Hey! I know Kotlin has lazy properties with its de...

# getting-started

[bazbo.222]

Hey! I know Kotlin has lazy properties with its delegates feature. Is there any way to add lazy evaluation (or, equivalently for my use case, memoization) such that the first time I call foo(arg) , foo(arg) gets calculated, but the second, third etc. times that I call foo after that, foo(arg) just returns the already computed, stored value.

E.g. in Python, this exists in the form of the @Memoize operator, if anyone is familiar with that.

Is there any way to achieve this in Kotlin or should I just implement the memoization mechanism myself.

[Joffrey]

There is no built-in feature for this. However, you could roll your own higher-order function to support it yourself. Something like cached(key, args) { ... }

[Joffrey]

Doing this requires asking yourself lots of questions, though. Like how to synchronize things between threads, how to clean up the cache, etc.

[Joffrey]

Different use cases call for different solutions

[Youssef Shoaib [MOD]]

Arrow has MemoizedDeepRecursiveFunction I believe, but yes you can just as well do your own higher-order function

[bazbo.222]

Thanks for the replies! I've heard of Arrow from a thread talking about how its Either<U,V> behaves a lot like Rust's Result<T,E> enum or like TypeScript's union types; so Youssef, you've made me all the more curious about checking it out. Thanks, guys!

[Joffrey]

@Youssef Shoaib [MOD] how does arrow's cache eviction work? Does everything stay in memory forever? Are calls to this function thread-safe?

[Youssef Shoaib [MOD]]

There's some basic built-in options, but also an integration module with cache-4k, which has all the cool cache eviction stuff you want. See https://arrow-kt.io/learn/collections-functions/recursive/

[joseph_ivie]

Short sample of doing it yourself (which is honestly short enough to perhaps not worry about a shortcut):

private val results = HashMap<Int, Int>()
fun memoizedFib(number: Int): Int = if(number <= 1) number else results.getOrPut(number) {
    memoizedFib(number - 1) + memoizedFib(number - 2)
}

memoizedFib(11)

Note this will have concurrency issues unless you use ConcurrentHashMap instead.

[bazbo.222]

Hey, @joseph_ivie! Yeah, I've tried out the getOrPut way of doing things. It's fun :))

[bazbo.222]

Question for the people in this thread: say I want to cache a function foo(bar, baz, qux) of three (or more) parameters. What's the most idiomatic way of doing it? I've perused [this StackOverflow answer](https://stackoverflow.com/questions/822322/how-to-implement-a-map-with-multiple-keys), and the idea of a MultiKeyMap seems enticing. I've tried just using a plain MutableMap<Pair<K1, K2>,V>, but maybe due to how pairs are hashed under the hood or something, it didn't have the expected / desired behaviour for my use case.

So, would you implement n maps for a function that takes n arguments? Or is there a smarter approach to the underlying data structure here?

[Joffrey]

Contrary to the SO question you linked, you want the tuple to act as a single key, you don't want multiple independent keys. So the solution to have several non-nested maps will not work for you, but the single map with a pair (or list) as key should work.

Could you please describe what you observed when you tried it?

[bazbo.222]

Oh okay, maybe I did something wrong then. Let me double check and I'll get back to you with my issue if it persists.