Thread overview
Member delegate/fp to another member in same object?
May 02, 2017
Juanjo Alvarez
May 02, 2017
Juanjo Alvarez
May 02, 2017
Adam D. Ruppe
May 02, 2017
Hi!

I would like to have a "proxy" delegate, let's call it "foo" that could point to a method or another, let's call them "fast_foo" or "slow_foo" on the same object. This way depending on some conditions I could switch at runtime from one set of methods to others with a "switchFoo(" fast") method while the rest of the code inside the object would continue happily calling "foo".

The problem I have is that since the delegate keep the state of the object at the moment of the assignment I can't use it for this is pretty since the "fast/slow" methods need the value of the members of the enclosing object members at the point of usage,  I guess that I could pass the "this" pointer in the rebinding method but since the targets are also methods of the same object I wouldn't be surprised to find that there is an obvious/elegant  way to get access to "this" on the delegate-pointed method implementing.

Example:

struct S {
  int someState;
  void some_foo() { return this. someState;}

  void delegate() foo;

  void enable() {
    foo = &some_foo;
  }
}

unittest {
  S s;
  s.someState = 1;
  enable();
  s.someState = 2;
  assert(foo == 2);
  // fails because the delegate keeps
  // the state of the object at the
  // assignment point
}


May 02, 2017
On Tuesday, 2 May 2017 at 17:08:11 UTC, Juanjo Alvarez wrote:
> Example:
>
> struct S {
>   int someState;
>   void some_foo() { return this. someState;}
>
>   void delegate() foo;
>
>   void enable() {
>     foo = &some_foo;
>   }
> }
>
> unittest {
>   S s;
>   s.someState = 1;
>   enable();
>   s.someState = 2;
>   assert(foo == 2);
>   // fails because the delegate keeps
>   // the state of the object at the
>   // assignment point
> }

Forget it. I just noticed the simplified example that I just posted works (once the typo of the return value is corrected) but my more complex real code won't, will try to get a simple snippet where I can reproduce the problem.
May 02, 2017
On Tuesday, 2 May 2017 at 17:08:11 UTC, Juanjo Alvarez wrote:
> struct S {
>   int someState;
>   void some_foo() { return this. someState;}
>
>   void delegate() foo;
>
>   void enable() {
>     foo = &some_foo;
>   }
> }

That's actually illegal in D. It will compile, but has undefined behavior because the compiler is free to move the struct around without giving you a chance to update the delegate. You are liable for random crashes doing that.

You'd be better off using a function pointer instead of a delegate and making the user pass `this` to it explicitly, or making it a class rather than a struct, which the compiler will not move. (or a struct only ever used by pointer, a diy class basically)