Improve type inference

[beikov]

Right now, we always try to determine the MappingModelExpressible in BaseSqmToSqlAstConverter by looking at an SQM node, but for certain expressions/predicates e.g. binary arithmetic or comparison predicate, we could instead make use of Expression#getExpressionType of the "other side". For example, if we first translate the LHS and then the RHS of a comparison predicate like path = :param, it would be much better to use the expression type of the LHS to infer the type of the RHS.
A possible downside of this would be, that we probably wouldn't figure out issues in the determineValueMapping function so fast because usually the LHS in expressions/predicates is a typed node.

Any reason for not applying this improvement?

We currently do something like this:

inferrableTypeAccessStack.push( () -> determineValueMapping( rightOperand ) );
final Expression lhs = toSqlExpression( leftOperand.accept( this ) );
inferrableTypeAccessStack.pop();

inferrableTypeAccessStack.push( () -> determineValueMapping( leftOperand ) );
final Expression rhs = toSqlExpression( rightOperand.accept( this ) );
inferrableTypeAccessStack.pop();

whereas we could do this:

inferrableTypeAccessStack.push( () -> determineValueMapping( rightOperand ) );
final Expression lhs = toSqlExpression( leftOperand.accept( this ) );
inferrableTypeAccessStack.pop();

inferrableTypeAccessStack.push( () -> lhs.getExpressionType() );
final Expression rhs = toSqlExpression( rightOperand.accept( this ) );
inferrableTypeAccessStack.pop();

[sebersole]

Really not sure what you are proposing here exactly, so hard to say.

Should we "improve" things where there is an opportunity? Yes, of course. Kind of depends on details.

[beikov]

I added an example to the description above now.

[beikov]

It doesn't seem worth it. I'll not do that for now.