Swift memoizing/caching lazy variable in a struct - swift

I drank the struct/value koolaid in Swift. And now I have an interesting problem I don't know how to solve. I have a struct which is a container, e.g.
struct Foo {
var bars:[Bar]
}
As I make edits to this, I create copies so that I can keep an undo stack. So far so good. Just like the good tutorials showed. There are some derived attributes that I use with this guy though:
struct Foo {
var bars:[Bar]
var derivedValue:Int {
...
}
}
In recent profiling, I noticed a) that the computation to compute derivedValue is kind of expensive/redundant b) not always necessary to compute in a variety of use cases.
In my classic OOP way, I would make this a memoizing/lazy variable. Basically, have it be nil until called upon, compute it once and store it, and return said result on future calls. Since I'm following a "make copies to edit" pattern, the invariant wouldn't be broken.
But I can't figure out how to apply this pattern if it is struct. I can do this:
struct Foo {
var bars:[Bar]
lazy var derivedValue:Int = self.computeDerivation()
}
which works, until the struct references that value itself, e.g.
struct Foo {
var bars:[Bar]
lazy var derivedValue:Int = self.computeDerivation()
fun anotherDerivedComputation() {
return self.derivedValue / 2
}
}
At this point, the compiler complains because anotherDerivedComputation is causing a change to the receiver and therefore needs to be marked mutating. That just feels wrong to make an accessor be marked mutating. But for grins, I try it, but that creates a new raft of problems. Now anywhere where I have an expression like
XCTAssertEqaul(foo.anotherDerivedComputation(), 20)
the compiler complains because a parameter is implicitly a non mutating let value, not a var.
Is there a pattern I'm missing for having a struct with a deferred/lazy/cached member?

Memoization doesn't happen inside the struct. The way to memoize is to store a dictionary off in some separate space. The key is whatever goes into deriving the value and the value is the value, calculated once. You could make it a static of the struct type, just as a way of namespacing it.
struct S {
static var memo = [Int:Int]()
var i : Int
var square : Int {
if let result = S.memo[i] {return result}
print("calculating")
let newresult = i*i // pretend that's expensive
S.memo[i] = newresult
return newresult
}
}
var s = S(i:2)
s.square // calculating
s = S(i:2)
s.square // [nothing]
s = S(i:3)
s.square // calculating

The only way I know to make this work is to wrap the lazy member in a class. That way, the struct containing the reference to the object can remain immutable while the object itself can be mutated.
I wrote a blog post about this topic a few years ago: Lazy Properties in Structs. It goes into a lot more detail on the specifics and suggest two different approaches for the design of the wrapper class, depending on whether the lazy member needs instance information from the struct to compute the cached value or not.

I generalized the problem to a simpler one: An x,y Point struct, that wants to lazily compute/cache the value for r(adius). I went with the ref wrapper around a block closure and came up with the following. I call it a "Once" block.
import Foundation
class Once<Input,Output> {
let block:(Input)->Output
private var cache:Output? = nil
init(_ block:#escaping (Input)->Output) {
self.block = block
}
func once(_ input:Input) -> Output {
if self.cache == nil {
self.cache = self.block(input)
}
return self.cache!
}
}
struct Point {
let x:Float
let y:Float
private let rOnce:Once<Point,Float> = Once {myself in myself.computeRadius()}
init(x:Float, y:Float) {
self.x = x
self.y = y
}
var r:Float {
return self.rOnce.once(self)
}
func computeRadius() -> Float {
return sqrtf((self.x * self.x) + (self.y * self.y))
}
}
let p = Point(x: 30, y: 40)
print("p.r \(p.r)")
I made the choice to have the OnceBlock take an input, because otherwise initializing it as a function that has a reference to self is a pain because self doesn't exist yet at initialization, so it was easier to just defer that linkage to the cache/call site (the var r:Float)

Related

Why does this compile without mutating keyword?

I feel like something is broken with the value semantics here. Consider:
struct A {
var b = 0
mutating func changeB() {
b = 6
}
}
struct B {
var arr = [A]()
func changeArr() {
/* as expected this won't compile
unlss changeArr() is mutating:
arr.append(A())
*/
// but this compiles! despite that changeB is mutating!
for var a in arr {
a.changeB()
}
}
}
Why can this example mutate the struct contents without marking the function as mutating? In true value semantics, any time you change any part of the value, the whole value should be considered changed, and this is usually the behavior in Swift, but in this example it is not. Further, adding a didSet observer to var arr reveals that changeArr is not considered mutation of the value.
for var a in arr {
a.changeB()
}
This is copying an element from arr out into a and leaving arr unchanged.
If you directly access the elements inside arr via their indexes, then it will mutate and require the mutating keyword.
The reason changeArr is not mutating is because it isn't really doing anything since it is working on local copies of the A objects. If you really want the method to do something meaningful it needs to be changed to
mutating func changeArrForReal() {
for index in arr.indices {
arr[index].changeB()
}
}

Can I set computed property in get block in Swift?

var myProperty: PropertyType {
get {
if let alreadyComupted = savedValue {
return alreadyComputed
}
return computeAndSave(someParam: "Hello")
}
set {
// is it possible to move *computeAndSave* body here somehow so I can call *set* method in above get definition?
}
}
private var savedValue: PropertyType?
private func computeAndSave(someParam: SomeType) -> PropertyType {
// perform computations and assign value to *savedValue*
}
I am fairly new to swift language, not sure if this is even standard by coding practice or not.
Basically you are describing a lazy variable. It calculates its initializer once, when the value is first fetched, and from then on uses the stored value (unless it is replaced). You can combine this with a define-and-call initializer:
lazy var myProperty: PropertyType = {
let p = // perform some expensive one-time calculation here
return p
}()
The outcome is that the first time you ask for the value of myProperty, the initializer method runs; but after that the previous result is used, and the initializer method never runs again.

how do people deal with iterating a Swift struct value-type property?

Here's an obvious situation that must arise all the time for people:
struct Foundation {
var columns : [Column] = [Column(), Column()]
}
struct Column : CustomStringConvertible {
var cards = [Card]()
var description : String {
return String(describing:self.cards)
}
}
struct Card {}
var f = Foundation()
for var c in f.columns {
c.cards.append(Card())
}
That code is legal but of course it has no effect on f, because var c is still a copy — the actual columns of f are unaffected.
I am not having any difficulties understanding why that happens. My question is what people generally do about it.
Clearly I can just make the whole matter go away by declaring Column a class instead of a struct, but is that what people usually do? (I'm trying to follow a mental stricture that one should avoid classes when there's no need for dynamic dispatch / polymorphism / subclassing; maybe I'm carrying that too far, or maybe there's something else people usually do, like using inout somehow.)
As of Swift 4, a compromise is to iterate over the indices of a mutable collection instead of the elements themselves, so that
for elem in mutableCollection {
// `elem` is immutable ...
}
or
for var elem in mutableCollection {
// `elem` is mutable, but a _copy_ of the collection element ...
}
becomes
for idx in mutableCollection.indices {
// mutate `mutableCollection[idx]` ...
}
In your example:
for idx in f.columns.indices {
f.columns[idx].cards.append(Card())
}
As #Hamish pointed out in the comments, a future version of Swift may implement a mutating iteration, making
for inout elem in mutableCollection {
// mutate `elem` ...
}
possible.

what is the difference between these two in swift 3 [duplicate]

In the Introduction to Swift WWDC session, a read-only property description is demonstrated:
class Vehicle {
var numberOfWheels = 0
var description: String {
return "\(numberOfWheels) wheels"
}
}
let vehicle = Vehicle()
println(vehicle.description)
Are there any implications to choosing the above approach over using a method instead:
class Vehicle {
var numberOfWheels = 0
func description() -> String {
return "\(numberOfWheels) wheels"
}
}
let vehicle = Vehicle()
println(vehicle.description())
It seems to me that the most obvious reasons for choosing a read-only computed property are:
Semantics - in this example it makes sense for description to be a property of the class, rather than an action it performs.
Brevity/Clarity - prevents the need to use empty parentheses when getting the value.
Clearly the above example is overly simple, but are there other good reasons to choose one over the other? For example, are there some features of functions or properties that would guide your decision of which to use?
N.B. At first glance this seems like quite a common OOP question, but I'm keen to know of any Swift-specific features that would guide best practice when using this language.
It seems to me that it's mostly a matter of style: I strongly prefer using properties for just that: properties; meaning simple values that you can get and/or set. I use functions (or methods) when actual work is being done. Maybe something has to be computed or read from disk or from a database: In this case I use a function, even when only a simple value is returned. That way I can easily see whether a call is cheap (properties) or possibly expensive (functions).
We will probably get more clarity when Apple publishes some Swift coding conventions.
Well, you can apply Kotlin 's advices https://kotlinlang.org/docs/reference/coding-conventions.html#functions-vs-properties.
In some cases functions with no arguments might be interchangeable
with read-only properties. Although the semantics are similar, there
are some stylistic conventions on when to prefer one to another.
Prefer a property over a function when the underlying algorithm:
does not throw
complexity is cheap to calculate (or caсhed
on the first run)
returns the same result over invocations
While a question of computed properties vs methods in general is hard and subjective, currently there is one important argument in the Swift's case for preferring methods over properties. You can use methods in Swift as pure functions which is not true for properties (as of Swift 2.0 beta). This makes methods much more powerful and useful since they can participate in functional composition.
func fflat<A, R>(f: (A) -> () -> (R)) -> (A) -> (R) {
return { f($0)() }
}
func fnot<A>(f: (A) -> Bool) -> (A) -> (Bool) {
return { !f($0) }
}
extension String {
func isEmptyAsFunc() -> Bool {
return isEmpty
}
}
let strings = ["Hello", "", "world"]
strings.filter(fnot(fflat(String.isEmptyAsFunc)))
There is a difference:
If you use a property you can then eventually override it and make it read/write in a subclass.
Since the runtime is the same, this question applies to Objective-C as well. I'd say, with properties you get
a possibility of adding a setter in a subclass, making the property readwrite
an ability to use KVO/didSet for change notifications
more generally, you can pass property to methods that expect key paths, e.g. fetch request sorting
As for something specific to Swift, the only example I have is that you can use #lazy for a property.
In the read-only case, a computed property should not be considered semantically equivalent to a method, even when they behave identically, because dropping the func declaration blurs the distinction between quantities that comprise the state of an instance and quantities that are merely functions of the state. You save typing () at the call site, but risk losing clarity in your code.
As a trivial example, consider the following vector type:
struct Vector {
let x, y: Double
func length() -> Double {
return sqrt(x*x + y*y)
}
}
By declaring the length as a method, it’s clear that it’s a function of the state, which depends only on x and y.
On the other hand, if you were to express length as a computed property
struct VectorWithLengthAsProperty {
let x, y: Double
var length: Double {
return sqrt(x*x + y*y)
}
}
then when you dot-tab-complete in your IDE on an instance of VectorWithLengthAsProperty, it would look as if x, y, length were properties on an equal footing, which is conceptually incorrect.
From the performance perspective, there seems no difference. As you can see in the benchmark result.
gist
main.swift code snippet:
import Foundation
class MyClass {
var prop: Int {
return 88
}
func foo() -> Int {
return 88
}
}
func test(times: u_long) {
func testProp(times: u_long) -> TimeInterval {
let myClass = MyClass()
let starting = Date()
for _ in 0...times {
_ = myClass.prop
}
let ending = Date()
return ending.timeIntervalSince(starting)
}
func testFunc(times: u_long) -> TimeInterval {
let myClass = MyClass()
let starting = Date()
for _ in 0...times {
_ = myClass.prop
}
let ending = Date()
return ending.timeIntervalSince(starting)
}
print("prop: \(testProp(times: times))")
print("func: \(testFunc(times: times))")
}
test(times: 100000)
test(times: 1000000)
test(times: 10000000)
test(times: 100000000)
Output:
prop: 0.0380070209503174
func: 0.0350250005722046
prop: 0.371925950050354
func: 0.363085985183716
prop: 3.4023300409317
func: 3.38373708724976
prop: 33.5842199325562
func: 34.8433820009232
Program ended with exit code: 0
In Chart:
There are situations where you would prefer computed property over normal functions. Such as: returning the full name of an person. You already know the first name and the last name. So really the fullName property is a property not a function. In this case, it is computed property (because you can't set the full name, you can just extract it using the firstname and the lastname)
class Person{
let firstName: String
let lastName: String
init(firstName: String, lastName: String){
self.firstName = firstName
self.lastName = lastName
}
var fullName :String{
return firstName+" "+lastName
}
}
let william = Person(firstName: "William", lastName: "Kinaan")
william.fullName //William Kinaan
Semantically speaking, computed properties should be tightly coupled with the intrinsic state of the object - if other properties don't change, then querying the computed property at different times should give the same output (comparable via == or ===) - similar to calling a pure function on that object.
Methods on the other hand come out of the box with the assumption that we might not always get the same results, because Swift doesn't have a way to mark functions as pure. Also, methods in OOP are considered actions, which means that executing them might result in side effects. If the method has no side effects, then it can safely be converted to a computed property.
Note that both of the above statements are purely from a semantic perspective, as it might well happen for computed properties to have side effects that we don't expect, and methods to be pure.
Historically description is a property on NSObject and many would expect that it continues the same in Swift. Adding parens after it will only add confusion.
EDIT:
After furious downvoting I have to clarify something - if it is accessed via dot syntax, it can be considered a property. It doesn't matter what's under the hood. You can't access usual methods with dot syntax.
Besides, calling this property did not require extra parens, like in the case of Swift, which may lead to confusion.
An updated/fixed version of Benjamin Wen's answer incorporating Cristik's suggestion.
class MyClass {
var prop: Int {
return 88
}
func foo() -> Int {
return 88
}
}
func test(times: u_long) {
func testProp(times: u_long) -> TimeInterval {
let myClass = MyClass()
let starting = CACurrentMediaTime()
for _ in 0...times {
_ = myClass.prop
}
let ending = CACurrentMediaTime()
return ending - starting
}
func testFunc(times: u_long) -> TimeInterval {
let myClass = MyClass()
let starting = CACurrentMediaTime()
for _ in 0...times {
_ = myClass.foo()
}
let ending = CACurrentMediaTime()
return ending - starting
}
print("prop: \(testProp(times: times))")
print("func: \(testFunc(times: times))")
}
test(times: 100000)
test(times: 1000000)
test(times: 10000000)
test(times: 100000000)

Is a static boolean a reference type in Swift?

I'm making some switches. In my MenuScene class there's some booleans that are static variables, booleans, to represent the states of these switches.
Are these addressable as reference types, so I can be sure that other objects are able to change their state with a unique reference to them?
The dream, in my dreamy pseudo code, I'm hoping changes to iAmOn impact the state of myButtonABC_state
class MenuScene {
static var myButtonABC_state: Bool = false
static var myButtonXYZ_state: Bool = false
override onDidMoveToView {
let buttonABC = Button(withState: MenuScene.myButtonABC_state)
let buttonXYZ = Button(withState: MenuScene.myButtonXYZ_state)
}
}
In a button class
class Button {
var iAmOn: Bool = false
init(withState state: Bool){
iAmOn = state
}
override onTouchesBegun(... etc...){
if iAmOn { iAMOn = false }
else { iAmOn = true}
}
}
Bool is a struct in Swift; structs are value types. It doesn't matter if it's static var, class var, let, var, etc., the type is what matters--so no, Bool is value type.
I think you are not 100% on all of the terminology (mostly because Apple doesn't really cover it much in documentation as usual, lol).
There are "Swift Types" (Bool, Int, your classes/structs, etc), and "Variable/Constant Types" (which hold data in a memory register, such as references or actual-values), as well as "Memory Register Write/Read Types" (variable vs vonstant, mutable vs immutable, var vs let).
Don't be frustrated.. It's a bit confusing for everyone... Especially at first and without great documentation. (I tried learning C++ pointers early age and it was way over my head).
Here's a good reference material: (towards the bottom)
https://developer.apple.com/library/content/documentation/Swift/Conceptual/Swift_Programming_Language/ClassesAndStructures.html
Basically, if you want to hold a reference to something, you have to use a Reference Type memory register. This means using a class instance Static makes no difference:
/* Test1: */
struct Hi {
static var sup = "hey"
}
var z = Hi.sup
Hi.sup = "yo"
print(z) // prints "hey"
/* Test 2: */
class Hi2 {
static var sup = "hey"
}
var z2 = Hi2.sup
Hi2.sup = "yo"
print(z2) // Prints "hey"
If you feel like you need a pointer to something that isn't inside of a class, then you can use UnsafeMutablePointer or something like that from OBJc code.
Or, you can wrap a bool inside of a class object (which are always references).
final class RefBool {
var val: Bool
init(_ value: Bool) { val = value }
}
And here is some interesting behavior for reference types using let:
let someBool: RefBool
someBool = RefBool(true)
someBool = RefBool(false) // wont compile.. someBool is a `let`
someBool.val = false // will compile because of reference type and member is `var`