Update TSQL grammar so that INTERSECT precedence is handled there

[asnare]

This PR updates the TSQL grammar and accompanying IR processing so that the precedence of INTERSECT is handled by the grammar instead of during transportation to our IR.

Resolves #1225.

[github-actions]

Coverage tests results

0 tests   - 464   0 ✅  - 427   0s ⏱️ -4s
0 suites  -   6   0 💤  -  37 
0 files    -   6   0 ❌ ±  0 

Results for commit 71e160c. ± Comparison against base commit 220b303.

[ericvergnaud]

I suspect this and the below can be greatly simplified using something such as

queryExpression
 : querySpecification
 | queryExpression UNION ALL queryExpression
 | queryExpression INTERSECT queryExpression
etc...

[asnare]

@ericvergnaud: This is interesting. I don't fully understand why the grammar is as it is. I already came up with this which also seems to be correct and is closer to what I would think of as the 'classical' way to express these rules:

queryExpression
    : LPAREN queryExpression RPAREN
    | queryExpression (UNION ALL? | EXCEPT) queryExpression
    | queryExpression INTERSECT queryExpression
    | querySpecification
    ;

Are there performance/implementation reasons for preferring the existing folding (??) style over this?

[jimidle]

This is probably correct, precedence is generally best done as one rule with precedence then being explicit, then label the alts and the visitor is more simple as well.

Let me look at the existing grammar

[asnare]

Thanks @jimidle, @vil1 indeed suggested:

queryExpression
    : LPAREN queryExpression RPAREN                                         #inParen
    | left = queryExpression (UNION ALL? | EXCEPT) right = queryExpression  #union
    | left = queryExpression INTERSECT right = queryExpression              #intersect
    | querySpecification                                                    #simple
    ;

[jimidle]

Generally, don' tuse the labels left/right as they generate more code. ctx.queryExpression(0), ctx.queryExpression(1) is more efficient. See the read me about working with antlr grammar, labels are useful for disambiguation, but even then it might mean splitting rules or labeling the alts is better.

[jimidle]

@ericvergnaud: This is interesting. I don't fully understand why the grammar is as it is. I already came up with this which also seems to be correct and is closer to what I would think of as the 'classical' way to express these rules:

queryExpression
    : LPAREN queryExpression RPAREN
    | queryExpression (UNION ALL? | EXCEPT) queryExpression
    | queryExpression INTERSECT queryExpression
    | querySpecification
    ;

Are there performance/implementation reasons for preferring the existing folding (??) style over this?

BTW the grammar is like this because it was just typed in by reading the MS docs, which don't encapsulate such things. You should have seen it 6 months ago ;). These are the occasions where we get to fix it up.