I've been looking into (basic) memory management within D. Through IRC conversation and reading the article on memory management on dlang.org (which seems to be a bit out-of-date), I've concluded that using a global (or static member) function and emplace() in std.conv is a simple solution for providing alternative allocation methods. However, I cannot, by default, scope my custom allocations. Take this for instance:

void main() {
	MyClass inst = alloc!(MyClass)();
	inst.do_something();
	dealloc(inst);
}

Now, I want my dealloc function to be called automagically. One safe measure is:

void main() {
	MyClass inst = alloc!(MyClass)();
	scope(exit) dealloc(inst);
	inst.do_something();
}

But I'm lazy and I'm looking for shorter methods.

void main() {
	mixin AutoRef(MyClass, "inst");
	inst.do_something();
}

Any ideas?

You can use structs for this kind of problems, somethin like

struct Scoped(T)
{
    ~this()
    {
        //handle deallocation here
    }
    T inst;
}

if you create a Scoped!MyClass in a scope, the struct's destructor will be invoked whenever the program leaves the scope, even if it is because of an exception.

On Tue, Jan 31, 2012 at 2:40 PM, Zachary Lund <admin@computerquip.com> wrote:
> I've been looking into (basic) memory management within D. Through IRC conversation and reading the article on memory management on dlang.org (which seems to be a bit out-of-date), I've concluded that using a global (or static member) function and emplace() in std.conv is a simple solution for providing alternative allocation methods. However, I cannot, by default, scope my custom allocations. Take this for instance:
>
> void main() {
>        MyClass inst = alloc!(MyClass)();
>        inst.do_something();
>        dealloc(inst);
> }
>
> Now, I want my dealloc function to be called automagically. One safe measure is:
>
> void main() {
>        MyClass inst = alloc!(MyClass)();
>        scope(exit) dealloc(inst);
>        inst.do_something();
> }
>
> But I'm lazy and I'm looking for shorter methods.
>
> void main() {
>        mixin AutoRef(MyClass, "inst");
>        inst.do_something();
> }
>
> Any ideas?

> However, I cannot, by default, scope my custom allocations.
> Any ideas?

std.typecons.scoped

On Tuesday, 31 January 2012 at 15:19:00 UTC, Trass3r wrote:
>> However, I cannot, by default, scope my custom allocations.
>> Any ideas?
>
> std.typecons.scoped

I looked into this and I'm unsure of its exact use. It says, "Allocates a class object right inside the current scope" which doesn't really define how it's allocated nor does it explain how this would work with custom de/allocators. Also, it claims it "avoids" the overhead of new of which I'm not entirely sure of what it means. Could some clarification be made?

On Tue, 31 Jan 2012 16:42:52 +0100, Zachary Lund <admin@computerquip.com> wrote:

> On Tuesday, 31 January 2012 at 15:19:00 UTC, Trass3r wrote:
>>> However, I cannot, by default, scope my custom allocations.
>>> Any ideas?
>>
>> std.typecons.scoped
>
> I looked into this and I'm unsure of its exact use. It says, "Allocates a class object right inside the current scope" which doesn't really define how it's allocated nor does it explain how this would work with custom de/allocators. Also, it claims it "avoids" the overhead of new of which I'm not entirely sure of what it means. Could some clarification be made?

If scoped is used inside a struct or class, it allocates the space within
that struct or class. If used inside a function, the space is allocated
on the stack. The size of the allocated buffer must be known at compile
time, so you may not put an instance of a subclass into a scoped variable.

Because the class may be allocated on the stack, returning it from a
function is a Bad Idea™, unless it's part of a struct or class you know
will survive for a while longer.

The reason it avoids the overhead of new is it does not allocate using
new. Basically, Scoped does this:

class A {}

A scoped = cast(A)alloca(sizeof(A));
A.__ctor(args);