Add support to identify different set operators & allow chain of multiple set operators

[Dencrash]

No description provided.

[Dencrash]

Thanks for the comments and spotting some of the oversights!

[mrks]

What's the status on this? Once the current master is merged, is this ready for a final review?

[Dencrash]

@mrks that depends. There is one bug still in there that is not yet closed. The bug is not something we added but already exists for the current system. The solution for that is relatively easy, but ugly as sin and we were stalling last week to see, if we can find a better solution.

We could do the following:
Let every query have a vector of set operation instead of only allowing one linked and nested set operation. This would mean that we don't overwrite old set operations anymore when specific parenthesis are used, but it means that we mix a vector with linked lists.

This is the only thing that is currently not done.

[mrks]

Can't you iterate over the chained list of select statements and add the set operation to the end? Something like

$$ = $1;
while ($$->setOperator) $$ = $$->setOperator;
$$->setOperator = $2;

No idea if bison likes that though.

Also, could you add some tests with similar (multiple, different, nested, ...) set operations in the WITH part of the query?

[Dencrash]

Hanging onto the event would mean that we have the wrong order of operation. We also cannot use the other Select statement because the set operation there could also be overwritten.

We can add test with WITH Statements and more nested queries

[mrks]

Let's discuss this on Wednesday.

[Dencrash]

@mrks I have added an explanation to the evaluation order of SetOperators as a comment. From our side this should be ready for your final review

[mrks]

Nothing fundamental here, just trying to gain favors with posterity.

[Dencrash]

@mrks I think that I added all the requested changes. Let me know, if you think this is ready to merge

[Daniel-Xu]

Hi, @Dencrash. From the test case, I don't see that the setOperations has multiple children. Is there any case that requires the setOperations to be a vector instead of a selectOption pointer?

[Daniel-Xu]

From the comment in the source code.
I believe (SELECT * FROM students INTERSECT SELECT * FROM students_2) UNION SELECT * FROM students_3 ORDER BY grade ASC; this is the case I was talking about ↑

[mweisgut]

Hi, @Dencrash. From the test case, I don't see that the setOperations has multiple children. Is there any case that requires the setOperations to be a vector instead of a selectOption pointer?

Do you mean a setOperation pointer rather than a selectOption pointer?

I think you're right. Thank's for pointing to that issue. #199 contains a patch addressing it.

[Daniel-Xu]

Hi, @Dencrash. From the test case, I don't see that the setOperations has multiple children. Is there any case that requires the setOperations to be a vector instead of a selectOption pointer?

Do you mean a setOperation pointer rather than a selectOption pointer?

I think you're right. Thank's for pointing to that issue. #199 contains a patch addressing it.

yea, selectOption was a typo

[klauck]

Hi @Daniel-Xu , hi @mweisgut,
if setOperations is only a pointer, we cannot express the order of set operations anymore,
e.g., there is no way to express the difference between (Q1 EXCEPT Q2) EXCEPT Q3 and Q1 EXCEPT (Q2 EXCEPT Q3).
Using this example: For (Q1 EXCEPT Q2) EXCEPT Q3, Q1 must have two children and setOperations must be a vector. For Q1 EXCEPT (Q2 EXCEPT Q3), Q1 and Q2 have one child each.

[klauck]

Nevertheless, in the current version Q1 EXCEPT Q2 EXCEPT Q3 == Q1 EXCEPT (Q2 EXCEPT Q3), which I think is a bug.
It should be Q1 EXCEPT Q2 EXCEPT Q3 == (Q1 EXCEPT Q2) EXCEPT Q3.

[klauck]

@mweisgut pointed out that INTERSECT binds more tightly than UNION and EXCEPT. [https://www.postgresql.org/docs/14/queries-union.html]
This is not yet implemented in the parser.

[Daniel-Xu]

Hi @Daniel-Xu , hi @mweisgut, if setOperations is only a pointer, we cannot express the order of set operations anymore, e.g., there is no way to express the difference between (Q1 EXCEPT Q2) EXCEPT Q3 and Q1 EXCEPT (Q2 EXCEPT Q3). Using this example: For (Q1 EXCEPT Q2) EXCEPT Q3, Q1 must have two children and setOperations must be a vector. For Q1 EXCEPT (Q2 EXCEPT Q3), Q1 and Q2 have one child each.

Thanks for the response, this makes sense now.