Diagnostic is unclear when passing ref readonly to ref readonly parameter using 'ref'

[RikkiGibson]

Diagnostic is unclear when passing ref readonly to ref readonly parameter using 'ref'

[jjonescz]

Note that you would get the same error with in parameter (so this issue is orthogonal to the ref readonly feature):

class C
{
    void M0(in int i) { }
    void M1(in int i)
    {
        M0(ref i); // error CS8329: Cannot use variable 'i' as a ref or out value because it is a readonly variable
    }
}

ref modifier simply cannot be used with readonly references. But I agree the diagnostic message could be clearer.

[RikkiGibson]

I am still interested to know why the design requires writability for a ref readonly argument passed using ref keyword.

The reason I found this design so confusing is that it is inconsistent with all the existing places that I can create ref readonlys, including assignment to ref readonly locals and return values.

int i = 1;
ref readonly int ri = ref i; // ok
ri = ref i; // reassignment also ok
M(ref i); // ok?
M(ref ri); // error!

ref readonly int M(ref readonly int i) => ref i; // ok

So, when you get started using this feature, and ref keyword works for some arguments, it leads user to think: ok, I'm supposed to use ref for these arguments, like all the other places that use ref readonlys in the language. Until they get around to actually passing a readonly reference by-ref, in which case an error occurs.

If we really have a strong reason for requiring writability here and not in the other places, then OK. Let's resolve by simply issuing a diagnostic which is consistent with all the other places that a wrong argument ref kind is being used: error CS1620: Argument 1 must be passed with the 'in' keyword.

[jjonescz]

I am still interested to know why the design requires writability for a ref readonly argument passed using ref keyword.

The reasons are explained in the proposal under alternatives:

[RikkiGibson]

Thanks Jan!