SourceGenerators & partial methods

[DoctorVanGogh]

Writing a source generator I've come across a situation with partial methods where I am not sure if my scenario is just unsupported "iffy", or I'm doing something wrong.

Imagine a source generator which outputs source that contains partial method stubs:

Foo.generated.cs:

/// <autogenerated />
public partial class Foo {
  public partial object Bar(int wiz, string bang);
}

Since that is partial method with both a visibility modifier and a return type (either would suffice) it must have an implementation part. (See https://learn.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/partial-classes-and-methods#partial-methods).

Now imagine a second manually written source file containing said implementation:
Foo.cs

public partial class Foo {
  public partial object Bar(int wiz, string bang) {
     return $"{wiz}-{bang}";
  }
}

If both files were part of a regular project this would compile fine. If the manually written implementation was missing, you would get a nice compiler error for that.

---

Now if the defining declaration comes out of a source generator the compiler may error with
No defining declaration found for implementing declaration of partial method 'Foo.Bar(int, string)'

I've specifically had those errors when using a non incremental ISourceGenerator - presumably since the non incremental generator seems to run after all other compilations.
Using IIncrementalGenerator I've been able to sidestep this (mostly*) since the generated source seems to be created early enough for roslyn to include it in the analysis of the manually crafted file.

I'm assuming that combining my value providers with the CompilationProvider would once again put an arrow in my knee here.

Now my question is: Is this intentional? Shouldn't roslyn only give diagnostic errors for potentially multi-file issues once it has all files ?

[CyrusNajmabadi]

Shouldn't roslyn only give diagnostic errors for potentially multi-file issues once it has all files ?

Yes.

If you are seeing otherwise, it is a bug. We'd need a repro case demonstrating the issue so someone could investigate. Thanks!

[jasonmalinowski]

Indeed, there should be no change of behavior whether you're using ISourceGenerator or IIncrementalGenerator.

[DoctorVanGogh]

Okay, that's rather good news (and bad). 😅

I was having significant issues getting the source generator to work in all build situations due to FileNotFound exceptions with (transitive) assembly dependencies. The missing definition errors may just have coincided with those initialization errors (and at those times been correct from a compiler perspective).
I think I eliminated all my initialization issues at the same time as switching to the incremental generator. So it might just have been that (and me thinking "Of course it works now: The incremental generator creates the source as soon as the value provider chain is satisfied so the generated code exists before any actual build")

I'll give it another look and maybe put in an issue or two regarding dependencies (and any other weird behavior I've seen).