@safe
class Y { }

@safe
class X { }

@safe
class Z
{
    int x;

    this()
    {
        if (typeid(X) == typeid(Y)) x = 1; // Compile Error
        else x = 2;
    }
}

void main() { new Z; }

// test.d(19): Error: safe function 'test.Z.this'
// cannot call system function 'object.opEquals'

Isn't this analagous to saying that the "instanceof" operator
in java endangers the GC?

Is it correct to replace '==' with 'is'?


-Eric

On Monday, November 24, 2014 22:12:08 Eric via Digitalmars-d-learn wrote:
>
> @safe
> class Y { }
>
> @safe
> class X { }
>
> @safe
> class Z
> {
>      int x;
>
>      this()
>      {
>          if (typeid(X) == typeid(Y)) x = 1; // Compile Error
>          else x = 2;
>      }
> }
>
> void main() { new Z; }
>
> // test.d(19): Error: safe function 'test.Z.this'
> // cannot call system function 'object.opEquals'
>
> Isn't this analagous to saying that the "instanceof" operator in java endangers the GC?
>
> Is it correct to replace '==' with 'is'?

It's not that it's inherently unsafe. The problem is a combination of the
fact that stuff in druntime that pre-existed @safe hasn't been made @safe
yet (particularly, stuff in TypeInfo) and the fact that Object shouldn't
even have opEquals, opCmp, toHash, or toString on it, because that restricts
which attributes can be used
( https://issues.dlang.org/show_bug.cgi?id=9769 ), though I think that with
@safe, we can work around that (unlike with const). However, for whatever
reason, TypeInfo's opEquals function hasn't been marked with @safe or
@trusted, so it's considered @system. That will need to be fixed, but I
don't know if there are any implementation issues preventing it. It _looks_
like it could probably be marked @trusted, but I haven't actually dug into
it in detail.

In any case, you should be able to just mark the constructor as @trusted for now to work around the issue, and at some point in the future opEqualso or TypeInfo should be @trusted or @safe.

- Jonathan M Davis

On Tuesday, 25 November 2014 at 02:48:43 UTC, Jonathan M Davis via Digitalmars-d-learn wrote:
> On Monday, November 24, 2014 22:12:08 Eric via Digitalmars-d-learn wrote:
>>
>> @safe
>> class Y { }
>>
>> @safe
>> class X { }
>>
>> @safe
>> class Z
>> {
>>      int x;
>>
>>      this()
>>      {
>>          if (typeid(X) == typeid(Y)) x = 1; // Compile Error
>>          else x = 2;
>>      }
>> }
>>
>> void main() { new Z; }
>>
>> // test.d(19): Error: safe function 'test.Z.this'
>> // cannot call system function 'object.opEquals'
>>
>> Isn't this analagous to saying that the "instanceof" operator
>> in java endangers the GC?
>>
>> Is it correct to replace '==' with 'is'?
>
> It's not that it's inherently unsafe. The problem is a combination of the
> fact that stuff in druntime that pre-existed @safe hasn't been made @safe
> yet (particularly, stuff in TypeInfo) and the fact that Object shouldn't
> even have opEquals, opCmp, toHash, or toString on it, because that restricts
> which attributes can be used
> ( https://issues.dlang.org/show_bug.cgi?id=9769 ), though I think that with
> @safe, we can work around that (unlike with const). However, for whatever
> reason, TypeInfo's opEquals function hasn't been marked with @safe or
> @trusted, so it's considered @system. That will need to be fixed, but I
> don't know if there are any implementation issues preventing it. It _looks_
> like it could probably be marked @trusted, but I haven't actually dug into
> it in detail.
>
> In any case, you should be able to just mark the constructor as @trusted for
> now to work around the issue, and at some point in the future opEqualso or
> TypeInfo should be @trusted or @safe.
>
> - Jonathan M Davis

Thanks for reminding me about @trusted.  I'm finding it really hard
to write robust classes in D due to all the problems with Object.

-Eric

On Tuesday, 25 November 2014 at 03:42:50 UTC, Eric wrote:
> I'm finding it really hard to write robust classes in D
> due to all the problems with Object.

I wish Object didn't have any methods. One thing I do is kinda act as if it didn't and make my own interfaces instead.