I'm working on a game written in Kotlin and was looking into improving GC churn. One of the major sources of churn are for-loops called in the main game/rendering loops that result in the allocation of iterators.
Turning to the documentation, I found this paragraph:
A for loop over an array is compiled to an index-based loop that does not create an iterator object.
If you want to iterate through an array or a list with an index, you can do it this way:
for (i in array.indices)
print(array[i])
Note that this “iteration through a range” is compiled down to optimal implementation with no extra objects created.
https://kotlinlang.org/docs/reference/control-flow.html#for-loops
Is this really true? To verify, I took this simple Kotlin program and inspected the generated byte code:
fun main(args: Array<String>) {
val arr = arrayOf(1, 2, 3)
for (i in arr.indices) {
println(arr[i])
}
}
According to the quote above, this should not result in any objects allocated, but get compiled down to a good old pre-Java-5 style for-loop. However, what I got was this:
41: aload_1
42: checkcast #23 // class "[Ljava/lang/Object;"
45: invokestatic #31 // Method kotlin/collections/ArraysKt.getIndices:([Ljava/lang/Object;)Lkotlin/ranges/IntRange;
48: dup
49: invokevirtual #37 // Method kotlin/ranges/IntRange.getFirst:()I
52: istore_2
53: invokevirtual #40 // Method kotlin/ranges/IntRange.getLast:()I
56: istore_3
57: iload_2
58: iload_3
59: if_icmpgt 93
This looks to me as if a method called getIndices is called that allocates a temporary IntRange object to back up bounds checking in this loop. How is this an "optimal implementation" with "no extra objects created", or am I missing something?
UPDATE:
So, after toying around a bit more and looking at the answers, the following appears to be true for Kotlin 1.0.2:
Arrays:
for (i in array.indices): range allocation
for (i in 0..array.size): no allocation
for (el in array): no allocation
array.forEach: no allocation
Collections:
for (i in coll.indices) range allocation
for (i in 0..coll.size): no allocation
for (el in coll): iterator allocation
coll.forEach: iterator allocation
To iterate an array without allocating extra objects you can use one of the following ways.
for-loop
for (e in arr) {
println(e)
}
forEach extension
arr.forEach {
println(it)
}
forEachIndexed extension, if you need to know index of each element
arr.forEachIndexed { index, e ->
println("$e at $index")
}
As far as I know the only allocation-less way to define a for loop is
for (i in 0..count - 1)
All other forms lead to either a Range allocation or an Iterator allocation. Unfortunately, you cannot even define an effective reverse for loop.
Here is an example of preparing a list and iterate with index and value.
val list = arrayListOf("1", "11", "111")
for ((index, value) in list.withIndex()) {
println("$index: $value")
}
Output:
0:1
1:11
2:111
Also, following code works similar,
val simplearray = arrayOf(1, 2, 3, 4, 5)
for ((index, value) in simplearray.withIndex()) {
println("$index: $value")
}
Related
How to "take(N)" iteratively - get a Sequence<Sequence>, each inner sequences having next N elements?
I am writing a high-load application in Kotlin.
I have tens of thousands of entries to insert to a database.
I want to batch them by, say, 1000.
So I created a loop:
val itemsSeq = itemsList.iterator().asSequence()
while (true) {
log.debug("Taking $BATCH_SIZE from $itemsSeq")
val batchSeq = itemsSeq.take(BATCH_SIZE)
val squareBatch = applySomething(batchSeq, something)
?: break
}
fun applySomething(batch: Sequence<Item>, something: Something) {
/* Fully consumes batch. Bulk-loads from DB by IDs, applies, bulk-saves. */
}
I thought that take() would advance the itemsSeq and the next call to take() would give me a sequence "view" of itemsSeq starting at the 10th item.
But with this code, I am getting:
DEBUG Taking 10 from kotlin.sequences.ConstrainedOnceSequence#53fe15ff
Exception in thread "main" java.lang.IllegalStateException: This sequence can be consumed only once.
at kotlin.sequences.ConstrainedOnceSequence.iterator(SequencesJVM.kt:23)
at kotlin.sequences.TakeSequence$iterator$1.<init>(Sequences.kt:411)
at kotlin.sequences.TakeSequence.iterator(Sequences.kt:409)
So it seems that the take() "opens" the itemsSeq again, while that can be consumed only once.
As a workaround, I can use chunked():
public fun <T> Sequence<T>.chunked(size: Int): Sequence<List<T>> {
But I would prefer not to create Lists, rather Sequences.
What I am looking for is something between take() and chunked().
Is there anything such in Kotlin SDK?
I can possibly create my own sequence { ... } but for readability, I would prefer something built-in.
There is a way to construct a Sequence by handing it over an Iterator, see Sequence.
Given an iterator function constructs a Sequence that returns values
through the Iterator provided by that function. The values are
evaluated lazily, and the sequence is potentially infinite.
Wrapped in an extension function it could look like this:
fun <T> Iterable<T>.toValuesThroughIteratorSequence(): Sequence<T> {
val iterator = this.iterator()
return Sequence { iterator }
}
Quick test:
data class Test(val id: Int)
val itemsList = List(45) { Test(it) }
val batchSize = 10
val repetitions = itemsList.size.div(batchSize) + 1
val itemsSeq = itemsList.toValuesThroughIteratorSequence()
(0 until repetitions).forEach { index ->
val batchSeq = itemsSeq.take(batchSize)
println("Batch no. $index: " + batchSeq.map { it.id.toString().padStart(2, ' ') }.joinToString(" "))
}
Output:
Batch no. 0: 0 1 2 3 4 5 6 7 8 9
Batch no. 1: 10 11 12 13 14 15 16 17 18 19
Batch no. 2: 20 21 22 23 24 25 26 27 28 29
Batch no. 3: 30 31 32 33 34 35 36 37 38 39
Batch no. 4: 40 41 42 43 44
Background
First of all, we need to be aware there is a big difference between an object that we can iterate over and object that represents a "live" or already running iteration process. First group means Iterable (so List, Set and all other collections), Array, Flow, etc. Second group is mostly Iterator or old Java Enumeration. The difference could be also compared to file vs file pointer when reading or database table vs database cursor.
Sequence belongs to the first group. Sequence object does not represent a live, already started iteration, but just a set of elements. These elements can be produced lazily, sequence could have unbounded size and usually internally it works by using iterators, but conceptually sequence is not an iterator itself.
If we look into the documentation about sequences it clearly compares them to Iterable, not to Iterator. All standard ways to construct sequences like: sequenceOf(), sequence {}, Iterable.asSequence() produce sequences that return the same list of items every time we iterate over them. Iterator.asSequence() also follows this pattern, but because it can't re-produce same items twice, it is intentionally protected against iterating multiple times:
public fun <T> Iterator<T>.asSequence(): Sequence<T> = Sequence { this }.constrainOnce()
Problem
Your initial attempt with using take() didn't work, because this is a misuse of sequences. We expect that subsequent take() calls on the same sequence object will produce exactly the same items (usually), not next items. Similarly as we expect multiple take() calls on a list always produce same items, each time starting from the beginning.
Being more specific, your error was caused by above constrainOnce(). When we invoke take() multiple times on a sequence, it has to restart from the beginning, but it can't do this if it was created from an iterator, so Iterator.asSequence() explicitly disallows this.
Simple solution
To fix the problem, you can just skip constrainOnce() part, as suggested by #lukas.j. This solution is nice, because stdlib already provides tools like Sequence.take(), so if used carefully, this is the easiest to implement and it just works.
However, I personally consider this a kind of workaround, because the resulting sequence doesn't behave as sequences do. It is more like an iterator on steroids than a real sequence. You need to be careful when using this sequence with existing operators or 3rd party code, because such sequence may work differently than they expect and as a result, you may get incorrect results.
Advanced solution
We can follow your initial attempt of using subsequent take() calls. In this case our object is used for live iteration, so it is no longer a proper sequence, but rather an iterator. The only thing we miss in stdlib is a way to create a sub-iterator with a single chunk. We can implement it by ourselves:
fun main() {
val list = (0 .. 25).toList()
val iter = list.iterator()
while (iter.hasNext()) {
val chunk = iter.limited(10)
println(chunk.asSequence().toList())
}
}
fun <T> Iterator<T>.limited(n: Int): Iterator<T> = object : Iterator<T> {
var left = n
val iterator = this#limited
override fun next(): T {
if (left == 0)
throw NoSuchElementException()
left--
return iterator.next()
}
override fun hasNext(): Boolean {
return left > 0 && iterator.hasNext()
}
}
I named it limited(), because take() suggests we read items from the iterator. Instead, we only create another iterator on top of the provided iterator.
Of course, sequences are easier to use than iterators and typical solution to this problem is by using chunked(). With above limited() it is pretty straightforward to implement chunkedAsSequences():
fun main() {
val list = (0 .. 25).toList()
list.asSequence()
.chunkedAsSequences(10)
.forEach { println(it.toList()) }
}
fun <T> Sequence<T>.chunkedAsSequences(size: Int): Sequence<Sequence<T>> = sequence {
val iter = iterator()
while (iter.hasNext()) {
val chunk = iter.limited(size)
yield(chunk.asSequence())
chunk.forEach {} // required if chunk was not fully consumed
}
}
Please also note there is a tricky case of chunk being not fully consumed. chunkedAsSequences() is protected against this scenario. Previous simpler solutions aren't.
Working on an Advent of Code puzzle I had found myself defining a function to transpose matrices of integers:
fun transpose(xs: Array<Array<Int>>): Array<Array<Int>> {
val cols = xs[0].size // 3
val rows = xs.size // 2
var ys = Array(cols) { Array(rows) { 0 } }
for (i in 0..rows - 1) {
for (j in 0..cols - 1)
ys[j][i] = xs[i][j]
}
return ys
}
Turns out that in the following puzzle I also needed to transpose a matrix, but it wasn't a matrix of Ints, so i tried to generalize. In Haskell I would have had something of type
transpose :: [[a]] -> [[a]]
and to replicate that in Kotlin I tried the following:
fun transpose(xs: Array<Array<Any>>): Array<Array<Any>> {
val cols = xs[0].size
val rows = xs.size
var ys = Array(cols) { Array(rows) { Any() } } // maybe this is the problem?
for (i in 0..rows - 1) {
for (j in 0..cols - 1)
ys[j][i] = xs[i][j]
}
return ys
}
This seems ok but it isn't. In fact, when I try calling it on the original matrix of integers I get Type mismatch: inferred type is Array<Array<Int>> but Array<Array<Any>> was expected.
The thing is, I don't really understand this error message: I thought Any was a supertype of anything else?
Googling around I thought I understood that I should use some sort of type constraint syntax (sorry, not sure it's called like that in Kotlin), thus changing the type to fun <T: Any> transpose(xs: Array<Array<T>>): Array<Array<T>>, but then at the return line I get Type mismatch: inferred type is Array<Array<Any>> but Array<Array<T>> was expected
So my question is, how do I write a transpose matrix that works on any 2-dimensional array?
As you pointed out yourself, the line Array(cols) { Array(rows) { Any() } } creates an Array<Array<Any>>, so if you use it in your generic function, you won't be able to return it when Array<Array<T>> is expected.
Instead, you should make use of this lambda to directly provide the correct value for the correct index (instead of initializing to arbitrary values and replacing all of them):
inline fun <reified T> transpose(xs: Array<Array<T>>): Array<Array<T>> {
val cols = xs[0].size
val rows = xs.size
return Array(cols) { j ->
Array(rows) { i ->
xs[i][j]
}
}
}
I don't really understand this error message: I thought Any was a supertype of anything else?
This is because arrays in Kotlin are invariant in their element type. If you don't know about generic variance, it's about describing how the hierarchy of a generic type compares to the hierarchy of their type arguments.
For example, assume you have a type Foo<T>. Now, the fact that Int is a subtype of Any doesn't necessarily imply that Foo<Int> is a subtype of Foo<Any>. You can look up the jargon, but essentially you have 3 possibilities here:
We say that Foo is covariant in its type argument T if Foo<Int> is a subtype of Foo<Any> (Foo types "vary the same way" as T)
We say that Foo is contravariant in its type argument T if Foo<Int> is a supertype of Foo<Any> (Foo types "vary the opposite way" compared to T)
We say that Foo is invariant in its type argument T if none of the above can be said
Arrays in Kotlin are invariant. Kotlin's read-only List, however, is covariant in the type of its elements. This is why it's ok to assign a List<Int> to a variable of type List<Any> in Kotlin.
Context: I've been benchmarking the difference between using invokedynamic and manually generating bytecode (this is in the context of deciding whether a compiler targeting the JVM should emit more verbose "traditional" bytecode or just an invokedynamic call with a clever bootstrap method). In doing this, it has been pretty straightforward to map bytecode into MethodHandles combinators that are at least as fast, with the exception of tableswitch.
Question: Is there a trick to mimic tableswitch using MethodHandle? I tried mimicking it with a jump table: using a constant MethodHandle[], indexing into that with arrayElementGetter, then calling the found handle with MethodHandles.invoker. However, that ended up being around 50% slower than the original bytecode when I ran it through JMH.
Here's the code for producing the method handle:
private static MethodHandle makeProductElement(Class<?> receiverClass, List<MethodHandle> getters) {
MethodHandle[] boxedGetters = getters
.stream()
.map(getter -> getter.asType(getter.type().changeReturnType(java.lang.Object.class)))
.toArray(MethodHandle[]::new);
MethodHandle getGetter = MethodHandles // (I)H
.arrayElementGetter(MethodHandle[].class)
.bindTo(boxedGetters);
MethodHandle invokeGetter = MethodHandles.permuteArguments( // (RH)O
MethodHandles.invoker(MethodType.methodType(java.lang.Object.class, receiverClass)),
MethodType.methodType(java.lang.Object.class, receiverClass, MethodHandle.class),
1,
0
);
return MethodHandles.filterArguments(invokeGetter, 1, getGetter);
}
Here's the initial bytecode (which I'm trying to replace with one invokedynamic call)
public java.lang.Object productElement(int);
descriptor: (I)Ljava/lang/Object;
flags: (0x0001) ACC_PUBLIC
Code:
stack=3, locals=3, args_size=2
0: iload_1
1: istore_2
2: iload_2
3: tableswitch { // 0 to 2
0: 28
1: 38
2: 45
default: 55
}
28: aload_0
29: invokevirtual #62 // Method i:()I
32: invokestatic #81 // Method java/lang/Integer.valueOf:(I)Ljava/lang/Integer;
35: goto 67
38: aload_0
39: invokevirtual #65 // Method s:()Ljava/lang/String;
42: goto 67
45: aload_0
46: invokevirtual #68 // Method l:()J
49: invokestatic #85 // Method java/lang/Long.valueOf:(J)Ljava/lang/Long;
52: goto 67
55: new #87 // class java/lang/IndexOutOfBoundsException
58: dup
59: iload_1
60: invokestatic #93 // Method java/lang/Integer.toString:(I)Ljava/lang/String;
63: invokespecial #96 // Method java/lang/IndexOutOfBoundsException."<init>":(Ljava/lang/String;)V
66: athrow
67: areturn
The good thing about invokedynamic is that it allows to postpone the decision, how to implement the operation to the actual runtime. This is the trick behind LambdaMetafactory or StringConcatFactory which may return composed method handles, like in your example code, or dynamically generated code, at the particular implementation’s discretion.
There’s even a combined approach possible, generate classes which you compose to an operation, e.g. settling on the already existing LambdaMetafactory:
private static MethodHandle makeProductElement(
MethodHandles.Lookup lookup, Class<?> receiverClass, List<MethodHandle> getters)
throws Throwable {
Function[] boxedGetters = new Function[getters.size()];
MethodType factory = MethodType.methodType(Function.class);
for(int ix = 0; ix < boxedGetters.length; ix++) {
MethodHandle mh = getters.get(ix);
MethodType actual = mh.type().wrap(), generic = actual.erase();
boxedGetters[ix] = (Function)LambdaMetafactory.metafactory(lookup,
"apply", factory, generic, mh, actual).getTarget().invokeExact();
}
Object switcher = new Object() {
final Object get(Object receiver, int index) {
return boxedGetters[index].apply(receiver);
}
};
return lookup.bind(switcher, "get",
MethodType.methodType(Object.class, Object.class, int.class))
.asType(MethodType.methodType(Object.class, receiverClass, int.class));
}
This uses the LambdaMetafactory to generate a Function instance for each getter, similar to equivalent method references. Then, an actual class calling the right Function’s apply method is instantiated and a method handle to its get method returned.
This is a similar composition as your method handles, but with the reference implementation, no handles but fully materialized classes are used. I’d expect the composed handles and this approach to converge to the same performance for a very large number of invocations, but the materialized classes having a headstart for a medium number of invocations.
I added a first parameter MethodHandles.Lookup lookup which should be the lookup object received by the bootstrap method for the invokedynamic instruction. If used that way, the generated functions can access all methods the same way as the code containing the invokedynamic instruction, including private methods of that class.
Alternatively, you can generate a class containing a real switch instruction yourself. Using the ASM library, it may look like:
private static MethodHandle makeProductElement(
MethodHandles.Lookup lookup, Class<?> receiverClass, List<MethodHandle> getters)
throws ReflectiveOperationException {
ClassWriter cw = new ClassWriter(ClassWriter.COMPUTE_FRAMES);
cw.visit(V1_8, ACC_INTERFACE|ACC_ABSTRACT,
lookup.lookupClass().getName().replace('.', '/')+"$Switch", null,
"java/lang/Object", null);
MethodType type = MethodType.methodType(Object.class, receiverClass, int.class);
MethodVisitor mv = cw.visitMethod(ACC_STATIC|ACC_PUBLIC, "get",
type.toMethodDescriptorString(), null, null);
mv.visitCode();
Label defaultCase = new Label();
Label[] cases = new Label[getters.size()];
for(int ix = 0; ix < cases.length; ix++) cases[ix] = new Label();
mv.visitVarInsn(ALOAD, 0);
mv.visitVarInsn(ILOAD, 1);
mv.visitTableSwitchInsn(0, cases.length - 1, defaultCase, cases);
String owner = receiverClass.getName().replace('.', '/');
for(int ix = 0; ix < cases.length; ix++) {
mv.visitLabel(cases[ix]);
MethodHandle mh = getters.get(ix);
mv.visitMethodInsn(INVOKEVIRTUAL, owner, lookup.revealDirect(mh).getName(),
mh.type().dropParameterTypes(0, 1).toMethodDescriptorString(), false);
if(mh.type().returnType().isPrimitive()) {
Class<?> boxed = mh.type().wrap().returnType();
MethodType box = MethodType.methodType(boxed, mh.type().returnType());
mv.visitMethodInsn(INVOKESTATIC, boxed.getName().replace('.', '/'),
"valueOf", box.toMethodDescriptorString(), false);
}
mv.visitInsn(ARETURN);
}
mv.visitLabel(defaultCase);
mv.visitTypeInsn(NEW, "java/lang/IndexOutOfBoundsException");
mv.visitInsn(DUP);
mv.visitVarInsn(ILOAD, 1);
mv.visitMethodInsn(INVOKESTATIC, "java/lang/String",
"valueOf", "(I)Ljava/lang/String;", false);
mv.visitMethodInsn(INVOKESPECIAL, "java/lang/IndexOutOfBoundsException",
"<init>", "(Ljava/lang/String;)V", false);
mv.visitInsn(ATHROW);
mv.visitMaxs(-1, -1);
mv.visitEnd();
cw.visitEnd();
lookup = lookup.defineHiddenClass(
cw.toByteArray(), true, MethodHandles.Lookup.ClassOption.NESTMATE);
return lookup.findStatic(lookup.lookupClass(), "get", type);
}
This generates a new class with a static method containing the tableswitch instruction and the invocations (as well as the boxing conversions we now have to do ourselves). Also, it has the necessary code to create and throw an exception for out-of-bounds values. After generating the class, it returns a handle to that static method.
I don't know of your timeline. But it is likely there will be a MethodHandles.tableSwitch operation in Java 17. It is currently being integrated via https://github.com/openjdk/jdk/pull/3401/
Some more discussion about it here:
https://mail.openjdk.java.net/pipermail/core-libs-dev/2021-April/076105.html
The things is, tableswitch isn't always compiled to a jump table. For a small number of labels, like in your example, it's likely to act as a binary search. Thus using a tree of regular "if-then" MethodHandles will be the closest equivalent.
I put together a dummy problem to illustrate my point: Say that we have the following handy function to display information about a particular sorting algorithm:
fun sort(name: String, array: Array<Int>, sortingAlgorithm: (Array<Int>) -> Array<Int>) {
println(name)
sortingAlgorithm(array).forEach { print(" $it ") }
println()
}
You would use it like this:
sort("Selection Sort - Θ(n^2)", arrayOf(2, 3, 1), ::selectionSort)
And this works because the signature of selectionSort is simple: fun selectionSort(array: Array<Int>): Array<Int> {
But say I have another sorting algorithm with the following signature
fun quickSort(array: Array<Int>,
start: Int = 0,
end: Int = array.size - 1): Array<Int> {
The last two arguments are optional, so in theory you could call quickSort the same way you call selectionSort. That is to say, it stil respects the signature (Array<Int>) -> Array<Int> Right?
Unfortunately when I try to call sort("Quick Sort", arrayOf(2, 3, 1), ::quickSort) I get:
I think that the compiler isn't being smart enough to realise that those two arguments are optional. How can avoid this problem, other than overloading the sort method to accept a high order function with the signature ?
There is no avoiding this problem since it would contradict 2 corner stones of Kotlin type system:
1) Every expression has a type (strong typing)
2) The receiver side does not affect an expression's type (local inference)
For example if you could do that, the following would not to work, which is a simple refactoring of your example:
val algorithm = ::quickSort
sort("Quick Sort", arrayOf(2, 3, 1), algorithm)
Anyways, the sort("Quick Sort", { quickSort(unsorted) }) workaround is too simple for Kotlin developers to spend spend time on the problem.
This may be a duplicate. I don't know. I couldn't understand the other answers well enough to know that. :)
Rust version: rustc 1.0.0-nightly (b47aebe3f 2015-02-26) (built 2015-02-27)
Basically, I'm passing a bool to this function that's supposed to build an iterator that filters one way for true and another way for false. Then it kind of craps itself because it doesn't know how to keep that boolean value handy, I guess. I don't know. There are actually multiple lifetime problems here, which is discouraging because this is a really common pattern for me, since I come from a .NET background.
fn main() {
for n in values(true) {
println!("{}", n);
}
}
fn values(even: bool) -> Box<Iterator<Item=usize>> {
Box::new([3usize, 4, 2, 1].iter()
.map(|n| n * 2)
.filter(|n| if even {
n % 2 == 0
} else {
true
}))
}
Is there a way to make this work?
You have two conflicting issues, so let break down a few representative pieces:
[3usize, 4, 2, 1].iter()
.map(|n| n * 2)
.filter(|n| n % 2 == 0))
Here, we create an array in the stack frame of the method, then get an iterator to it. Since we aren't allowed to consume the array, the iterator item is &usize. We then map from the &usize to a usize. Then we filter against a &usize - we aren't allowed to consume the filtered item, otherwise the iterator wouldn't have it to return!
The problem here is that we are ultimately rooted to the stack frame of the function. We can't return this iterator, because the array won't exist after the call returns!
To work around this for now, let's just make it static. Now we can focus on the issue with even.
filter takes a closure. Closures capture any variable used that isn't provided as an argument to the closure. By default, these variables are captured by reference. However, even is again a variable located on the stack frame. This time however, we can give it to the closure by using the move keyword. Here's everything put together:
fn main() {
for n in values(true) {
println!("{}", n);
}
}
static ITEMS: [usize; 4] = [3, 4, 2, 1];
fn values(even: bool) -> Box<Iterator<Item=usize>> {
Box::new(ITEMS.iter()
.map(|n| n * 2)
.filter(move |n| if even {
n % 2 == 0
} else {
true
}))
}