Re: C++17

[Shachar Shemesh via Digitalmars-d]

On 27/01/16 16:24, Minas Mina wrote:

On Wednesday, 27 January 2016 at 14:22:18 UTC, Shachar Shemesh wrote:

On 26/01/16 11:33, deadalnix wrote:

On Tuesday, 26 January 2016 at 09:16:47 UTC, Ola Fosheim Grøstad wrote:

[...]

Now if one want to use that, D is very capable of doing it already. Just
won't make it the default (like it is not the default in C++ either).

I bring it up every time the subject comes up, in the hopes that at
some point it will sink in.

No, D is not capable of doing it already. Without 100% reliable
destructors, RAII is simply not implementable.

D's destructors are not guaranteed to run on 100% of fully initialized
structs, which means that a RAII container has no way to make sure its
resource is actually freed. It is up to the implementer. This
eradicates almost all of the utility RAII was meant to provide.

Shachar

Can you show a case where this happens?

Sure. Here it is, again:

import std.stdio;

struct RAII {
    int num;

    ~this() {
        writefln("Destructed %s", num);
    }
}

struct Container {
    int num;
    RAII raii;

    this(int num) {
        writefln("Constructing %s", num);
        this.num = num;
        raii = RAII(num);

        if( num==0 )
            throw new Exception("Ooops");
    }
}

void test(int num) {
    try {
        auto c = Container(num);
        writefln("Point %s", num);
    } catch(Exception ex) {
        writefln("Caught %s", ex.msg);
    }
}

void main()
{
    test(1);
    test(0);
}