
      return null;
    }
  }
}
return true;

The matching will succeeded, which is not correct. Though all input scans of query exist in MV, the counts are different.

THX again for review this :)

[hsyuan]

No, this is what I mean:

if (queryInputs.size() == targetInputs.size()) {
  for (MutableRel rel： queryInputs) {
   int index = targetInputs.indexOf(rel);
    if (index == -1) {
      return null;
    } else {
      targetInputs.remove(index);
    }
  }
}

You don't need to remove twice. Does this make sense?

[jinxing64]

Ah~ I got it~ Thanks ~ refine soon.

[jinxing64]

Thanks @hsyuan for review this ~
I refined by your comments and squashed the commits :)

[hsyuan]

+1

[jinxing64]

Thanks for approve ~