More natural derived properties

[joshaber]

I've found when doing reactive UI stuff, I end up using -rac_deriveProperty:from: a lot. It's awesome for the most part, but it doesn't read very naturally. What I really want to write and read is something like:

self.titleLabel.textColor = someSubscribable;

I've come up with two different ways to write something more like that, both with tradeoffs, so I'd like to get some feedback.

Keyed subscripting:

rac(self.titleLabel.textColor) = someSubscribable;

That abuses keyed subscripting. It works for both object and non-object properties. The downside is that the class has to implement -setObject:forKeyedSubscript:. We can create a macro that does it for them, but it's an extra step users have to remember per-class.

Proxy bounce

self.titleLabel.rac.textColor = (id) someSubscribable;

// ooooh also:
[self.createButton.rac setTitleColor:(id) someSubscribable forState:UIControlStateNormal];

This uses a proxy to bounce message sends through. It's kinda awesome because you can use it with methods that aren't just plain setters, which happens to be the case a fair amount in UIKit land. But this approach has two downsides: (1) you have to cast the subscribable, (2) it won't work with non-object properties.

So I'd be curious to see if anyone has any other ideas or feels strongly about either of those two options.

[jwilling]

The proxy looks fantastic. If I had a choice between the two, I'd rather go with the proxy and suffer the loss of not being able to use it with non-object properties.

However, would it be impractical to offer both? Subscripting primarily for scalars, and proxies for objects?

[jspahrsummers]

I like the proxy much less now that I realize subscribables would always have to be cast. I think I would honestly rather see them wrapped in a dummy object made specifically for this purpose, like:

self.titleLabel.rac.textColor = someSubscribable.propertySomethingOrOther;

Where the subscribable's property would simply return id.

Can go you into more detail about how exactly the keyed subscripting works?

[joshaber]

@jspahrsummers Yeah I agree the cast is weak sauce :(

The keyed subscripting works something like:

#define RAC(keypath) self[RAC_KEYPATH_SELF(keypath)]

- (void)setObject:(id)obj forKeyedSubscript:(id<NSCopying>)key {
    [self rac_deriveProperty:(NSString *) key from:obj];
}

RAC(self.blah.thing) = subscribable;

The really annoying bit is that the compiler has to be able to see the definition of -setObject:forKeyedSubscript: otherwise it won't compile. So we can't just inject it at runtime.

[jspahrsummers]

@joshaber So that wouldn't work for objects that aren't relative to self? I don't know – the keyed subscripting seems pretty picky, and probably not worth it.

[joshaber]

@jspahrsummers Well, you could imagine a version that does:

#define RAC_OBJ(obj, keypath) obj[RAC_KEYPATH(obj, keypath)]

But that object would need to to implement -setObject:forKeyedSubscript:.

Honestly, I don't really care much about that case. The case I'm specifically thinking of is UI where there are a lot of values derived from a few basic values, most/all of which are on the class itself.

[andymatuschak]

Wait, couldn't you make the proxy-bounce situation way nicer by defining rac to return instancetype? Then you wouldn't have to cast. Also, you could totally make it work with non-object properties.

[andymatuschak]

Ah, wait, totally misunderstood the problem; nevermind. I've got nothing.

[joshaber]

Actually I think there's probably a way to make the keyed subscripting version more generic. I'll give a try.

The proxy bounce is really cool since it lets you pass in a subscribable in the place of a normal object but maybe there's a better / more specific way to surface that.

[jspahrsummers]

Also, what happens if you proxy bounce on a RACSubscribable property?

Does it have to be a subscribable of subscribables? How would it know?

[joshaber]

That linked PR adds the subscripting-abusive assignment. It's a little more abusive but it works without needing the class to implement -setObject:forKeyedSubscript:.

@jspahrsummers I'm pretty sure the universe implodes. Pretty sure.