Skip to content

πŸ’» A fast and flexible O(n) difference algorithm framework for Swift collection.

License

Notifications You must be signed in to change notification settings

wireapp/DifferenceKit

Repository files navigation

A fast and flexible O(n) difference algorithm framework for Swift collection.
The algorithm is optimized based on the Paul Heckel's algorithm.

Swift4 Release CocoaPods Carthage
Build Status Platform Lincense


Features

βœ… Automate to calculate operations for batch-updates of UITableView, UICollectionView, NSTableView and NSCollectionView

βœ… O(n) difference algorithm optimized for performance in Swift

βœ… Supports both linear and sectioned collection

βœ… Supports calculating differences with best effort even if elements or section contains duplicates

βœ… Supports all operations for animated UI batch-updates including section reloads


Sample app is imitated from DataSources πŸ’Ύ


Algorithm

The algorithm used in DifferenceKit is optimized based on the Paul Heckel's algorithm.
See also his paper "A technique for isolating differences between files" released in 1978.
RxDataSources and IGListKit are also implemented based on his algorithm.
This allows all types of differences to be computed in linear time O(n).

However, in performBatchUpdates of UITableView and UICollectionView, there are combinations of operations that cause crash when applied simultaneously.
To solve this problem, DifferenceKit takes an approach of split the set of differences at the minimal stages that can be perform batch-updates with no crashes.

Implementation is here.


Documentation

⚠️ Since DifferenceKit is currently released in alpha version, updates may contain breaking changes.
See docs in GitHub Pages.
Documentation is generated by jazzy.


Getting Started

Example codes

The type of the element that to take the differences must be conform to the Differentiable protocol.
The differenceIdentifier's type is determined generic by the associated type:

struct User: Differentiable {
    let id: Int
    let name: String

    var differenceIdentifier: Int {
        return id
    }

    func isContentEqual(to source: User) -> Bool {
        return name == source.name
    }
}

In the case of definition above, id uniquely identifies the element and get to know the user updated by comparing equality of name of the elements in source and target.

There are default implementations of Differentiable for the types that conformed to Equatable or Hashable:

public extension Differentiable where Self: Equatable {
    func isContentEqual(to source: Self) -> Bool {
        return self == source
    }
}

public extension Differentiable where Self: Hashable {
    var differenceIdentifier: Self {
        return self
    }
}

So, you can simply:

extension String: Differentiable {}

Calculates the differences by creating StagedChangeset from two collections of elements conforming to Differentiable:

let source = [
    User(id: 0, name: "Vincent"),
    User(id: 1, name: "Jules")
]
let target = [
    User(id: 1, name: "Jules"),
    User(id: 0, name: "Vincent"),
    User(id: 2, name: "Butch")
]

let changeset = StagedChangeset(source: source, target: target)

If you want to include multiple types conformed to Differentiable in the collection, use AnyDifferentiable:

let source = [
    AnyDifferentiable("A"),
    AnyDifferentiable(User(id: 0, name: "Vincent"))
]

In the case of sectioned collection, the section itself must have a unique identifier and be able to compare whether there is an update.
So each section must conform to DifferentiableSection protocol, but in most cases you can use ArraySection that general type conformed to it.
ArraySection requires a model conforming to Differentiable for differentiate from other sections:

enum Model: Differentiable {
    case a, b, c
}

let source: [ArraySection<Model, String>] = [
    ArraySection(model: .a, elements: ["A", "B"]),
    ArraySection(model: .b, elements: ["C"])
]
let target: [ArraySection<Model, String>] = [
    ArraySection(model: .c, elements: ["D", "E"]),
    ArraySection(model: .a, elements: ["A"]),
    ArraySection(model: .b, elements: ["B", "C"])
]

let changeset = StagedChangeset(source: source, target: target)

You can perform incremental updates on UITableView and UICollectionView using the created StagedChangeset.

⚠️ Don't forget to synchronously update the data referenced by the data-source, with the data passed in the setData closure. The differences are applied in stages, and failing to do so is bound to create a crash:

tableView.reload(using: changeset, with: .fade) { data in
    dataSource.data = data
}

Batch-updates using too large amount of differences may adversely affect to performance.
Returning true with interrupt closure then falls back to reloadData:

collectionView.reload(using: changeset, interrupt: { $0.changeCount > 100 }) { data in
    dataSource.data = data
}

Comparison with Other Frameworks

Made a fair comparison as much as possible in features and performance with other popular and awesome frameworks.
This does NOT determine superiority or inferiority of the frameworks. I know that each framework has different benefits.
The frameworks and its version that compared is below.

Features comparison

- Supported collection

Linear Sectioned Duplicate Element/Section
DifferenceKit βœ… βœ… βœ…
RxDataSources ❌ βœ… ❌
FlexibleDiff βœ… βœ… βœ…
IGListKit βœ… ❌ βœ…
ListDiff βœ… ❌ βœ…
DeepDiff βœ… ❌ βœ…
Differ βœ… βœ… βœ…
Dwifft βœ… βœ… βœ…

Linear means 1-dimensional collection.
Sectioned means 2-dimensional collection.

- Supported element differences

Delete Insert Move Reload Move across sections
DifferenceKit βœ… βœ… βœ… βœ… βœ…
RxDataSources βœ… βœ… βœ… βœ… βœ…
FlexibleDiff βœ… βœ… βœ… βœ… ❌
IGListKit βœ… βœ… βœ… βœ… ❌
ListDiff βœ… βœ… βœ… βœ… ❌
DeepDiff βœ… βœ… βœ… βœ… / ❌ ❌
Differ βœ… βœ… βœ… ❌ ❌
Dwifft βœ… βœ… ❌ ❌ ❌

- Supported section differences

Delete Insert Move Reload
DifferenceKit βœ… βœ… βœ… βœ…
RxDataSources βœ… βœ… βœ… ❌
FlexibleDiff βœ… βœ… βœ… βœ…
IGListKit ❌ ❌ ❌ ❌
ListDiff ❌ ❌ ❌ ❌
DeepDiff ❌ ❌ ❌ ❌
Differ βœ… βœ… βœ… ❌
Dwifft βœ… βœ… ❌ ❌

Performance comparison

Performance was measured using XCTestCase.measure on iPhoneX simulator with -O -whole-module-optimization.
Use Foundation.UUID as an element.

⚠️ If Move is included in the difference, performance may obviously decrease in some frameworks.
⚠️ DeepDiff may had increased the processing speed by misuse of Hashable in algorithm.

- From 5,000 elements to 500 deleted and 500 inserted

Time(second)
DifferenceKit 0.0032
RxDataSources 0.0078
FlexibleDiff 0.0168
IGListKit 0.0412
ListDiff 0.0388
DeepDiff 0.0150
Differ 0.3260
Dwifft 33.600

- From 10,000 elements to 1,000 deleted and 1,000 inserted

Time(second)
DifferenceKit 0.0076
RxDataSources 0.0143
FlexibleDiff 0.0305
IGListKit 0.0891
ListDiff 0.0802
DeepDiff 0.0300
Differ 1.3450
Dwifft ❌

- From 100,000 elements to 10,000 deleted and 10,000 inserted

Time(second)
DifferenceKit 0.087
RxDataSources 0.179
FlexibleDiff 0.356
IGListKit 1.329
ListDiff 1.026
DeepDiff 0.334
Differ ❌
Dwifft ❌

Requirements

  • Swift4.2+
  • iOS 9.0+
  • tvOS 9.0+
  • OS X 10.9+
  • watchOS 2.0+ (only algorithm)

Installation

To use only algorithm without extensions for UI, add the following:

use_frameworks!

target 'TargetName' do
  pod 'DifferenceKit/Core'
end

iOS/tvOS

To use DifferenceKit with UIKit extension, add the following to your Podfile:

target 'TargetName' do
  pod 'DifferenceKit'
end

or

target 'TargetName' do
  pod 'DifferenceKit/UIKitExtension'
end

macOS

To use DifferenceKit with AppKit extension, add the following to your Podfile:

target 'TargetName' do
  pod 'DifferenceKit/AppKitExtension'
end

watchOS

There is no UI extension for watchOS.
You can use only the algorithm.

Add the following to your Cartfile:

github "ra1028/DifferenceKit"

And run

carthage update

Contribution

Welcome to fork and submit pull requests.
Before submitting pull request, please ensure you have passed the included tests.
If your pull request including new function, please write test cases for it.


License

DifferenceKit is released under the MIT License.

About

πŸ’» A fast and flexible O(n) difference algorithm framework for Swift collection.

Resources

License

Code of conduct

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Swift 97.1%
  • Ruby 1.7%
  • Other 1.2%