TypeAdapterConfig.GlobalSettings.Compile() throws exception when open generics are configured

[cjung95]

When a mapping for a generic type is configured like described here TypeAdapterConfig.GlobalSettings.Compile() throws a Mapster.CompileException: 'Operation is not valid due to the current state of the object.' with an inner ArgumentException: Type ClassA`1[T] is a generic type definition (Parameter 'type').

Minimal working example:

using Mapster;

TypeAdapterConfig.GlobalSettings.ForType(typeof(ClassA<>), typeof(ClassB<>));
TypeAdapterConfig.GlobalSettings.Compile();

var classA = new ClassA<int> { Variable = 15 };
var classB = classA.Adapt<ClassB<int>>();

class ClassA<T> { public T? Variable { get; set; } = default; }
class ClassB<T> { public T? Variable { get; set; } = default; }

When removing TypeAdapterConfig.GlobalSettings.Compile(); the mapping works just fine.

The exception is thrown in Mapster.Adapters.BaseAdapter.CreateAdaptFunc(CompileArgument arg) when calling Expression.Parameter(arg.SourceType).

Mapster version 10.0.7

[DocSvartz]

Hi @cjung95 I checked, this isn't a regression in v10. It also threw an error in version 7.4.0.
It seems to actually work like a hidden Inherits - when a child class copies the parent's settings. But the OpenGeneric configuration itself ends up in the shared settings store.
And since it's impossible to create an OpenGeneric instance, it doesn't compile, as it's impossible to create a function that instantiates OpenGeneric.

[cjung95]

Hi @DocSvartz,
thanks for you reply.
I'm running the Compile() function to validate my configuration, so I need a solution here.
Should I figure out a solution myself and create a pull request, or will someone on your team take on this issue?

[DocSvartz]

I'm currently working on a new way to create config for OpenGeneric
using marker

TypeAdapterConfig<ClassA<IOpenGeneric>, ClassB<IOpenGeneric>>
.NewConfig()
.Map(dest => dest.Variable2, src => src.Variable);

This problem won't occur there.

In your case, can remove the error by removing the check for Opengeneric types.
Because the check result is actually valid. 😅

[cjung95]

I don't quite follow your argument. Registering open generic types is valid according to the specification. However, when you run the function provided in the specification to validate the configuration, it throws an exception. Yet you say that the validation result is correct. Could you please explain that to me?

[DocSvartz]

Compile is the creation of a function to convert a source type to a destination type.
By calling Compile, you're actually trying to do the following:

var result = new ClassB<>() // - Which is an invalid operation

This classA.Adapt<ClassB<int>>(); in turn means:
Convert ClassA<int> to ClassB<int> by copying the mapping settings for the types ClassA<> to ClassB<> if it exist

var result = new ClassB<int>() // which is now a valid operation

[cjung95]

Then the logical conclusion would be to generally exclude open generic types from the Compile function, otherwise, the documentary is incorrect.
Do you disagree?

// TypeAdapterConfig.cs
foreach (var key in keys)
{
    try
    {
        if (key.Source == typeof(void))
            continue;

        // Added
        if (key.Source.ContainsGenericParameters)
            continue;

        _mapDict[key] = Compiler(CreateMapExpression(key, MapType.Map));
        _mapToTargetDict[key] = Compiler(CreateMapExpression(key, MapType.MapToTarget));
    }
    catch (Exception ex)
    {
        if (failFast)
        {
            throw;
        }

        exceptions.Add(ex);
    }
}

[DocSvartz]

Then the logical conclusion would be to generally exclude open generic types from the Compile function, otherwise, the documentary is incorrect.

Yes, this is the fastest solution. But in this case, there won't be any validation at all.
As far as I understand, the validation is run mainly to ensure that all fields are mapped.

[DocSvartz]

@cjung95 I can add this disabling of the check as a temporary solution.
But really, we need to come up with a proper, permanent solution 😒

[cjung95]

If that's okay with you, I'll take a closer look at this issue next week and find a solution that allows for proper validation even with open generics. Let's not make any hasty changes here.

[DocSvartz]

@cjung95 Ok,

solution using the OpenGeneric definition via proxy stub

TypeAdapterConfig<ClassA<IOpenGeneric>, ClassB<IOpenGeneric>>
.NewConfig()
.Map(dest => dest.Variable2, src => src.Variable);

solves this problem. However, Include() doesn't work correctly yet. 😅

[DocSvartz]

Hi @cjung95. This should be fixed in https://www.nuget.org/packages/Mapster/10.0.8-pre04

But this is only for the basic case, when the generic has one parameter.