Better callback system about rhea

[hfossli]

I tried to make this change

And then use it like this

containerHeight.on_change = [self](const rhea::variable &v) {
    std::cout << "containerHeight changed to: " << v.value() << std::endl;
};

But it seems like that this is a different instance.. Look at the pointer address in highlighted in blue (they are pointing at the correct value).

So I guess it is not straight forward...

[Nocte-]

You're right, that would get unwieldy after a while. You were on the right track, but the callbacks should be handled in abstract_variable (variable is not much more than a convenience wrapper around a smart pointer).

[hfossli]

Okay. I will have a go at that. 👍

[hfossli]

Aaaaand here's the result.

https://github.com/hfossli/rhea/compare/callbacks

I have no clue how I can make the API better. First time I'm coding c++ ever so this is quite the challenge. Ideally I would like something like

x.on_change([](float old_value, float new_value) {
    // code
});

Instead of

x.on_change([]() {
    // code
});

[Nocte-]

Looking at the code, I wonder why there are separate calls for 'set' and 'change'. I'd like to try and simplify that before adding the callback.

[hfossli]

I agree. 👍

[hfossli]

I'm not entirely sure, but I think the callback should be called when adding a variable to the solver as well. So that I from a user perspective can link the current value to properties on my views.

Maybe on_change should rather be apply or something like that? What do you think?

[Nocte-]

I think I have found a good way to extend the API, while you still only pay for what you use. You can either link an existing variable to a rhea::variable:

float kitty = 41.0f;
rhea::variable x (kitty, rhea::linked());

If you link to an integer, the library does the rounding for you.
Or use a callback if you need one:

rhea::variable x ([&](double v){ aargh(v); }, 41.0);

[hfossli]

Hm. Seems interesting, I don't understand the mechanisms though. :-) Can you explain a little bit more? Will the value of x change whenever the float kitty changes?

[Nocte-]

Yes, but don't do that. ;) It's meant to write the result of the solution back to kitty. It is basically a more efficient shortcut for:

rhea::variable x ([&](double v){ kitty = static_cast<float>(v); }, 41.0);

[hfossli]

Ah, I see :) Looks nice

[hfossli]

It would be awesome to be able to have n subscribers and not just one

[Nocte-]

If you know your subscribers ahead of time, that wouldn't be too difficult:

variable x ([&](double v){ some_var = other_var = yet_another = v; });
variable y ([&](double v){ call_me(v); call_me_on_the_line(); call_me_anytime(v); });

But if you need a more elaborate subscriber mechanism, I'd suggest picking a library that suits your needs (Boost signal2, Sigslot, FastDelegate, ...) and writing a small helper class to connect the variable to the signal.

I have the feeling there might be an easier way though. :) Could I take a look at your project?

[hfossli]

Well, for me it is sufficient that I initialize the variable with a callback.

Rhea variables are instantiated just by declaring them. How can I declare an instance variable and do the initialization myself?

@implementation BoxModel  {
    rhea::variable left;
}

- (id)init {
    self = [super init];
    if(self) {
        // I could write
        left = rhea::variable([self](double newValue) {
             [self leftChangedTo:newValue];
        });

       // But then a rhea::variable is allocated 1 time excessively, right?
    }
    return self;
}

@end

[hfossli]

I would be happy to let you have a look at some time, but not right now :-) Still a couple of proof of concepts left

[Nocte-]

Yeah indeed, 'left' starts as a normal variable and gets swapped out for a callback variable later. This is probably the best solution; the overhead is negligible (unless you're instancing BoxModel so often it starts showing up on the profiler) and 'left' is always in a well-defined state.

[hfossli]

I agree!

Will the memory be okay? The old variable will get released automatically? (I'm not very familiar with std and c++)

[Nocte-]

Yep!

[hfossli]

👍