Why Y? Deriving the Y Combinator in JavaScript

http://raganwald.com/2018/09/10/why-y.html

by braythwayt • 8 years ago

200 68 8 years ago

sjroot• 8 years ago

Joking aside, I thought this was a very approachable article. Thank you for sharing.

cakeface• 8 years ago

It's nice to see an article like this every once and a while to remind myself that there is so much about programming still to learn. I particularly liked the javascript examples so I could play with them in my scratch pad. I am still not sure I could describe what a Y Combinator is and where to use it, but I'm much closer than I was 15 minutes ago.

arethuza2• 8 years ago

It's over 30 years since I was taught about the Y combinator and I still find it amazing - it allows you to define recursion in a language which has no concept of recursion or named functions - such as λ-calculus.

The fact that you then define this in terms of amazingly simple functions like the S and K combinators just, in my opinion, adds to how wonderful it is.

guelo• 8 years ago

In those 30 years did you ever encounter the need to use these combinators?

wcarss• 8 years ago

I slightly wish the exponentiation function from the original article weren't the more complex "faster" version -- it's not really relevant to the article, and it makes playing through the calls in your head just a little bit harder with a topic that already stretches your mind the first several times around. Regular exponentiation is a fine example without the added fun of working out why and how exactly this alternative form works.

That said, as always raganwald's stuff is a pleasant and informative read.

braythwayt14• 8 years ago

Good feedback, thank you.

paulddraper2• 8 years ago

I disagree. It nice to have a non-FizzBuzz problem. I.e. code that you could actually ship.

anonytrary• 8 years ago

I disagree. If you are writing shippable code in an article, it's because the article is about how to write shippable code. This article is not about that, so writing shippable code risks detracting from the main purpose of the article.

mFixman• 8 years ago

There's a nice Haskell library for dealing with combinators, where the types of most functions are intuitive enough to understand what they do: http://hackage.haskell.org/package/data-aviary-0.4.0/docs/Da...

The Y combinator isn't there and can't be defined trivially since Haskell only allows non-recursive data types. However, there's a nice solution using non-monotonic data types[1]:

    newtype Mu a = Mu (Mu a -> a)
    y f = (\h -> h $ Mu h) (\x -> f . (\(Mu g) -> g) x $ x)

[1] https://stackoverflow.com/a/5885270/305597

danjoc3• 8 years ago

So we start out with an attempt to build an anonymous recursive function. That's cool I suppose. But doing it in a language like javascript (or java)? It turns out that tail recursion == stack overflow. Not daunted, the plucky author decides that we can solve this particular issue by creating a closure over stack variables and put them on the heap instead!

You know, if you want to use Haskell, use Haskell. Trying to shoehorn this stuff into languages designed for loops instead of recursion just means impressionable JS developers are going to see this and rush out to implement it in some project to prove how smart they are. Then everyone suffers. The implementation is slower because it goes to heap. The memory usage explodes because it goes to heap.

Purely for illustration of a Y combinator, this is great. But there should be a big red warning somewhere on the page as a reminder of the downsides of this approach.

empath75• 8 years ago

You don't even use the y combinator in haskell. It's an academic exercise that's mostly relevant in lambda calculus.

braythwayt14• 8 years ago

We actually don’t start out with an attempt to build anonymous recursion.

That is the goal of the Y Combinator in the Lambda Calculus and Combinatory Logic, but here the initial goal was to decouple an anonymous function from itself so that we could decorate it.

danjoc3• 8 years ago

Going out to main memory is orders of magnitude slower than hitting cache. Even a stack overflow would be preferable to debug vs an out of memory error. At least with the stack overflow, the stack trace pinpoints the problem.

Like I said, I have no problem with this article as an illustration of a ycombinator for JS devs. The problem is it does not warn them sufficiently of the very major drawbacks in using it. This is why we have Gate's law; Developers using inefficient constructs because they're neato.

maze-le• 8 years ago

Nice. Very thorough article, that touches more concepts of functional programming than just Y like tail recursion.

For someone who never had contact with functional programming, the concept of the Y combinator can seem a bit... strange. But trying to figure out what it is and does can be a real eye opener.

9dev• 8 years ago

I always thought windows was the only exe opener

Edit: ah, you fixed the typo ;)

golergka3• 8 years ago

While I enjoyed the subject matter, I can't help but feel that Y Combinator is to functional programming as AbstractFactoryFactory is to object-oriented programming. My inner geek revels at these abstractions, but my inner engineer and team manager wants something simpler and more robust.

jacquesm• 8 years ago

Even simpler?

golergka3• 8 years ago

Simpler in understanding and mental overhead. I want code in my projects being written in a way so that it's maintainable by developers this business can reasonably hire. Which means that they didn't read SICP or even Knuth, don't have experience with functional language and overall are not the x10 developers everybody pretends to have.

dualscyther• 8 years ago

But it's not really meant to be used as a pattern for writing actual code in an actual programming language. It exists as a way to implement recursion when we can't refer to named functions.

Edit: unless you are specifically talking about the use case of this article.

pwm• 8 years ago

The difference is that the Y combinator has a deep meaning in theoretical CS, making recursion essentially out of thin air in the untyped lambda calculus which had far reaching consequences. In contrast an AbstractFactoryFactory is pretty mundane.

epx• 8 years ago

Y cromossome :P

janci• 8 years ago

This feels like Alice in Wonderland

braythwayt14• 8 years ago

Author here. I take that as a compliment.

Not everything is meant to be deployed to production.

DonaldPShimoda• 8 years ago

Are you suggesting that I shouldn't be writing production code in the lambda calculus?

...hm.

nikodunk• 8 years ago

This article sounds like poetry. Beautifully written!

seanmcdirmid• 8 years ago

Is there something like the Y combinator for self types in F bounded OO generic type systems? I just realized it’s the same kind of problem at the type level. E.g.

abstract class FooImpl<Self extends FooImpl<Self>> { Self doIt() { .... } }

As a class with a self type, analogous to an open recursive function, and then:

class FooFinal extends FooImpl<FooFinal> {...}

As it’s call.

abetusk• 8 years ago

For those of us who don't know what a Y-Combinator is or why it's useful, I found a SO answer [1] that makes sense to me:

    A Y-combinator is a "functional" (a function that
    operates on other functions) that enables recursion,
    when you can't refer to the function from within
    itself. In computer-science theory, it generalizes
    recursion, abstracting its implementation, and
    thereby separating it from the actual work of the
    function in question. The benefit of not needing a
    compile-time name for the recursive function is
    sort of a bonus. =)

[1] https://stackoverflow.com/questions/93526/what-is-a-y-combin...

amelius2• 8 years ago

Wondering if anybody here has ever used the Y-combinator in practice (and not as an academic exercise) ...

(though I get that it can be a useful intermediate construct inside a compiler)

billsix2• 8 years ago

It's not useful in practice. But defining recursion via a function is a good mental exercise.

I'm unaware of compilers using the Y combinator. Perhaps you were thinking of continuation passing style

viksit• 8 years ago

I’ve used it in a semi professional way (internal tooling in clojure). I’ve also written about it here (Practical applications of the ycombinator in clojure) [1].

[1] http://www.viksit.com/tags/clojure/practical-applications-y-...

flatline• 8 years ago

I’ve used it for recursion in anonymous functions in C#. It is straightforward enough to implement. Was it the best choice? Probably not, but I took the opportunity when presented.

mehrdadn• 8 years ago

Nope, but I felt good when I finally managed to write one from scratch, because I knew for sure there was a time when I couldn't. So maybe the same would happen for you too.

scrame• 8 years ago

There was a good practical ruby example for auto-vivification of hashes using the applicative ycombinator many years back, but I'm having trouble finding it.

kadoban• 8 years ago

I've use something very similar in practice. Haskell's fixed point operator is:

  fix f = let x = f x in x

This allows you to turn a non-recursive function into a recursive one, it factors the recursion out. It turns out that you can write a similar combinator (usually called mfix or memofix as I recall) that will result in a memoized version of the same recursive function.

So it allows you to abstract away the memoization, write it once and then use it on any recursive function. That way you can write quite natural recursive functions and just memoize them in one wrapper call.

https://hackage.haskell.org/package/memoize-0.8.1/docs/Data-... looks similar to what I ended up with

I used this in competitive programming for a while, it was nice for certain types of dynamic programming problems where the interesting part is the recursive definition, so once I found that there wasn't any more to write. Quite satisfying when it applied.

KayEss• 8 years ago

Useful when you have a lambda that needs to call itself, but the language doesn't bind the name until after the lambda is parsed (so you can't call by name). This happens in a few languages I think, C++ and C# for sure.

dionidium• 8 years ago

This SO answer is posted every time this topic comes up, so I've seen it many times (and, in fact, I provided one of the answers there, too) and it's still the one and only place I've ever seen "functional" used as a noun like that. I'm always tempted to edit it, but maybe I'm missing something?

dataflow• 8 years ago

It's correct as a noun. I believe it's taken from math: https://en.wikipedia.org/wiki/Functional_(mathematics)

The CS term is "higher-order function": https://en.wikipedia.org/wiki/Higher-order_function

markc• 8 years ago

I find Matt Might's post on this topic (Y in JavaScript) the best I seen (even though it's at least 8 years old by now): http://matt.might.net/articles/implementation-of-recursive-f...

It's clear, minimalist, insightful, and even has an in-page live code demo. Bonus points for including fixed-points as a key concept, and memoization as an extremely useful technique.

braythwayt14• 8 years ago

I've always liked that post, and I think it inspired me to do a similar thing.

That being said, I'm a big fan of decorators, so I expressed the memoization using a decorator, rather than baking it into Ymem.

On the other hand, trampolining is the perfect application for a special-purpose Y combinator, and thus the decoupled Trampoline.

skybrian• 8 years ago

It's interesting to see how the Y combinator is derived, but it seems like it's being oversold here?

I like the mockingbird better since it's less magical. Passing "self" as the first argument seems idiomatic from an object-oriented point of view; this is sort like how Python has an explicit "self" parameter. We can take the next step and pass "self" explicitly as well.

The ability to change how a self-call is handled seems quite like overriding a method.

yiyus• 8 years ago

AFAIU the problem, from a theoretical point of view, is not that you need to pass the self function, but how to define it. In pure lambda calculus, there are no named functions, just lambdas, so you actually need the Y combinator for recursion.

oneweekwonder• 8 years ago

So anybody up for a short snippet or explanation on how to use a @decorator_function to implement a Y Combinator in Python?

zelon88• 8 years ago

I would argue this is Google's fault.

I tested the theory myself. I went to Boston recently and followed my GPS. It took me over the Tobin bridge through solid bumper-to-bumper gridlock. When I arrived at the gridlock my GPS said 30 minutes left. 45 minutes later I had travelled about .5 miles and it still said 25 minutes left. Google was completely wrong, but I swear it had me right where it wanted me the entire time.

After getting fed up with traffic, I eventually turned my GPS off and got into the right lane to head towards Cambridge. My logic was that GPS could navigate from anywhere, to anywhere. I would simply go somewhere with less traffic and try the GPS again.

Sure enough I made it to Edward Land blvd and turned my GPS back on. I had a clear 15 minute drive with average traffic for the rest of my trip.

Google saw the traffic I was dealing with, and queued me right in behind it all. It could have recalculated around it, but it legitimately "thought" I only had 30 minutes to go. I guarantee it told all 500 other cars in front of me the same thing.

So we were all following our GPS's, and our GPS's were taking all of us the fastest route. It created a traffic jam and then optimistically pretended there was no traffic jam.

If street lights were truly "smart", with cameras and machine learning to maximize throughput, and GPS's programmed to load balance their own congestion, this problem wouldn't be a problem anymore.

braythwayt14• 8 years ago

FridgeSeal• 8 years ago

I think you commented on the wrong article lol.

arkades• 8 years ago

Yet the topmost response.

adiusmus• 8 years ago

Given I’ve never had reason to use a y-combinator, mocking bird or eta or any of the other colourful terms from the article this post about GPS seems vaguely appropriate.

Which means that in around 14 days I’ll suddenly need to revisit this article and learn all about it because it’s suddenly useful.

Life is strange like that. I do like the term “why bird” though.

curryhoward• 8 years ago

This is actually the Z combinator, which is the eta expansion of the Y combinator. The Y combinator only works in languages with non-strict semantics, but JavaScript is call-by-value.

braythwayt14• 8 years ago

This is included as a footnote in TFA.

gfaure• 8 years ago

Excellent article, Reginald.

I followed it entirely with one exception: I can see that `maker(maker)` has the right type, but why is it the right value? How do we know, as we fill in the blanks, that this is the implementation that yields the right result?

braythwayt14• 8 years ago

This is the best comment!

I wondered this myself. I mean, I can test it and see that it appears to give the correct result, but how do I prove that it gives the correct result?

I didn't include it, but you can do reductions, successively replacing the names with their expansions, and you see that it works out was we want.

And in Combinatory Logic, there really is nothing else. If an expression has the correct "type," then it works. There's nothing except rearranging terms, duplicating terms, and erasing terms.

But that being said, I don't know if that's a "proof," and I especially don't claim that a proof about Combinatory Logic would say anything at all about JavaScript.

Why Y? Deriving the Y Combinator in JavaScript

Related Stories