Hacker News Reader

Comment by notb | Hacker News Reader

Seems like a disadvantage to me. It's mostly just confusing and the only thing it affords is the ability to use :keywords as functions (:like so) which is just sugar for property access, so["like"]. This is broken, though.

The coolest thing about :keywords in Clojure is that they really are functions and you can do things like (map :keywords ontoSomething) to extract the same property out of many things. It don't work in Wisp.

View on HN

What you're not realizing, though, is that each time you use the string "like" as a key for a map/dictionary, you're allocating a new object on the heap. So what if you have many keys, and they're all strings? What if you have many maps that share the same keys? That's a lot of memory to allocate, and strings are an extremely inefficient way to implement what is essentially a unique identifier to address map values. Instead, use an integer as a unique identifier.

That's basically what keywords do -- they are human-readable and compile down (prob. using some hash) into some integer, and the integer is stored globally only once. The biggest savings in memory comes when using maps, and that's how & why keywords get their name.

Javascript strings are immutable and literals are almost certainly preallocated and stored in a table of constants. Why should using a string keys cause heap allocations?

If you have a VM where strings are immutable and interned, you're right. Java doesn't guarantee that strings are interned (though you can ask the VM to intern it), so Clojure treats them differently.

Also, Java declared java.lang.String as final, so Strings can't be extended to implement clojure interfaces to get nice behavior, like IFn and IKeywordLookup.

The coolest thing about :keywords in Clojure ... don't work in Wisp.

Ok. That sucks. So pretty much anyone already a bit serious about Clojure wont be enticed by this then.

Edit: From the page:

    ;;    Keywords can be invoked as functions, that desugars to plain
    ;;    associated value access in JS
    (:bar foo) ;; => foo["bar"]

That seems pretty close to what's expected though, doesn't it?

I'd expect `(:bar foo)` to desugar to `foo["bar"]()` since the keyword is being used as a function. Property access in clojurescript uses the ugly `(.-bar foo)` syntax.

No, to get foo["bar"]() in CLJS you write ((:bar foo)). :bar in this case is a function, which returns a value by a key ":bar" from "foo", it's not a property.

Reply on news.ycombinator.com