Thread overview
struct in class feature
Dec 09, 2012
js.mdnq
Dec 11, 2012
d coder
Dec 12, 2012
Rob T
Dec 12, 2012
js.mdnq
December 09, 2012
stemming from:

http://forum.dlang.org/thread/azqevtjaoafbapiadgjz@forum.dlang.org

I believe the following idea would offer good benefit for encapsulation in classes(increases the divide and conquer approach):

For structs in classes(call them cstructs if you want):

prototype:

class A
{
    cstruct B // Can only be used inside class A
    {
        int x;
        void func()
        {
            // if func accesses any member of A then a hidden "cthis" is passed to it to allow it to look them up
            a = x*45;
        }
    }

    B b;
    int a;
}

A `cstruct` then, is useful for encapsulating data and functionality in a class into smaller self-contained parts. It would be very useful because it allows us to avoid collisions.


class A
{
    cstruct B
    {
        opAssign;
    }

    opAssign;
}

even though, functionally both opAssigns accept the same this pointer.

The reason being is that since a cstruct is fixed inside class A(any value type of type cstruct has a fixed offset from the start of class A), all the fields of both the class and struct are calculated by simple offsets.

i.e., the struct this ptr is simply an offset from the class this ptr. This allows us to use one or the other but we don't need to store a ptr to a type A inside the struct to access is(and we can also get at the private members of A).

Having such a feature makes it easy to wrap types in a struct and provide custom  operator overloading but still being able to access the class members(which is not possible as it stands):

class A
{
    cstruct sInfo
    {
        void print() { writeln(Name); }
    }

    string Name;
    sInfo Info;

    void print() { writeln("It's howdy doody time!"); }
}


This not also provides better encapsulation it also avoids collisions. The above case is easily fixed but it is impossible to do for build in overloads since we can't specify the type we want to overload(it works on the class, not a member of a class):

e.g., the above can be written as
class A
{

    string Name;
    void printInfo() { writeln(Name); }
    void print() { writeln("It's howdy doody time!"); }
}

If one wants(although, we lose the hierarchy potential).

But what about

class A
{
    cstruct B
    {
        opAssign(int) { writeln("Inside " ~ Name); ... }
    }
    string Name;
    B b;
    opAssign
}

Both a and b have their own opAssigns, b's opAssign can access A's members(without storing a ptr to an A). But we can't "flatten" the hierarchy like the first case:

class A
{


    string Name;
    int b;
    opAssign
    opAssign(int) { writeln("Inside " ~ Name); ... } // works on A, not int b!
}

One can say that all we need to do to get such functionality is store a ptr of type A and set it in the structs. This is a huge waste of memory though and not necessary since we can just pass a "this"(cthis) ptr to the methods inside.

Obviously such structs can only be used in fields inside the class, not arbitrarily, and so are not quite the same thing as pass by value structs(which, when copied, would lose their relationship to their parent class). We would not allow a struct to be copied because then it would be disassociated from it's container object which would invalidate cthis.

(although we could convert it to a normal struct, one that doesn't use methods involving cthis and it should work just like any other struct... or "lift" the structs to include a this ptr inside that is used(but possibly create unintended consequences))

Any ideas if this is achievable or anything fundamentally wrong with it? My purpose for it is to provide additional functionality to fields inside classes but not have to include a this ptr in them for no reason. It is achievable using offsets and pointers but requires a lot of manual work that I feel the compiler could do easily.










December 11, 2012
Greetings

I too find this very useful. I am also having this situation often when there is an array of nested struct instances inside a class environment. This is a common scenario and deserves attention. Also I believe (looking at D1 forums) that there was a time in the past when D1 nested structs did support 'outer'.

But unfortunately structs are neglected part of D, so I am not sure if developers give an ear to this thread. To make D truly a systems language we need to make structs more robust and feature rich.

Regards
- Puneet


December 12, 2012
On Tuesday, 11 December 2012 at 03:28:00 UTC, d coder wrote:
> Greetings
>
> I too find this very useful. I am also having this situation often when
> there is an array of nested struct instances inside a class environment.
> This is a common scenario and deserves attention. Also I believe (looking
> at D1 forums) that there was a time in the past when D1 nested structs did
> support 'outer'.
>
> But unfortunately structs are neglected part of D, so I am not sure if
> developers give an ear to this thread. To make D truly a systems language
> we need to make structs more robust and feature rich.
>
> Regards
> - Puneet

If I want to manually pass the parent pointer to a child struct, how do I do it? The compiler won't allow me to use "this" from the parent until after the parent has been instantiated.

The solution I came up with is to create a default constructor for the parent that sets the child struct with the parents pointer. It works but maybe there's a better way?

BTW, the same situation happens with structs defined inside structs, they will not have access to the parent struct members unless you manually pass a pointer into the child struct at some point after the parent struct's address becomes known. You cannot create a default constructor for a struct, so reliably initializing the child struct with the parents pointer is problematic.

--rt

December 12, 2012
On Wednesday, 12 December 2012 at 01:36:55 UTC, Rob T wrote:
> On Tuesday, 11 December 2012 at 03:28:00 UTC, d coder wrote:
>> Greetings
>>
>> I too find this very useful. I am also having this situation often when
>> there is an array of nested struct instances inside a class environment.
>> This is a common scenario and deserves attention. Also I believe (looking
>> at D1 forums) that there was a time in the past when D1 nested structs did
>> support 'outer'.
>>
>> But unfortunately structs are neglected part of D, so I am not sure if
>> developers give an ear to this thread. To make D truly a systems language
>> we need to make structs more robust and feature rich.
>>
>> Regards
>> - Puneet
>
> If I want to manually pass the parent pointer to a child struct, how do I do it? The compiler won't allow me to use "this" from the parent until after the parent has been instantiated.
>
> The solution I came up with is to create a default constructor for the parent that sets the child struct with the parents pointer. It works but maybe there's a better way?
>
> BTW, the same situation happens with structs defined inside structs, they will not have access to the parent struct members unless you manually pass a pointer into the child struct at some point after the parent struct's address becomes known. You cannot create a default constructor for a struct, so reliably initializing the child struct with the parents pointer is problematic.
>
> --rt

The problem with this method is that it requires an extra ptr in the child to the parent class. This is not necessary and a waste of space. While it work just fine it's not "efficient". Also, the compiler should have no issue accessing parent fields because they are a simple compile time const difference.

In some sense, if ptr_s is the this variable for a struct. Then ptr_s + x would point to a field in the struct. x is known at compile time as it depends on the size of the types in the struct. A nested struct would simply use ptr_s - y + z to access the elements of the parent. ptr_s - y = ptr_p, which is the this ptr to the parent class.

Hence, the only thing that would change is the compiler would have to compute where the struct is located in the class, then subtract that value from the struct's this ptr to get the this ptr of the parent object containing that struct object.

I wrote a post here that does something similar but ultimately the compiler can do it all for us and would not waste any space or extra cycles.

http://forum.dlang.org/thread/dgleowuonachabwxmlrx@forum.dlang.org

The main thing to realize is that the offsets are all known at compile time(obviously) and the relationship between a nested struct object and it's parent is also known at compile time. By having to "hard code" stuff to get it to work though, makes it difficult to manage.