Standard: Binary serialization is broken when building against netstandard and running on netfx.

[safern]

If an assembly is built against netstandard and run it against netfx (desktop) and it has a type marked as Serialized and this types have any of the event methods that use either OnDeserializedAttribute, OnSerializingAttribute, OnSerializedAttribute, and OnDeserializingAttribute it would throw an exception like this one:

Type <type name> in assembly <assembly name> has method <method name> with an incorrect signature for the serialization attribute that it is decorated with

This would happen even if the method follows the guidelines stated in the msdn documentation:

To use the OnSerializingAttribute, the method must contain a StreamingContext parameter.

We are currently hitting this issue on the System.Runtime.Serialization.Formatters.Tests that have a type with this characteristics (source code) when the test is built against netstandard and we run it against netfx.

After investigating with @weshaggard he found out that there a bug in the methodtablebuilder.cpp when verifying the signature for this method (requiring a StreamingContext parameter) it doesn't follow the typeforwards correctly on the case it was defined on a shim like in the case of netstandard that is a typeforward to mscorlib so it throws an exception that the method signature is bad.

This bug is in Full Framework VM but we should decide if we want to document that we don't support applications/libraries using binary serialization built against netstandard and running on netfx.

See: https://github.com/dotnet/corefx/issues/17575#issuecomment-293657902

[jarenduan]

Could dropping support for these attributes in netstandard 2.0 be an alternative solution?

[weshaggard]

@jarenduan yes that is an alternative approach. @terrajobst what do you think?

[danmoseley]

@AlexGhiondea we might want to port this into full FX?

[danmoseley]

Oh it's TFS 407239 already

[weshaggard]

Yes we should try and get this fixed in .NET Framework.

[ericstj]

Had a crazy idea for a workaround: have the netstandard lib reference an aliased MSCorlib from desktop targeting pack and then the user assembly can reference the StreamingContext from there. That'd get the right typeref in the user assembly that'd make desktop happy.

[weshaggard]

I thought about that but that would cause all sorts of issues as netstandard is a core assembly and it cannot reference any other assemblies otherwise trouble will ensue.

[ericstj]

I'm not talking about doing that in netstandard.dll. I'm talking about having the user assembly do it as a workaround until desktop could be fixed.

[weshaggard]

Are you suggesting that the person building against .NET Standard also add a reference to the desktop mscorlib and then alias it just for these types? Wouldn't that cause some crazy amount of other ambiguous references? Or are you suggesting we produce a custom mscorlib with only StreamContext and whatever else is needed from the signature?

While that might work it seems very heavy handed and might cause issues on other platforms.

[ericstj]

some crazy amount of other ambiguous references

No: because of the alias and the limited use of the serialization event methods (the runtime is the only one who typically calls them).

I'm not considering this as a serious long term solution, just a crazy idea for workaround if someone happens to be blocked. I agree its heavy handed.

[Levvente]

Hi!
Are there any updates regards this issue?
This was a really big hit for us, as we tried to port a huge codebase to netstandard2 and after extensive refactoring and working around not supported classes when we reached to testing all of our serialization codes are broken.
To say the least, I'm really disappointed regards the .Net Standard (Sorry, i'm not being rude, I'm just tired and stressful)

If you have any possible workarounds or any other information related to this bug please answer, I Hope there is some progress since August.

[weshaggard]

Sorry to hear about your issues @Levvente. Unfortunately there isn't a lot we can do on the .NET Standard side of things to address this. It is a bug in the .NET Framework which takes a lot of time to update due to the risk of making changes to it. There isn't any great workarounds for the issue beyond not using the methods marked with the OnSerial*Attributes and instead do the serialization in other ways like the constructor (see https://github.com/dotnet/corefx/blob/a9bcc7965f6da027cd23d5d74c14b4ef6c61b313/src/System.Runtime.Serialization.Formatters/tests/SerializationTypes.cs#L282-L312 as an example).

[ViktorHofer]

What Wes said is true, I just want to add that we are indeed working on a fix which will hopefully be ready for the next version of .NET Framework.

[ViktorHofer]

TODO in corefx: When net472 is released, add its version to our version table in PlatformDetection and conditionalize the existing test to a) only run >=net472 or b) to fail with a TypeLoadException

[safern]

Also would need to remove netfx from the tests configurations tu actually build ve netstandard and run in netfx.

https://github.com/dotnet/corefx/blob/master/src/System.Runtime.Serialization.Formatters/tests/Configurations.props

And remove the comment referencing the issue.

[rockfordlhotka]

Hello, I am wondering if there is any chance this will get resolved in the near future?

Until this is resolved many of us who build components and libraries for the community are stuck releasing a version for full .NET, and a version for netstandard 2.0, because the two just aren't compatible at runtime.

It is a PITA for us as community contributors, and an even bigger PITA for everyone out there who tries to use our NuGet packages!

[jherby2k]

I believe it’s fixed in 4.7.2 but I haven’t confirmed personally.

[ilkengin]

I had also got the error and it is now working like a charm after I moved to .NET Framework 4.7.2. Thanks a lot!

[wtgodbe]

@ViktorHofer @safern is there anything still to be done here?

[ViktorHofer]

Nothing to be done one the corefx side...

[safern]

I don't think so, this would require a netfx fix and there is already an actionable item there.