Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Defunctionalization: Everybody Does It, Nobody Talks About It

blog.sigplan.org

Post reply on HN
Print view View original

11–20 of 23 posts

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#11

Post by The_rationalist »

duelingjello wrote: post #8
mikelyons wrote: post #2

> defunctionalization, a transformation that replaces higher-order functions with equivalent first-order ones. For anyone else who got nothing from the title.

I admit to fawning for functional languages like Haskell, Clojure and OCaml for fun, but it seems like they are not more popular because having many additional, higher-order constructs makes code “obsfucated” behind clever features rather than written as simply as possible (but no simpler). Maybe for a coding shop full of insane geniuses (i.e., quants, CalTech aerospace, physicists or Stanford research department) it…

Clearly, but a subset of functional features help to reduce cognitive load. And this subset is what modern languages target (see my comment below).

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#12

Post by pkolaczk »

The_rationalist wrote: post #10
Gravityloss wrote: post #9

Earlier quoted context omitted.

What are the ergonomics-first languages of the functional world?

I would say that rust and kotlin are good at providing most of the useful concepts from functional programming (higher order functions, common functors (map, filter, etc), tail call, pattern matching, destructuring, expressions, lazy evaluation, immutable collections, actor model, algebraic data types) Sadly they lack some features at the type system level (higher kinded types, intersection and union types) TypeScrip…

Looks like the language you wanted to mention is Scala. Pragmatic and has all the things you mentioned.

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#13

Post by kod »

The_rationalist wrote: post #10
Gravityloss wrote: post #9

Earlier quoted context omitted.

What are the ergonomics-first languages of the functional world?

I would say that rust and kotlin are good at providing most of the useful concepts from functional programming (higher order functions, common functors (map, filter, etc), tail call, pattern matching, destructuring, expressions, lazy evaluation, immutable collections, actor model, algebraic data types) Sadly they lack some features at the type system level (higher kinded types, intersection and union types) TypeScrip…

Kotlin doesn't have destructuring pattern matching.

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#14

Post by aseipp »

touisteur wrote: post #5

The authors work seems very interesting, and could maybe lead to the automatic generation of a language server with semantic query and refactoring capabilities, for any programming language. Or portable (in the 'language to language' sense) static analysis tools or interpreters or symbolic execution tools... Just wondering (out loud) how you formalize language semantics so they can be consumed... I know about grammar…

Most people formalize "kernel languages" that contain all the bare essentials of the language at hand. There are a variety of tools to do this; a lot of people just end up embedding their language semantics in Coq or something and writing proofs about the theories there. Alternatively, a lot of people will tend to just write literal interpreters for their kernel language in a language like Haskell or OCaml; interpreters can be seen as a lightweight machine-runnable definition. And they have the added bonus you can play with them!

Another popular tool I'm aware of for this is Ott[1], which lets you write a high-level semantics description, much like you'd write in a LaTeX paper. It can then compile it to Coq, LaTeX, etc for further work. Ott is probably a much better place to start without getting too far into the weeds, if you want to write real high-level sketches.

In general, I (unfortunately) don't think there's anything like a "library" of language semantics definitions that are easily reusable or anything like that. Modular definitions of semantics that can be reused like a library/API is basically a programming language abstraction problem, but I'm not sure what the current forefront of research on that is. In general, very few languages have fully specified machine-checked semantics, or even unproven ones.

[1]: https://www.cl.cam.ac.uk/~pes20/ott/

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#15

Post by AnimalMuppet »

zozbot234 wrote: post #7
skybrian wrote: post #6

> For instance, I teach programmers to tune out the noise of refactoring catalogs and learn how most familiar refactorings instead follow from a few primordial algebraic laws Why "instead?" Knowing more math is good, but I think the author is doing his students a disservice. You don't need to memorize refactoring catalogs, but if someone says "maybe you should use a visitor pattern here" it's helpful to look it up an…

They can learn that stuff after they've become familiar with the fundamentals. Visitor pattern is just a way to implement pattern matching starting from the primitives Java and the like provide.

That's like saying that people can learn arithmetic after they learn group theory, which they can learn after they learn category theory. No, learn the concrete first, then learn the abstractions, so that you have some concept of why the abstractions are relevant.

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#16

Post by twic »

Gravityloss wrote: post #9
duelingjello wrote: post #8

Earlier quoted context omitted.

I admit to fawning for functional languages like Haskell, Clojure and OCaml for fun, but it seems like they are not more popular because having many additional, higher-order constructs makes code “obsfucated” behind clever features rather than written as simply as possible (but no simpler). Maybe for a coding shop full of insane geniuses (i.e., quants, CalTech aerospace, physicists or Stanford research department) it…

What are the ergonomics-first languages of the functional world?

LISPers would say LISP.

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#17

Post by strbean »

AnimalMuppet wrote: post #15
zozbot234 wrote: post #7

Earlier quoted context omitted.

They can learn that stuff after they've become familiar with the fundamentals. Visitor pattern is just a way to implement pattern matching starting from the primitives Java and the like provide.

That's like saying that people can learn arithmetic after they learn group theory, which they can learn after they learn category theory. No, learn the concrete first, then learn the abstractions, so that you have some concept of why the abstractions are relevant.

I'm not sure we do kids a service by skipping over fundamental aspects of mathematics in order to stick to the 'more concrete' stuff. The staggeringly vast majority of kids will never learn a lick of group theory or category theory, and I think that is unfortunate because the fundamentals aren't difficult or big. Most high school grads will hardly have touched even basic logic, and it is frighteningly common to run into people who don't understand that "A -> B; B; Therefore A" isn't valid.

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#18

Post by AnimalMuppet »

strbean wrote: post #17
AnimalMuppet wrote: post #15

Earlier quoted context omitted.

That's like saying that people can learn arithmetic after they learn group theory, which they can learn after they learn category theory. No, learn the concrete first, then learn the abstractions, so that you have some concept of why the abstractions are relevant.

I'm not sure we do kids a service by skipping over fundamental aspects of mathematics in order to stick to the 'more concrete' stuff. The staggeringly vast majority of kids will never learn a lick of group theory or category theory, and I think that is unfortunate because the fundamentals aren't difficult or big. Most high school grads will hardly have touched even basic logic, and it is frighteningly common to run i…

We would do a worse disservice by skipping over more concrete stuff to teach group theory or, worse, category theory.

First, even if you can teach group theory or category theory to an 8-year-old, if you wind up with an 8-year-old who knows category theory but can't add, is that a win? No, it isn't. They need to be able to add in a way that they don't need group theory.

But, second, you can't in any meaningful way teach category theory, or even group theory, to an 8-year-old. But you can teach them to add.

So again, I say, concrete first, then abstractions. First because they can function better knowing one concrete form and no abstract than they can knowing the abstract and zero concrete forms. And second, they are able to learn the most-useful concrete form younger and with less background than they are able to learn the abstraction.

Re: Defunctionalization: Everybody Does It, Nobody Talks About It

#20

Post by kazinator »

Gravityloss wrote: post #9
duelingjello wrote: post #8

Earlier quoted context omitted.

I admit to fawning for functional languages like Haskell, Clojure and OCaml for fun, but it seems like they are not more popular because having many additional, higher-order constructs makes code “obsfucated” behind clever features rather than written as simply as possible (but no simpler). Maybe for a coding shop full of insane geniuses (i.e., quants, CalTech aerospace, physicists or Stanford research department) it…

What are the ergonomics-first languages of the functional world?

TXR Lisp, for one.
Post reply on HN