On another thread I pondered that the D2.0 syntax for const member function was ambiguous. I've now come to believe it's also wrong.

What we're talking about is a function which would be declared like this in C++ (with const at the end)

class C
{
    int f() const { /* ... */ }
}

in D2.0, apparently the syntax is:

class C
{
    invariant int f() { /* ... */ }
}

Here's my problem. f isn't invariant, by D's use of the word.

What makes f a member function is that it has access to a hidden variable called "this". What makes f a /const/ member function is that the type of "this" is const(C). NOT invariant(C). f gets a read-only view of "this" - so, even by D2.0's keyword names, it should be "const", not "invariant".

As followers of this know, I advocate rename invariant -> const, and const -> readonly in any case, so under the renamed scheme, we would call f a readonly function, because it has a readonly view of "this".

My suggestion for the syntax is to position the word "readonly" immediately to the left of the function name. Thus:

class C
{
    readonly int f() { /* ... */ } /* this has type C; return type is
readonly(int) */
    int readonly f() { /* ... */ ) /* this has type readonly(C);
return type is int */
}

You could even use the bracket syntax if you wanted...

class C
{
    readonly(int)f() { /* ... */ } /* this has type C; return type is
readonly(int) */
    int readonly(f()) { /* ... */ ) /* this has type readonly(C);
return type is int */
}

Janice Caron wrote:
> On another thread I pondered that the D2.0 syntax for const member
> function was ambiguous. I've now come to believe it's also wrong.
> 
> What we're talking about is a function which would be declared like
> this in C++ (with const at the end)
> 
> class C
> {
>     int f() const { /* ... */ }
> }
> 
> in D2.0, apparently the syntax is:
> 
> class C
> {
>     invariant int f() { /* ... */ }
> }
> 
> Here's my problem. f isn't invariant, by D's use of the word.

Personally, I don't like the dual use of 'invariant' anyway.  That using it for const required that class invariant suddenly needed empty parens to distinguish them suggested to me that it was a poor choice--it was a breaking change.  I've advocated 'view' in place of 'readonly' in the past, but it's perhaps too common a word.


Sean

Sean Kelly wrote:
> I've advocated 'view' in place of 'readonly' in the past, but it's perhaps too common a word.

On the other hand, 'readonly' is kind of a lot to type for something that's going to be used so frequently.  Maybe 'roview'?

Thanks,
Nathan Reed