Hacker News Reader

Comment by cies | Hacker News Reader

I encountered HM a lot in my working life as a programmer. I did not expect that when I first encountered the names.

From the article:

> HM type systems can't implement heterogenous lists [...] > [well it does when wrapping the elements in a sum type, but then] > it can only accept types you know about in advance. > [which makes it] a massive pain to work with.

Sorry, I totally disagree... Heterogeneous lists without the sum types are a massive pain! Maybe it is easier for the initial development effort, but if you need to maintain code with heterogeneous lists (or dynamically typed code in general) the pain will come at some point and likely when already in production.

I much rather fix compile errors than runtime errors. And a massive thanks to Hindley and Milner for advancing type systems!

Found their photos in gratitude just now:

https://www.researchgate.net/profile/J-Hindley

https://en.wikipedia.org/wiki/Robin_Milner

View on HN

They're only a pain if you need to know the run-time types of the elements. If you're just operating on them through a narrowly defined interface, that's just dynamic polymorphism, which is indeed more flexible with regards to future expandability, as long as you define the interface appropriately.

> as long as you define the interface appropriately

Well, if you have an interface that all entries adhere to it is not a heterogeneous list in my book. In Haskell that could be achieve without a sum type wrapper, but with a type class.

https://fluoridealert.org/key-topics/tsca-fluoride-lawsuit/

I mean, a heterogeneous list is literally a list where its elements may be of different types. To use a tired example, if a list that contains both a Cat and a Dog is not heterogeneous just because both adhere to the Mammal interface, then what is a heterogeneous list?

I think my position is:

Native heterogeneous lists will often be convenient, and sometimes when you've used them they're going to turn out in retrospect to have been a bad idea but sometimes they'll be just fine, especially if they stay fairly locally scoped. I haven't done any serious work in a language that supports them for years, so I don't have a strong opinion about how often each of those cases happens.

But if you decide you want heterogeneous lists, but your language doesn't support them, so you try to implement that with a sum type, then that's basically always going to be a massive pain.

> I haven't done any serious work in a language that supports them for years

Is this not silently admitting that strong types to some degree have won?

Don't get me wrong: I like Clojure/LISPs/Ruby. But I would not choose them for a new project these days.

(and I do not like JS, which has them too)

I'm still working at the same company as when I wrote this, and that company is still using Haskell (now mostly Typescript instead of Elm). If I did move on I'd ideally want to keep using Haskell, or failing that some other strongly typed language.

But I don't expect that my own experience says much about the language ecosystem in general. I don't particularly have an opinion on whether or not strong types have "won", and I didn't intend to comment on that question.

I would definitely choose a Lisp if it had an Intellij-like IDE. Especially since the type system of CL is good, though not static, obviously. But it's a tradeoff between having Haskell during compile time and CL during development time, for me.

What's an example use case where heterogeneous lists turn out to be fine but can't be modelled with eg the OCaml difflist technique, eg https://github.com/yawaramin/re-web/blob/0d6c62fb432f85cc437... ?

I'm not familiar with that technique and don't know what's going on from that snippet.

In Haskell, any time a heterogeneous list turns out to be fine, I expect to be able to model it. Often it'll look like "I'm applying a polymorphic function to every one of these list elements", and then you can either do a sum type (as discussed in the post) or an existential (which doesn't need you to list up front all the types you might use). If the function is "is negative?", it'll look something like (untested)

    data SomeNum = forall a. Num a => SomeNum a

    isNegative :: SomeNum -> Bool
    isNegative (SomeNum n) = n < 0

    numbers = [SomeNum (3 :: Int), SomeNum (5.2 :: Double), SomeNum valOfUnexpectedNumericType]
    anyNegative = any isNegative numbers

...but it'll often be easier to just apply the `< 0` check to every element before putting them in the list. (If you have several functions you want to apply to them all, that's when the existential trick becomes more useful.)

So you can model heterogeneous lists in this case, and it's safer (because you can't put in a value that can't be compared to 0) but also less convenient. Whether that's an improvement or not will depend on the situation.

I also just learned about it. I found this explanation of what it is: https://drup.github.io/2016/08/02/difflists/

It's a slightly confusing name, though; it makes me think of a difference list, which seems to be a completely unrelated data structure (basically a rope).

http://h2.jaguarpaw.co.uk/posts/demystifying-dlist/

Thanks. Yeah, I don't understand the name.

You could construct a basically-equivalent data structure in Haskell, but I think normally you'd use an HList (defined in many places but e.g. https://hackage.haskell.org/package/HList-0.5.2.0/docs/Data-...). I've only occasionally had use for them myself, at any rate I don't think they're convenient for the "apply a function to all these elements" case.

In the technique I showed, I am using a difflist to enumerate a set of HTTP POST form fields and their expected types. This difflist type is defined in such a way that one of its type parameters gets inferred as a function type which takes the decoded values in the correct order and returns a record containing the decoded form. Eg from Field.[int "id"; string "name"] we get a type parameter int -> string -> 'a, where 'a is the type of the final decoded form.

This is the kind of real-world usage where difflists or heterogeneous lists shine. The same technique is used by this library to define type-safe GraphQL resolvers: https://github.com/andreas/ocaml-graphql-server

Reply on news.ycombinator.com