Hacker News Reader

Comment by ohyes | Hacker News Reader

I'm biased, but there's a big difference between "learning lisp" (well enough to transliterate that python program) and really getting the code as data concept.

I'd say that learning X really well is always a positive because whenever you learn something in depth you can then apply the concepts elsewhere. And there are still lisp-only concepts, so learning lisp is positive. ML is good too. Also spending time with a well designed concurrent language is beneficial.

View on HN

With Lisp I get the data thing but with ML you learn almost everything is a language.

I can't really explain it but the whole variant/ADT pattern matching really forces to make you think of your problem domain as a specification or language (e.g. DSL). It is one of the reason why I think so many compilers are written in ML (that and the toolset is awesome for it).

That description of ML reminds me what I really like about REBOL and RED as well.

With the “slight” difference that ML actually enforces that your program is consistent with your description of the problem domain.

Another important feature of ML is abstract data types (opaque signature ascription), which forces abstraction clients not to rely on implementation details. It's Dijkstra's “separation of concerns”, integrated into the type checking process! See: https://existentialtype.wordpress.com/2011/04/16/modules-mat...

Pattern matching is the feature i enjoy most in functional languages. It's simple to use and comes with great value.

Having looked up variants just now, I can definitely see how it would be useful. It seems (to me) to fall into the same category of macro-like things that enhance the expressiveness of the language. I think the most important point is the idea of lifting the programming language into the problem domain. When you can do this in a clean way, the result is a very readable solution that you can reason about.

I think there's a difference, though. With ML (IIUC), you build a language on top of ML that is a set of functions (and the data structures they operate on) that make it easy to express the program.

With Lisp macros, though, you change the syntax of Lisp. You take things that were not valid Lisp syntax, and you make them valid for your program.

So ML "language building" is in terms of semantics, but in Lisp, it's both semantics and syntax. (Again, IIUC).

I personally think the "code as data" concept is not so important in lisp. It is critical to writing macros, but you often don't write macros and beginners to lisp often abuse macros.

I've written Clojure professionally now for a couple years and am pretty fast and proficient at it, but when I write a function call, I don't think of it as a list, even though that is what it is. I don't think any lisper does. It's just a function with arguments and I'm writing in the syntax necessary, which happens to be a list.

In those rare occasions where I think a macro is necessary, sure, it helps that a function call is a list, but that's hardly the main role in a day's work.

With respect to Clojure, you don't write macros, but many of the really interesting language features within Clojure are built upon macros. (core.async and core.match come to mind). It would be difficult to add them without macros.

I do actually think of my code as a tree/graph structure. Not just in lisps but in other languages as well, I'm always manipulating a stack of trees and thinking about how the data flows through the branches.

> many of the really interesting language features within Clojure are built upon macros

That's true with any lisp. But that doesn't mean the the ability to write macros is a requirement for most development work on a daily basis, and homoiconicity is often touted as amazing specifically because of its help in writing macros. Therefore, since writing macros is not super common or necessary most of the time, I would say that "code as data" is also not a big deal most of the time.

I really like lisp, I just don't think some of its unique features are as big a deal as they advertise.

code as data, however, does become an important concept in becoming really good at Perl or Python (and I would assume Ruby etc), and understanding common pitfalls here also helps one spot, for example, LSP problems and provide better solutions (because most LSP problems can be solved by replacing a mutator with something that returns a new object).

In Javascript, also, effectively you have to work with code as data.

So this isn't just about writing macros. There are a lot of areas where this does touch things.

Reply on news.ycombinator.com