Annotating primitive mappings?

[kkukshtel]

Running this against a C library the bool values are mapped to byte. Instead I'd like to use bool with [UnmanagedType.I1] in both extern function params and struct members. Is this possible?

[tannergooding]

It would be possible, but it is largely "out of scope" for what ClangSharp is doing today.

The general issue is that bool is not considered "blittable" by default and so using something like [UnmanagedType.I1] involves marshalling. This represents a hidden cost that can be greatly undesirable for the types of interop bindings.

I plan to eventually add a feature switch that assumes [assembly: DisableRuntimeMarshallingAttribute] has been specified and therefore that using bool directly for C bool is "safe". However, this is a .NET 7+ only feature and it will not involve emitting [UnmanagedType.I1]

Adding support for marshalled bindings is an incredibly complex task and is basically impossible to do implicitly due to the wide range of scenarios that even something like char* in C could represent. Consider that for starters that the signedness of char in C is implementation defined. It could be signed or unsigned. Then consider that it is the backing type for string literals as well as 8-bit (typically) integers.

So, a char* could be:

• a null terminated ascii or utf-8 string
• a byte, sbyte, ascii character, or utf-8 character array
• a pointer to a single byte, sbyte, ascii, or utf-8 character
  • this could itself be read/write (equivalent to ref) or write first (equivalent to out)

This requires some mechanism for the consumer of ClangSharp to be able to specify complex annotations per method, field, parameter, etc. There exists some support for doing that today via --with-attribute and --with-type options alongside the name mapping schemes that are supported, but there isn't a trivial way to provide specifically the specifically the marshalling attributes.

[kkukshtel]

Okay gotcha, this makes sense! I also didn't know DisableRuntimeMarshallingAttribute would map to a native bool type instead of the Win32 COM mapping, thanks for the info! I agree based on what you said that Marshalled bindings is potentially far out of scope for a lib that just works to generate blittable signatures/types.