[Issue 13897] Wrong purity of inner function

[via Digitalmars-d-bugs]

https://issues.dlang.org/show_bug.cgi?id=13897

Kenji Hara <k.hara.pg at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |INVALID

--- Comment #2 from Kenji Hara <k.hara.pg at gmail.com> ---
(In reply to bearophile_hugs from comment #0)
> This compiles with no errors in dmd 2.067alpha:

It's an intentional language change.

Until 2.066, the nested function foo had been deduced as a strong purity
function. But if foo is declared in struct or class scope, it had been deduced
as a weak purity function. That was inconsistent.

// 2.066 and earlier
void main() {
    int x;
    void foo() pure {  // wrongly deduced as strong purity
        x++;    // modifying enclosing scope via context pointer was rejected.
    }
}
struct S {
    int x;
    void foo() pure {  // weak purity
        x++;    // OK
    }
}

In git-head, a nested function without any type qualifiers is deduced to weak
purity so it has a hidden mutable context pointer, and the semantics is
consistent with member functions.

// 2.067alpha
void main() {
    int x;
    void foo() pure {  // deduced as weak purity, so
        x++;    // modifying enclosing scope via context pointer is accepted.
    }
    void bar() immutable pure {  // deduced as strong purity
        x++;    // NG
    }
}
struct S {
    int x;
    void foo() pure {  // weak purity
        x++;
    }
    void bar() immutable pure {  // strong purity
        x++;    // NG
    }
}

See also issue 9148 comments for more detail.

--