Another semantic analysis order bug

[kinke]

This testcase is very similar to #21098 - the real-world code hasn't been fixed yet, so this is another dustmite reduction based on DMD stable (with #21132). As before, compiling b.d works, but a.d fails:

$ dmd -o- a.d
phobos.d(47): Error: expression `hasToString!(const(Nullable!(Type[])))` of type `void` does not have a boolean value

a.d:

import b;

b.d:

import phobos : Appender, Nullable;

struct Type {
    Nullable!(Type[]) templateArgs;
}

Type[] parseDeclarations() {
    Appender!(Type[]) members;
    return null;
}

enum ast = parseDeclarations();

phobos.d:

module phobos;



// module std.typecons;

struct Nullable(T)
{
    T _value;

    string toString() const
    {
        Appender!string app;
        formatValue(app, _value);
        return null;
    }
}



// module std.array;

struct Appender(A)
{
    InPlaceAppender!A impl;
}

struct InPlaceAppender(T)
{
    static void toStringImpl(const T[] data)
    {
        string app;
        formatValue(app, data);
    }
}



// module std.format.internal.write;

void formatValue(Writer, T)(Writer w, T val)
{
    static if (is(T == E[], E))
        formatValue(w, val[0]);
    else static if (is(T == struct) || is(T == union))
    {
        static if (hasToString!T)
            int dummy;
        formatValue(w, val.tupleof);
    }
}

template hasToString(T)
{
    static if (is(typeof(
        (T val) {
            val.toString(123);
        })))
    {
        enum hasToString = true;
    }
    else
    {
        enum hasToString = false;
    }
}

[kinke]

Interestingly, b.d starts failing with the same error when changing

struct Appender(A)
{
    InPlaceAppender!A impl;
}

struct InPlaceAppender(T)
{
    static void toStringImpl(const T[] data)
    {
        string app;
        formatValue(app, data);
    }
}

to

struct Appender(A)
{
    static void toStringImpl(const A[] data)
    {
        string app;
        formatValue(app, data);
    }
}