Missing configuration options for transformStrings

[mipastgt]

The feature transformStrings of the new plugin is interesting but I see a few problems with it.

1. Which high-level language actually supports returning two i32 values indicating the pointer and length of a string returned in memory?
2. How would you distinguish this pattern from other uses? E.g., in my code I return byte arrays and not strings in a similar way. In a real program there could also be a mixture of such uses. I don't see how these cases could be distinguished from each other without further configuration options.

[CharlieTap]

Most compilers (Kotlin, Go, Rust and more) default to the encoding strategy of two integers, either two I32s if running wasm32 or two I64 if running the wasm64 proposal. That being said I know some C to wasm compilers will do a single I32 if they're encoding c strings as these are null terminated.

I appreciate there's many strategies and ultimately it depends on the compiler/toolchain or even the program itself. In the latest version of the plugin (0.2.0) I've added support for configuration of the function codegen and specifically string encoding.

You can now do this:

chasm {
    modules {
        create("TestService") {
            binary = ...
            packageName = ...
            function("string_function") {
                stringReturnType(StringEncodingStrategy.NULL_TERMINATED)
            }
        }
    }
}

where the encoding strategy is one of the following:

enum class StringEncodingStrategy {
    /**
     * The pointer and length of the string are encoded in two I32s
     */
    POINTER_AND_LENGTH,

    /**
     * The pointer of the string is encoded in an I32 and end of the String is the first null byte
     * encountered after that pointer
     */
    NULL_TERMINATED,

    /**
     * The pointer of the string is encoded in an I32 and length of the string is encoded in the int
     * found at that pointer
     */
    LENGTH_PREFIXED,

    /**
     * The pointer and length of the string are encoded in a single I64
     */
    PACKED_POINTER_AND_LENGTH,
}