VB support for CallerArgumentExpression

[Youssef1313]

Proposal: dotnet/csharplang#287
Test plan: #52745

Addresses part of LDM notes: https://github.com/dotnet/csharplang/blob/main/meetings/2021/LDM-2021-06-14.md

[AlekseyTs]

CallerArgumentExpressionParameterIndex

This looks fragile and probably incorrect, at least for delegate EndInvoke method. This implementation assumes that the target parameter is still going to be present and will remain at the same position, which probably can be wrong. The more robust approach is to reinterpret the attribute in the new signature. But before we do this, let's create some tests and figure out if this code path is even reachable in interesting scenarios. Perhaps simply returningh -1 is going to be good enough. #Closed

[AlekseyTs]

BTW, I bet C# has similar class, we should check it as well.

[Youssef1313]

@AlekseyTs EndInvoke should only be cloning parameters that are ref or out:

roslyn/src/Compilers/CSharp/Portable/Symbols/Source/SourceDelegateMethodSymbol.cs

Lines 402 to 409 in 22a7203

	foreach (SourceParameterSymbol p in invoke.Parameters)
	{
	if (p.RefKind != RefKind.None)
	{
	var synthesizedParam = new SourceClonedParameterSymbol(originalParam: p, newOwner: this, newOrdinal: ordinal++, suppressOptional: true);
	parameters.Add(synthesizedParam);
	}
	}

So it is not going to ever clone a caller argument expression parameter. Hence, this path is unreachable for an EndInvoke. However, the implementation is needed for indexers and delegate invoke methods.

[AlekseyTs]

So it is not going to ever clone a caller argument expression parameter.

Why do you think so? Is there a test that demonstrates that?

[AlekseyTs]

However, the implementation is needed for indexers and delegate invoke methods.

I see nothing in this type that limits its application to these scenarios. Therefore, the implementation is not robust in the long term in my opinion.

[Youssef1313]

Why do you think so?

Per the code above that creates SourceClonedParameterSymbol for EndInvoke, a parameter is only created if its RefKind == RefKind.None.

CallerArgumentExpressionAttribute is only allowed for optional parameters (which should have RefKind.None.

Applying the attribute to a ref or out parameter generates a compile-time error:

error CS9006: The CallerArgumentExpressionAttribute may only be applied to parameters with default values

These are just my thoughts, I may be missing some interesting cases that you may have.

[AlekseyTs]

CallerArgumentExpressionAttribute is only allowed for optional parameters (which should have RefKind.None.

Is this accurate? One cannot do this in IL or VB?

Applying the attribute to a ref or out parameter generates a compile-time error:

error CS9006: The CallerArgumentExpressionAttribute may only be applied to parameters with default values

Why are we talking about C# error, when we are working in VB? What prevents applying this attribute in IL for example?

This is perfectly valid code:

class Test
    Shared Sub M(Optional ByRef x as String = "ddd")
    End Sub
    
    Shared Sub Test()
        M()
    End SUb
End Class

[AlekseyTs]

I see nothing in this type that limits its application to these scenarios. Therefore, the implementation is not robust in the long term in my opinion.

One way out could be making this type MustInherit, creating specialized NotInheritable derived types for specific scenarios, implementing the API in the derived types only.

[Youssef1313]

@AlekseyTs

Behavioral-wise:

• VB delegates doesn't allow optional parameter, hence, calling CallerArgumentExpressionIndex should never be done for EndInvoke invocations (the binder calls this property only if we're binding an optional parameter).

• For IL, shouldn't we be dealing with metadata symbols, not source symbols? or I'm misunderstanding the concept here?

---

Implementation-wise:

I'm okay with making the type abstract and introducing derived type.

[AlekseyTs]

I'm misunderstanding the concept here?

The discussion like that can go forever and will never end because reviewer's concerns can almost never be addressed by gueses, speculations or asumptions. The best way to alleviate reviewer's concerns is to create a test for the scenario in question and demonstrate the behavior, in this case the behavior of the API. Specifically here we are discussing an API behavior, whether there is an error reported or not has very little impact to the question what API's behavior is and what it should be. We are not discussing an issue with lowering/code generation. Since the type is not specific to any single particular scenario (a unique reason for creating a clone), each scenario should be tested individually and the tests should demonstrate that the implementation is in fact necessary and correct. Neccessity can be demonstrated by observing a break for a successful scenario when the implementation is removed/changed. For example, you are saying: "the implementation is needed for indexers and delegate invoke methods." Please demonstrate the fact with unit tests.

I'm okay with making the type abstract and introducing derived type.

I would prefer that we will create more specialized derived types. This way one can reason when instances are created and how they are used, can confirm that the code path is tested. Today anyone can create an instance of SourceClonedParameterSymbol for whatever reason and "take" an implementation of this API, whether appropriate or not, with it without even be aware of the possible impact. I think we should prevent this from happening.

[AlekseyTs]

CallerArgumentExpressionParameterIndex

Why is this the right value? Ae we testing this code path? #Closed

[Youssef1313]

@AlekseyTs I don't have enough info about COM, is it expected that order of parameters can change, etc?

Trying to test this code path locally with the following test, but there are few errors:

        <Fact>
        Public Sub ComClass()
            Dim source As String = "
Imports Microsoft.VisualBasic

<ComClass(ComClass1.ClassId, ComClass1.InterfaceId, ComClass1.EventsId)>
Public Class ComClass1
    ' Use the Region directive to define a section named COM Guids. 
#Region ""COM GUIDs""
    ' These  GUIDs provide the COM identity for this class 
    ' and its COM interfaces. You can generate 
    ' these guids using guidgen.exe
    Public Const ClassId As String = ""7666AC25-855F-4534-BC55-27BF09D49D46""
    Public Const InterfaceId As String = ""54388137-8A76-491e-AA3A-853E23AC1217""
    Public Const EventsId As String = ""EA329A13-16A0-478d-B41F-47583A761FF2""
#End Region

    Public Sub New()
        MyBase.New()
    End Sub

    Public Sub M(x As Integer, Optional y As String = """")
    End Sub
End Class

Module Program
    Sub Main()
        Dim comObj As New ComClass1()
        comObj.M(1+  0)
    End Sub
End Module
"
            Dim compilation = CreateCompilation(source, targetFramework:=TargetFramework.NetCoreApp, references:={Net451.MicrosoftVisualBasic}, options:=TestOptions.ReleaseExe, parseOptions:=TestOptions.RegularLatest)
            CompileAndVerify(compilation, expectedOutput:="").VerifyDiagnostics() ' PROTOTYPE(caller-expr): Figure out how to fix these:
            ' (5) : error BC35000: Requested operation is not available because the runtime library function 'System.Runtime.InteropServices.GuidAttribute..ctor' is not defined.
            ' (5) : error BC35000: Requested operation is not available because the runtime library function 'System.Runtime.InteropServices.ClassInterfaceAttribute..ctor' is not defined.
            ' (5) : error BC35000: Requested operation is not available because the runtime library function 'System.Runtime.InteropServices.DispIdAttribute..ctor' is not defined.
        End Sub

[AlekseyTs]

I don't have enough info about COM, is it expected that order of parameters can change, etc?

I do not remember these details. Since you are the PR author, it is on you to "defend" the changes. I would start with reviewing parts of the code that create symbols like this and analyze how and in what fashion this is happening. For tests, I would try to locate tests that are targeting this feature and would try to clone/adjust them as appropriate.

[Youssef1313]

@AlekseyTs I experimented with this. As far as I can tell, the SynthesizedComParameters are only exposed to emitter. It's not exposed to binder at all. It feels safe to just throw ExceptionUtilities.Unreachable if my conclusions are correct.

[AlekseyTs]

I experimented with this. As far as I can tell, the SynthesizedComParameters are only exposed to emitter. It's not exposed to binder at all. It feels safe to just throw ExceptionUtilities.Unreachable if my conclusions are correct.

Looks this way to me too.

[AlekseyTs]

So far, there are no tests.

---

In reply to: 862484319

[Youssef1313]

Thanks for the review @AlekseyTs! I'll work on it (but I may delay on this due to exams).

[AlekseyTs]

Done with review pass (commit 2)

[Youssef1313]

Roslyn Error: test timeout exceeded, dumping remaining processes

closing and reopening for new build.

[Youssef1313]

##[error]The request: 13684396 was abandoned due to an infrastructure failure. Notification of assignment to an agent was never received.

[Youssef1313]

Some jobs are failing with:

D:\workspace\_work\1\s\src\Scripting\CSharp\Microsoft.CodeAnalysis.CSharp.Scripting.csproj : error NU3037: Package 'System.Reflection.Emit.ILGeneration 4.3.0' from source 'https://pkgs.dev.azure.com/dnceng/public/_packaging/dotnet-public/nuget/v3/index.json': The repository primary signature validity period has expired. [D:\workspace\_work\1\s\src\NuGet\Microsoft.Net.Compilers.Toolset\Microsoft.Net.Compilers.Toolset.Package.csproj]
D:\workspace\_work\1\s\src\Scripting\CSharp\Microsoft.CodeAnalysis.CSharp.Scripting.csproj : error NU3037: Package 'System.Runtime.Handles 4.3.0' from source 'https://pkgs.dev.azure.com/dnceng/public/_packaging/dotnet-public/nuget/v3/index.json': The repository primary signature validity period has expired. [D:\workspace\_work\1\s\src\NuGet\Microsoft.Net.Compilers.Toolset\Microsoft.Net.Compilers.Toolset.Package.csproj]
D:\workspace\_work\1\s\src\Scripting\CSharp\Microsoft.CodeAnalysis.CSharp.Scripting.csproj : error NU3037: Package 'System.Text.Encoding 4.3.0' from source 'https://pkgs.dev.azure.com/dnceng/public/_packaging/dotnet-public/nuget/v3/index.json': The repository primary signature validity period has expired. [D:\workspace\_work\1\s\src\NuGet\Microsoft.Net.Compilers.Toolset\Microsoft.Net.Compilers.Toolset.Package.csproj]

and others with:

##[warning]There was a failure in sending the provision message: A timeout occurred while sending request to the remote provider.

I'm seeing CI issues with another recent PR too. Will give some time before re-triggering a new build.

[AlekseyTs]

@333fred, @dotnet/roslyn-compiler For the second review.

[333fred]

-2

Consider making this a constant, and not -2 since we already use -2 for reduced extension method receivers. #Closed

[333fred]

IdentifierComparison.Equals(parameters(i).Name, parameterName)

Couple of notes:

1. I think the C# version of this method is broken, as it's using default string comparison. It should be using Ordinal.
2. This is potentially going to mean differing behavior between C# and VB. I think we likely should be using Ordinal here as well, not case-insensitive comparison. If VB defined a parameter named param but referred to it as Param in the attribute, we would accept that here, but not in C#. #Closed

[333fred]

I'm unsure why this reasoning applies to this parameter, but not the rest of the caller info attributes? #Closed

[Youssef1313]

@333fred It should apply, but I avoided changing the existing implementation. Let me change it so things are less confusing.

[333fred]

IdentifierComparison.Equals(parameters(i).Name, parameterName)

Same comments as PEParameterSymbol

[333fred]

P

Unless we're somehow adjusting this during emit to be the name that can be consumed by C#/F#, this doesn't seem correct. #Closed

[333fred]

Done review pass (commit 44)

[Youssef1313]

@333fred I addressed your feedback. Can you take another look? Thanks!

[AlekseyTs]

parameters(i).Name.Equals(parameterName, StringComparison.Ordinal)

I do not think I agree with this change. Languages are already different in terms of case-sensitivity. I believe, the difference in behavior can be observed with named arguments, for example. And probably with other similar scenarios. @333fred, if you believe language rules should not apply for this scenario, please bring this to LDM #Closed

[Youssef1313]

@AlekseyTs Should I add a prototype until this is discussed in LDM?

My 2 cents: The current language case-insensitivity doesn't affect consuming the code in C#, but in this case, it will affect it as @333fred stated. So I'm leaning towards being case-sensitive.

[AlekseyTs]

Should I add a prototype until this is discussed in LDM?

I would be fine with behavior on which I signed off (i.e. case-insensitive behavior consistent with named arguments) and a prototype comment to confirm that at LDM. I personally fine with putting the burden of interop with other languages (if one is required) on the developer and leave it to him/her to match the case.

[333fred]

I do disagree, but as long as we have a prototype comment to follow up on this I'm ok with merging.

The sign-off is dismissed due to recent changes.

[AlekseyTs]

If parameters(i).Name.Equals(parameterName, StringComparison.Ordinal) Then

I believe language rules should apply here. #Closed

[333fred]

LGTM (commit 47)

[AlekseyTs]

LGTM (commit 47)

[AlekseyTs]

@Youssef1313 Thanks for the contribution.