ISwitchExpressionOperation.IsExhaustive API extension

[bernd5]

fixes #52577

[bernd5]

@333fred or @CyrusNajmabadi please review

[bernd5]

Something to add or change?

[333fred]

Done review pass (commit 10)

[AlekseyTs]

    private ISwitchExpressionOperation CreateBoundSwitchExpressionOperation(BoundSwitchExpression boundSwitchExpression)

It looks like we can safely change this type to BoundConvertedSwitchExpression.

---

In reply to: 827918882

---

In reply to: 827918882

---

Refers to: src/Compilers/CSharp/Portable/Operations/CSharpOperationFactory.cs:2153 in 955cd47. [](commit_id = 955cd47, deletion_comment = False)

[AlekseyTs]

Done with review pass (commit 11), test are not looked at

---

In reply to: 827922761

---

In reply to: 827922761

[AlekseyTs]

It looks like ControlFlowGraphBuilder always adds SwitchExpressionException branch for a switch expression. Should we adjust that to be conditional based on the new property?

---

In reply to: 827926846

[bernd5]

@AlekseyTs : I tried to follow your comment - but I'm not quite sure if it is correct

[bernd5]

@AlekseyTs : thanks for your commits.
Could you look at the failing tests?

[AlekseyTs]

Could you look at the failing tests?

The best way to deal with test that are failing due to baseline difference is to update the baselines. Then we can review the changes and confirm that the baseline changes are expected.
Also, it looks like there is a formatting error:

##[error]src\Compilers\Test\Core\Compilation\TestOperationVisitor.cs(1289,1): error IDE0055: (NETCORE_ENGINEERING_TELEMETRY=Build) Fix formatting

[333fred]

Alright, I've dug through the test failures here. The problem is that we are introducing a branch for the following (example) tree:

        Block[B4] - Block
            Predecessors: [B2]
            Statements (0)
            Jump if False (Regular) to Block[B6]
                IIsPatternOperation (OperationKind.IsPattern, Type: System.Boolean) (Syntax: '_ => true')
                  Value: 
                    IFlowCaptureReferenceOperation: 2 (OperationKind.FlowCaptureReference, Type: System.Int32, IsImplicit) (Syntax: 'input')
                  Pattern: 
                    IDiscardPatternOperation (OperationKind.DiscardPattern, Type: null) (Syntax: '_') (InputType: System.Int32, NarrowedType: System.Int32)
                Leaving: {R2}

            Next (Regular) Block[B5]
        Block[B5] - Block
            Predecessors: [B4]
            Statements (1)
                IFlowCaptureOperation: 1 (OperationKind.FlowCapture, Type: null, IsImplicit) (Syntax: 'true')
                  Value: 
                    ILiteralOperation (OperationKind.Literal, Type: System.Boolean, Constant: True) (Syntax: 'true')

            Next (Regular) Block[B6]
                Leaving: {R2}
    }

    Block[B6] - Block
        Predecessors: [B3] [B4] [B5]
        Statements (1)
            IExpressionStatementOperation (OperationKind.ExpressionStatement, Type: null) (Syntax: 'result = in ... };')
              Expression: 
                ISimpleAssignmentOperation (OperationKind.SimpleAssignment, Type: System.Boolean) (Syntax: 'result = in ... }')
                  Left: 
                    IFlowCaptureReferenceOperation: 0 (OperationKind.FlowCaptureReference, Type: System.Boolean, IsImplicit) (Syntax: 'result')
                  Right: 
                    IFlowCaptureReferenceOperation: 1 (OperationKind.FlowCaptureReference, Type: System.Boolean, IsImplicit) (Syntax: 'input switc ... }')

        Next (Regular) Block[B7]
            Leaving: {R1}

Note how B6 is using flow capture 1, which is not initialized by B4. It feels like we need to skip branches for pure discard branches in the flow tree.

[AlekseyTs]

@333fred

I've dug through the test failures here. The problem is that we are introducing a branch for the following (example) tree:

What unit-test is that?

[333fred]

Alright, looking at the test failures there is quite a bit more that would need to be done to eliminate false jumps here. For example, boolVal switch { true => ..., false => ... } needs to not generate jumps on the false case because it's an unconditional fallthrough. @bernd5, can you please force this back before the changes to the CFG builder and we can file a followup to do this in a separate PR? It's far too complex to try and do right now.

[AlekseyTs]

LGTM (commit 33)

[333fred]

Thanks for the patience and persistence @bernd5!