Swiftpack.co -  marcprux/MemoZ as Swift Package
Swiftpack.co is a collection of thousands of indexed Swift packages. Search packages.
marcprux/MemoZ
Memoization Caching for Swift; Trade Space for Time on Linux, Windows & Apple Platforms
.package(url: "https://github.com/marcprux/MemoZ.git", from: "1.4.5")

MemoZ – Zero-line Memoization for Swift

Build Status Swift Package Manager compatible Platform

MemoZ is a cross-platform (Apple + Linux + Windows) Swift microframework for zero-line caching of computed properties on pure value types.

let value = SomeHashableValue()
let slow = value.slowProp // O(Nˣ)

import MemoZ // provides Hashable.memoz…

let stillSlow = value.memoz.slowProp // O(Nˣ) compute and cache the property
let VERY_FAST = value.memoz.slowProp // O(1) to retrieve from memoz cache

How does it work?

tl;dr: Magic

MemoZ is not really magic: it doesn't somehow speed up the result of an expensiveComputation property. Rather, it provides an extension to Hashable with the property memoz, which will return a Memoization instance that dynamically passes-through the subsequent property accessor and caches the result in a single global NSCache.

It is designed to operate on value types (structs & enums), and requires that the property computation be pure (no side-effects) and referentially transparent (no inputs other than the subject value).

Sample Usage

The recommended convention for memoizing expensive computations in the property prop is to expose a memoized version of the property named propZ.

import MemoZ

extension Sequence where Element : Numeric {
    /// Computes the sum of all the elements.
    /// - Complexity: O(N)
    var sum: Element { self.reduce(0, +) }
}

extension Sequence where Element : Numeric, Self : Hashable {
    /// Computes & memoizes the sum of all the elements.
    /// - Complexity: Initial: O(N) MemoiZed: O(1)
    var sumZ: Element { self.memoz.sum }
}

// Measure the performance of non-memoized & memoized `sum`
class MemoZDemo: XCTestCase {
    /// A sequence of integers ranging from -1M through +1M
    let millions = (-1_000_000...1_000_000)

    func testCalculatedSum() {
        // average: 1.299, relative standard deviation: 0.509%, values: [1.312717, 1.296008, 1.306766, 1.298375, 1.299257, 1.303043, 1.296738, 1.294311, 1.288839, 1.293301]
        measure { XCTAssertEqual(millions.sum, 0) }
    }

    func testMemoizedSum() {
        // average: 0.133, relative standard deviation: 299.900%, values: [1.332549, 0.000051, 0.000018, 0.000032, 0.000110, 0.000021, 0.000016, 0.000015, 0.000014, 0.000123]
        measure { XCTAssertEqual(millions.sumZ, 0) }
    }
    
    override func tearDown() {
        super.tearDown()
        MemoizationCache.shared.clear() // clear out the global cache
    }
}

Memoization

Wikipedia describes the technique of memoization as:

A memoized function "remembers" the results corresponding to some set of specific inputs. Subsequent calls with remembered inputs return the remembered result rather than recalculating it, thus eliminating the primary cost of a call with given parameters from all but the first call made to the function with those parameters. The set of remembered associations may be a fixed-size set controlled by a replacement algorithm or a fixed set, depending on the nature of the function and its use. A function can only be memoized if it is referentially transparent; that is, only if calling the function has exactly the same effect as replacing that function call with its return value. […]

Memoization is a way to lower a function's time cost in exchange for space cost; that is, memoized functions become optimized for speed in exchange for a higher use of computer memory space. The time/space "cost" of algorithms has a specific name in computing: computational complexity. All functions have a computational complexity in time (i.e. they take time to execute) and in space.

Setup

MemoZ is distributed as a source-level Swift Package.

Swift Package Manager (SPM)

The Swift Package Manager is a dependency manager integrated with the Swift build system. To learn how to use the Swift Package Manager for your project, please read the official documentation.

Add MemoZ to the dependencies of your Package.swift file and refer to that dependency in your target.

// swift-tools-version:5.0
import PackageDescription
let package = Package(
    name: "<Your Product Name>",
    dependencies: [
        .package(url: "https://github.com/marcprux/MemoZ.git", .upToNextMajor(from: "1.3.0"))
    ],
    targets: [
        .target(name: "<Your Target Name>", dependencies: ["MemoZ"])
    ]
)

After adding the dependency, you can fetch the library with:

$ swift package resolve

Xcode

You can add MemoZ to your Xcode Swift project using the "Swift Packages" feature.

  1. File > Swift Packages > Add Package Dependency…
  2. Add the MemoZ repository: https://github.com/marcprux/MemoZ
  3. Use 1.3.0 (or master for bleeding-edge)
  4. Add import MemoZ to any source file that will use memoz

Alternatively, if you are trying to minimize dependencies, you can simply copy the MemoZ.swift file into your project: all the code is in that single small file, which itself has no dependencies of its own (other than Foundation's NSCache).

Error Handling

MemoZ uses the keyPath as a key for the memoization cache, and as such, are performed on computed properties (which can be implemented via extensions). Computed property accessors cannot throw errors, but error handling can be accomplished using the Result type. For example:

extension BidirectionalCollection {
    /// Returns the first and last element of this collection, or else an error if the collection is empty
    var firstAndLast: Result<(Element, Element), Error> {
        Result {
            guard let first = first else {
                throw CocoaError(.coderValueNotFound)
            }
            return (first, last ?? first)
        }
    }
}

/// `Result.get` is used to convert `Result.failure` into a thrown error
XCTAssertThrowsError(try emptyArray.memoz.firstAndLast.get())

Parameterizing Memoization

Although you cannot memoize an arbitrary function call, you can parameterize the computation by implementing the keyPath as a subscript with Hashable parameters. For example, if you want to be able to memoize the results of JSON encoding based on various formatting properties, you might make this extension on Encodable:

extension Encodable {
    /// A JSON blob with the given formatting parameters.
    subscript(JSONFormatted pretty: Bool, sorted sorted: Bool? = nil, noslash noslash: Bool = true) -> Result<Data, Error> {
        Result {
            let encoder = JSONEncoder()
            var fmt = JSONEncoder.OutputFormatting()
            if pretty { fmt.insert(.prettyPrinted) }
            if sorted ?? pretty { fmt.insert(.sortedKeys) }
            if noslash { fmt.insert(.withoutEscapingSlashes) }
            encoder.outputFormatting = fmt
            return try encoder.encode(self)
        }
    }
}

Since all of the parameter arguments to the keypath are themselves Hashable, you can memoize the results with:

try instance.memoz[JSONFormatted: true].get() // pretty
try instance.memoz[JSONFormatted: false, sorted: true].get() // unformatted & sorted

Keying on KeyPath

The keyPath for memoizing can itself take a keyPath (since it is a Hashable parameter to the subscript), which allows memozied permutations on computations that work with key paths. For example:

extension Sequence {
    /// Sorts the sequence by the the given `keyPath` of the element
    subscript<T: Comparable>(sorting sortPath: KeyPath<Element, T>) -> [Element] {
        return self.sorted(by: {
            $0[keyPath: sortPath] < $1[keyPath: sortPath]
        })
    }
}

extension Array where Element: Collection & Hashable {
    /// "C", "BB", "AAA"
    var sortedByCountZ: [Element] {
        self.memoz[sorting: \.count]
    }
}

extension Array where Element: Comparable & Hashable {
    /// "AAA", "BB", "C"
    var sortedBySelfZ: [Element] {
        self.memoz[sorting: \.self]
    }
}

extension MemoZTests {
    func testMemoKeyedSubscript() {
        let strs = ["AAA", "C", "BB"]
        XCTAssertEqual(strs.sortedBySelfZ, ["AAA", "BB", "C"])
        XCTAssertEqual(strs.sortedByCountZ, ["C", "BB", "AAA"])
    }
}

Inline Function Caching

You may want to memoize a function inline without having to make a calculating keyPath in a separate extension. This can be accomplished by wrapping the arguments to the function in a Hashable struct, and using an instance as the subject of memoz:

/// Sum all the numbers from `from` to `to`
/// - Complexity: initial: O(to-from) memoized: O(1)
func summit(from: Int, to: Int) -> Int {
    /// Sum all the numbers from `from` to `to`
    /// - Complexity: O(to-from)
    func sumSequence(from: Int, to: Int) -> Int {
        (from...to).reduce(0, +)
    }

    /// Wrap the arguments to `sumSequence`
    struct Summer : Hashable {
        let from, to: Int
        var sum: Int { sumSequence(from: from, to: to) }
    }

    return Summer(from: from, to: to).memoz.sum
}

This approach does introduce a lot of boilerplate to the memoization process, but an advantage is that the entire implementation can be encapsulated, and other types don't need to be "polluted" with computed properties that otherwise won't be used.

Sequential Memoization

Note that the memoz only caches the adjacent keyPath. If you would like to memoize multiple sequential key paths, this can be done with multiple chained memoz calls, like so:

let costlyExpensiveSlowValue = instance.costlyProp.expensizeProp.slowProp
let fastQuickSpeedyValue = instance.memoz.costlyProp.memoz.expensizeProp.memoz.slowProp

Memoization vs. Laziness

Lazy evaluation allows expensive computations to be deferred until later. So whereas this function to filter by even numbers will need to only go through all million elements:

let million = 1...1_000_000
measure { // average: 1.158, relative standard deviation: 0.909%, values: [1.172126, 1.171695, 1.160608, 1.159658, 1.162762, 1.155741, 1.134647, 1.162453, 1.150810, 1.149786]
    XCTAssertEqual(2, million.filter(\.isEven).first)
}

The lazy version will just go though 2 elements:

let million = 1...1_000_000
measure { // average: 0.000, relative standard deviation: 245.818%, values: [0.001246, 0.000049, 0.000024, 0.000021, 0.000020, 0.000043, 0.000027, 0.000020, 0.000020, 0.000019]
    XCTAssertEqual(2, million.lazy.filter(\.isEven).first)
}

Memoization can be thought of as the opposite of laziness: rather than lazily deferring computations until the very last moment, memoization eagerly computes an entire value just once.

Both memoization and laziness require that subsequent operations be referentially transparent, but laziness imposes the additional restriction that any lazily-evaluated computations must be deferrable (via the @escaping closure parameter). You should avoid memoizing lazy sequences, since the creation of the lazy sequence itself is usually already very fast, and memoizing a lazy chain of operations will result in any escaping function closures being captured in the cache.

Thread Safety

MemoZ is as thread-safe as the underlying property computation. The cache is locked for reading and writing, but it should be noted that simultaneous executions of uncached property computations are not synchronized, which means that two computations can be performed simultaneously (one of whose results will be cached).

This example shows a DispatchQueue executing many memoizations in parallel:

extension MemoZDemo {
    /// A bunch of random numbers from the given offset
    func rangeLimts(count: Int = 20, offset: Int = 1_000_000) -> [Int] {
        (0..<count).map({ $0 + offset }).shuffled()
    }

    func testCalculatedSumParallel() {
        let ranges = rangeLimts()
        measure { // average: 7.115, relative standard deviation: 3.274%, values: [6.579956, 6.785192, 7.074619, 7.123436, 7.242951, 7.295850, 7.326060, 7.285277, 7.249500, 7.187203]
            DispatchQueue.concurrentPerform(iterations: ranges.count) { i in
                XCTAssertEqual((-ranges[i]...ranges[i]).sum, 0)
            }
        }
    }

    func testMemoziedSumParallel() {
        let ranges = rangeLimts()
        measure { // average: 0.671, relative standard deviation: 299.856%, values: [6.708572, 0.000535, 0.000298, 0.000287, 0.000380, 0.000400, 0.000337, 0.000251, 0.000225, 0.000183]
            DispatchQueue.concurrentPerform(iterations: ranges.count) { i in
                XCTAssertEqual((-ranges[i]...ranges[i]).sumZ, 0)
            }
        }
    }
}

Cache Cleanup

MemoZ uses NSCache, which automatically drops values when memory pressure is experienced. The exact details of this process on macOS are vague, but you can also clear out the cache manually.

A common place to perform manual cache clearing in a mobile app is when the app enters the background: emptying the cache will reduce the app's overall memory footprint and thereby reduce the chances that the system will terminate the app (at the cost of needing to re-build the memoization cache if & when the app is again reactivated).

For an AppDelegate-based app, you can add:

func applicationDidEnterBackground(_ application: UIApplication) {
    MemoizationCache.shared.clear() // clear out the global cache
}

Or for SwiftUI:

struct MyApp: App {
    @Environment(\.scenePhase) private var scenePhase

    var body: some Scene {
        WindowGroup {
            ContentView()
        }
        .onChange(of: scenePhase) { phase in
            if phase == .background {
                MemoizationCache.shared.clear() // clear out the global cache
            }
        }
    }
}

Cache Customization

Memoization caches can be partitioned into separate global caches, each with its own size limit:

extension MemoizationCache {
    /// A domain-specific cache that caps the number of cached instances at around 1,000
    static let domainCache = MemoizationCache(countLimit: 1_000)
}

func testCachePartition() {
    let uuids = (0...100_000).map({ _ in UUID() }) // a bunch of random strings
    measure {
        // the following two calls are the same, except the second one uses a custom cache rather than the default global cache
        XCTAssertEqual(3800038, uuids.memoz.description.count)
        XCTAssertEqual(3800038, uuids[memoz: .domainCache].description.count)
    }
}

The countLimit is an approximate maximum that the cache should hold. As per NSCache's documentation:

This is not a strict limit—if the cache goes over the limit, an object in the cache could be evicted instantly, later, or possibly never, depending on the implementation details of the cache.

An additional advantage of using your own cache is that you can enable & disable it (by setting it to nil) on a global basis from a single location and compare the performance of your app with caching disabled.

Measuring Performance

Any caching solution, memoization included, introduces some performance overhead in time and space. As such, care should be taken that the memoized computation is actually costly enough to make caching worthwhile.

For example, the following (pointless) memoization of a constant value would result in a 3x performance degradation:

struct Pointless : Hashable {
    var alwaysOne: Int { 1 } // 1, always
    var alwaysOneZ: Int { memoz.alwaysOne } // we'd be better off without memoization!
}

func testPointlessComputation() {
    let pointless = Pointless()
    measure { // average: 0.013, relative standard deviation: 10.553%, values: [0.017202, 0.014706, 0.013114, 0.012666, 0.012782, 0.012689, 0.012770, 0.012431, 0.012440, 0.013157]
        for _ in 1...10_000 {
            XCTAssertEqual(1, pointless.alwaysOne)
        }
    }
}

func testPointlessMemoization() {
    let pointless = Pointless()
    measure { // average: 0.040, relative standard deviation: 3.971%, values: [0.043689, 0.039205, 0.038396, 0.041741, 0.038969, 0.039193, 0.039041, 0.039618, 0.038810, 0.038761]
        for _ in 1...10_000 {
            XCTAssertEqual(1, pointless.alwaysOneZ)
        }
    }
}

An advantage of MemoZ's zero-line memoization is that you can easily put in measuring tests to ensure that memoization is actually worthwhile. A useful statistic is XCTest's measure function, which runs the block 10 times and reports the individual times and the standard deviation between the times. A higher standard deviation generally indicates that the memoization is yielding a performance gain.

Gotchas

Care must be taken that memozied computations are truly referentially transparent. It might be tempting to cache results of parsing dates or numbers using built-in static parsing utilities, but be mindful that these functions often take external environment settings (such as the current locale) into account, so if the environment changes between invocations, the memoized result will not be the same as the computed result.

In general, if your memoization subjects are pure value types (i.e., they transitively contain no properties that are reference types), and the memozied properties are dependent only on the state of the subject, then memoz is a safe and transparent operation.

Implementation Details

MemoZ is a coarse-grained caching library that maintains a single global cache keyed by a combination of a target Hashable and a key path. As such, it just works for most cases, but care must be taken that:

  1. the target Hashable instance is a value type
  2. the predicate keyPath is pure: is must have no side-effects and be referentially transparent

Platform Differences

MemoZ behaves similiarly on Apple platforms (macOS, iOS, tvOS, watchOS) and non-Apple platforms (Linux, Windows), but note that the underlying NSCache implementation differs. On none-Apple platforms, this will probably be using the reference NSCache.swift implementation, which is a simpified cache that does not offer any automatic purging (based on memory pressure or otherwise). The practical result of this is that on non-Apple platforms the application must perform cache maintenance manually in order to prevent unbounded growth and eventual memory exhaustion, whereas on Apple platforms, some amount of automatic cache cleanup should be expected.

“Zero-line”?

Many libraries advertise that their capabilities can be used in a single line of code. Since MemoZ's caching can be added to the same line of code as the property it is caching, it doesn't add any lines of code, and so it is a "zero-liner" instead of a "one-liner".

Technically, one might say it is a “six-character” API (“memoz.” ), but “zero-line” sounds awesomer.

Other Features

MemoZ also exposes its own Cache instance that wraps an NSCache and permits caching value types (NSCache itself is limited to reference AnyObject instances for keys and values).

The cache itself is not specific to memoization, so it can cache the results of an arbitrary function, at the cost of some additional book-keeping and cache key ceremony.

func testCacheAPI() {
    let cache = Cache<Int, String>()

    func cachedDescription(for number: Int) -> String {
        cache.fetch(key: number, create: { key in
            String(describing: key)
        })
    }

    XCTAssertEqual("123", cachedDescription(for: 123))
    XCTAssertEqual("123", cachedDescription(for: 123)) // this will return the cached result
    
    cache.clear()
    
    XCTAssertEqual("123", cachedDescription(for: 123)) // this will again be un-cached
}

Testing

Tests on the host computer can be run with:

swift test --enable-test-discovery

Container testing can be done with Docker. For examplem for Linux:

docker run --rm --interactive --tty --volume "$(pwd):/src" --workdir "/src" swift:latest swift test --enable-test-discovery

GitHub

link
Stars: 4
Last commit: 1 week ago

Ad: Job Offers

iOS Software Engineer @ Perry Street Software
Perry Street Software is Jack’d and SCRUFF. We are two of the world’s largest gay, bi, trans and queer social dating apps on iOS and Android. Our brands reach more than 20 million members worldwide so members can connect, meet and express themselves on a platform that prioritizes privacy and security. We invest heavily into SwiftUI and using Swift Packages to modularize the codebase.

Release Notes

1.4.0
27 weeks ago

Use NSRecursiveLock instead of DispatchQueue for per-cache-value locking

Swiftpack is being maintained by Petr Pavlik | @ptrpavlik | @swiftpackco | API