https://kotlinlang.org logo
Join Slack
Powered by
I just discovered that `Iterable<T>.map` ret...
# announcements
d
I just discovered that 
Iterable<T>.map
returns a 
List<R>
not the original data structure (whatever implemented the 
Iterable
. This seems very strange to me, is it expected? I would expect the 
map
function to affect only the contents of the data structure, not the data structure itself.
n
Well, an iterable doesn't have to actually come from a data structure
so I don't see how that could work, even in principle
v
Iterable<T>.map
could return 
Iterable<R>
But actually it does that as 
List<R>
implements 
Iterable<R>
🙂
d
I’m not sure how to write the code, not there in Kotlin yet, but Iteralble is an interface, can’t I return whatever type implemented the interface?
n
not necessarily, no
an iterable can be lazy for example
d
why does that matter?
n
well, at that point map would be behaving in a very different way
to discuss that we'd probably need to discuss why Kotlin has both Iterable<T> and Sequence<T>
t
someMap.map{ it.toString()}
... hmm?
👍 1
n
FYI, if you're coming from most other languages, what's surprising about 
map
in Kotlin isn't that it doesn't return the same data structure
it's that 
map
isn't just returning a lazy iterator, that you later decide to collect in a data structure
this is probably the most common approach
v
Also imagine a 
MyDatabaseResult
that lazily gives you the results from the database when you query them. How should the result of 
map
create an instance of that class to satisfy your requirement?
d
I don’t understand, isn’t Iterable just something that can (possibly) return a next value? Surely I could build a 
map
function on that that works lazily or strictly, depending on the iterator?
n
so, inKotlin, functions taking lambdas (like map) can be inline, or they can be lazy
they cannot be both
d
Iterable
interface have only one method: 
fun iterator(): Iterator<T>
. There is no information how to construct original type on which 
map
was called.
n
and inline is desirable in this context. So, Kotlin gives you both options. Iterable<T>.map is inline, and isn't lazy, it always processes the iterable immediately. So it has ot put it somewhere. So it needs a data structure.
Sequence<T>.map is not inline, but it is lazy
d
Copy code
class MySuperIterable : Iterable<String> {
   ...
}

fun test(it: MySuperIterable) {
    val result = it.map { it.length }
}
In this example there is no way to create 
MySuperIterable
which holds 
Int
d
ahhh, @dmitriy.novozhilov I now see the issue with the way it works
so Iterable is the wrong thing, 
map
would need it’s own interface
v
o_O
n
not really, no
d
well that’s a shame, it was quite unexpected to me that 
map
would change the data structure
n
do you have a specific example?
I guess you wanted to call map on a Map or something?
d
yes, I have a 
Page<T>
type which wraps up a list with some other info
n
okay, and Page<T> implements Iterable<T> I suppose?
Or even implements List<T>?
d
well I just added Collection in order to do this, I thought it would work without investigating enough
n
So, what you'd typically do is to do the map via sequences instead of iterables
myPage.asSequence().map { ... }.toPage();
v
Again, how should 
map
know how to instantiate a 
Page<R>
? And more so if the type bounds for 
T
maybe don't even allow 
R
.
d
the problem is that there isn’t a 
toPage
function because there is extra data in the Page object (page number, page size etc)
n
if you find that verbose, then you could write your own 
mapPage
or something like that. You could also choose to have Page<T> implement Sequence<T> rather than Iterable<T>
d
so I’m looking for a way to map over the values inside without changing anything else
n
Right, that's a good point. but yeah, it would also be unreasonable to think that a generic 
map
could magically know how to do that
d
I wrote 
mapPage
but that was a bit of a hack, I thought there would be some interface that I could implement
n
Probably the best solution is to write a function called 
transform
, which takes a lambda which accepts a List<T> and returns a List<T>
or accepts a sequence<T> and returns a sequence<T>
d
yeah, I’m coming from a different world where you have a 
map
interface
i.e. functor
n
that would allow you to use the full standard library collection of functions on this
Yeah, I guess you're coming from Scala or Haskell or something
d
Haskell/Purescript
n
Nothing is stopping you from implementing map yourself, just to be clear
d
yeah that’s what I’m doing
n
from my limited knowledge of haskell I'm not sure this concepts really carries over 1-1
The thing is though if you implement 
map
itself, then you'll have the same issue again with say 
filter
that's why I suggested 
transform
d
well, mapping over the contents of a container
or mapping a function over the contents of a container
anyway, no big deal
n
Yes, I understand, but like I said, you'll need to repeat this for every useful utility function in the stdlib
that you want to use
But, perhaps you only really see yourself using map and prefer the simplicity/convenience of targeting this one use case
it's just fairly common to also use things like filter, mapNotNull, etc
mapIndexed, etc
d
yeah, that’s the real shame, with a functor you can implement a lot of other functions
e
so this is something that Scala can do with its collection builders, and you can sort of manually do in Kotlin
n
yeah, but that's easily avoided, if you implement something like transform instead
e
e.g. 
.mapTo(builder) { transform(it) }
n
Copy code
Page<T>.transform(f: (List<T>) -> List<T>): Page<T>
that's trivial to implement just once, and now you'll be able to do whatever you want inside the transform
Copy code
newPage = oldPage.transform { oldList ->
    oldList.map { .... }.filter { ... } etc etc 
}
d
in the end I just created
Copy code
fun <A, B> Page<A>.map(f: (A) -> B): Page<B> {
    val bs = this.values.map(f)
    return Page.slice(bs, this.totalElements, Pageable(this.pageNumber, this.pageSize))
}
filter
is a bit of a weird function compared to map because by it’s nature it changes the data structure
@Nir this transform function is a bit dodgy because we don’t really want to change the size of the list underneath without changing 
this.pageSize
etc
thanks for all your help anyway!
n
Well, transform could of course handle that, since you could ask for the size of the List. better, this.pageSize should probably just be implemented as a property that returns the size of the List anyway
since it saves storage, and better guarantees correctness
np
r
Not sure if it's been mentioned, but if you're coming from a functional world, look into #C5UPMM0A0.
d
yeah, I took a look, I’m just trying not to push too much of that type of thing into my workplace (yet…). I was just hoping that some stuff would exist more natively (i.e. without a compiler plugin) and so far I have found lots of great things, map was not one of them though :)
I sort of took type classes for granted until I started using kotlin and found that it’s very difficult to implement a lot of nice things without them because you need to attach generic stuff to existing classes
n
Yeah, it's problematic that you cannot implement a trait for a class you don't control
d
exactly
n
it's probably my single biggest issue with kotlin tbh, it's terribly constraining
I come from C++ which is basically the wild west, okay, the generics are totally unconstrained, but at least that means I can implement generic things for anything I want, and have things work.
d
lol, an honest way to describe C++ 🙂
n
I tried to write a class to hold a json data structure in Kotlin, and it was just impossible to make it nice
in C++, yes it's a bit of extra work to have a recursive data structure, but you can do it very nicely
indeed 🙂 we know what we are.
t
n
not really looking for a json library, i was just commenting that Kotlin's type system puts fundamental limitaitons on some of the things you can do
writing an in-memory json representation is a way to see it
6 Views
Open in Slack
PreviousNext
Powered by