Consider adding a ref LocalState state parameter. #Closed

var rightState = this.State.Clone(); [](start = 12, length = 36)

Why do we need this clone? Is LearnFromNonNullTest changing the current state? Also, the rightState name is somewhat misleading, we haven't visited the right side yet, which is not in line with the naming pattern used in this component. #Closed

warning CS8602: Possible dereference of a null reference. [](start = 27, length = 57)

The warning feels misleading, e is not dereferenced here. #Closed

node.Kind == BoundKind.ThrowExpression || [](start = 25, length = 42)

Relaxing this assert doesn't feel like a right thing to do. Are we not setting _resultType properly for a throw expression? I think we should do that instead. #Closed

What do you think the _resultType should be?

Currently, we let _resultType be Exception (or whatever is thrown) from visiting the thrown value.
If we wanted to set it properly, it should probably be the converted type of the throw expression, which does not exist in the throw expression itself, only in the conversion that wraps the throw expression.

A possibility is to set the _resultType to null/default, since the BoundThrowExpression has a null Type. Then I would adjust this method to handle null/default _resultType, instead of special-casing throw expression.
Do you think that's better?

Do you think that's better?

In this visitor, after visiting each expression, we should set _resultType the result type should reflect the original type of the node plus added nullability information. It shouldn't matter if there is a conversion on top of the BoundThrowExpression node, it will be handled when its turn comes and the _resultType will be adjusted for it as appropriate. That said, there should be no special handling/treatment for BoundThrowExpression, but adjusting the assert adds it.

In reply to: 253510940 [](ancestors = 253510940)

Consider adding a test for string~. #Closed