Swift Algorithms Versions Save

Additions

• Sequences have new grouped(by:) and keyed(by:) methods, which simplify building dictionaries based on sequence elements: (#197)

  let evensOdds = (1...10).grouped(by: { $0.isMultiple(of: 2) ? "even" : "odd" })
  // ["even": [2, 4, 6, 8, 10], "odd": [1, 3, 5, 7, 9]]
  
  let animals = ["Aardvark", "Alpaca", "Armadillo", "Barracuda", "Buffalo", ...]
  let alphaAnimals = animals.keyed(by: \.first!, resolvingConflictsWith: { _, first, _ in first })
  // ["A": "Aardvark", "B": "Barracuda", ...]
  

• The endOfPrefix(while:) and startOfSuffix(while) methods are now public. These methods were previously implementation details of the trimming group of methods. (#211)

Fixes

• Documentation and performance improvements. (#202, #210)

The 1.2.0 release includes contributions from @amomchilov, @mattyoung, @natecook1000, @stephentyrone, and @ttsugriy. Thank you!

Additions

• New nonmutating partitioned(by:) methods that return both the failing and passing elements of a sequence or collection in separate arrays. (#152)

  let cast = ["Vivien", "Marlon", "Kim", "Karl"]
  let (longNames, shortNames) = cast.partitioned(by: { $0.count < 5 })
  
  print(longNames)
  // Prints "["Vivien", "Marlon"]"
  print(shortNames)
  // Prints "["Kim", "Karl"]"
  

• A new evenlyChunked(in:) method that chunks a sequence or collection into the specified number of evenly-sized subsequences. (#96)

  let numbers = Array(0..<10) 
  for chunk in numbers.evenlyChunked(in: 4) {
      print(chunk)
  }
  // Prints: 
  // [0, 1, 2]
  // [3, 4, 5]
  // [6, 7]
  // [8, 9]
  

Changes

• The package now has a minimum Swift version of Swift 5.4.

Fixes

• Improvements to documentation, including a new DocC-based documentation structure.

The 1.1.0 release includes contributions from @dickoff, @gwynne, @mdznr, @natecook1000, @ole, @stephentyrone, and @timvermeulen. Thank you!

Swift Algorithms is now source stable!

Changes

• Most sequence and collection types have been renamed, following a more consistent naming structure:
  • The Lazy prefix was dropped.
  • Either a Sequence or Collection suffix was added depending on whether or not the type is unconditionally a collection.
  • The base name was derived from the name of the method that produces it, including an argument label to disambiguate if necessary.
• Types that can only be produced from a lazy sequence chain now unconditionally conform to LazySequenceProtocol and wrap the base sequence instead of the lazy wrapper, making some return types slightly simpler.
  • e.g. [1, 2, 3].lazy.reductions(+) now returns ExclusiveReductionsSequence<[Int]>, not ExclusiveReductionsSequence<LazySequence<[Int]>>.
• The generic parameters of the ExclusiveReductions type have been swapped, putting the base collection first and the result type second.
• The Indices associated type of IndexedCollection now matches Base.Indices.

Removals

• Previously deprecated type and method names have been removed:
  • The Chain type alias for Chain2Sequence
  • The chained(with:) method which was replaced with the chain(_:_:) free function
  • The LazyChunked and Chunked type aliases for ChunkedByCollection
  • The rotate(subrange:at:) and rotate(at:) methods which were renamed to rotate(subrange:toStartAt:) and rotate(toStartAt:) respectively

Fixes

• The StridingSequence and StridingCollection types now conditionally conform to LazySequenceProtocol, allowing the striding(by:) method to properly propagate laziness in a lazy sequence chain.
• Fixed chunked(by:) to always compare two consecutive elements rather than each element with the first element of the current chunk. (#162)

Additions

Expanded versatility for two existing operations:

• A series of joined(by:) overloads concatenate a sequence of sequences using an element or a collection, either passed in or generated from consecutive elements via a closure. (#138)
• Additional trimming(while:) methods for trimming only the start or end of a collection, as well as mutating versions of all three variants. (#104)

Additions

Two new additions to the list of algorithms:

• adjacentPairs() lazily iterates over tuples of adjacent elements of a sequence. (#119)
• minAndMax() finds both the smallest and largest elements of a sequence in a single pass. (#90)

Changes

• When calling chunked(on:), the resulting collection has an element type of (Subject, SubSequence) instead of just SubSequence, making the subject value available when iterating.

  let numbers = [5, 6, -3, -9, -11, 2, 7, 6]
  for (signum, values) in numbers.chunked(on: { $0.signum() }) {
      print(signum, values)
  }
  // 1 [5, 6]
  // -1 [-3, -9, -11]
  // 1 [2, 7, 6]
  

Fixes

• Improvements to the documentation and PR templates.

Fixes

• Product2 associated type inference error in release build (#130)

Additions

• The compacted() method lazily finds the non-nil elements of a sequence or collection (#112).

Changes

• The uniqued() method now lazily computes the unique elements of a sequence or a collection (#71). Pass this resulting sequence to an Array initializer to recover the behavior of the previous release.
• Calling cycled(times:) now returns a new FiniteCycle type, which has the same conformances as its underlying collection (#106).
• The base collections of the sequence and collection wrapper types are no longer public (#85, #125), and the wrapper types no longer conform to the Equatable or Hashable protocols (#124). If you need those conformances, convert the wrapper type to an Array or other collection currrency type before storing. Please file an issue if these changes pose a problem for your use case.

Additions

More new algorithms to join the party:

• A lazy version of the standard library's two split methods. (#78)
• firstNonNil(_:) returns the first non-nil element from an optional-generating transform. (#31)
• uniquePermutations() skips duplicates when generating permutations of a collection. (#91)
• The reductions methods return all the in-between states of reducing a sequence or collection. (#46)

Fixes

• Methods and computed properties are more consistently marked as inlinable, resolving a performance regression.
• The Stride type now efficiently calculates distances between positions, supported by the underlying collection.
• Better test coverage and improved diagnostics for comparing sequences.
• Fixed links and improved documentation.

Additions

An exciting group of new algorithms, contributed by the community:

• trimming(while:) returns a subsequence of a bidirectional collection with the matching elements removed from the start and end. (#4)
• min(ofCount:) and max(ofCount:) find the smallest or largest elements in a collection. (#9, #77)
• windows(ofCount:) lets you iterate over all the overlapping subsequences of a particular length. (#20)
• striding(by:) iterates over every nth element of a sequence or collection. (#24)
• interspersed(with:) places a new element between every pair of elements in a sequence or collection. (#35)
• chunks(ofCount:) breaks a collection into subsequences of the given number of elements. (#54)
• suffix(while:) matches the standard library's prefix(while:), by returning the suffix of all matching elements from a bidirectional collection. (#65)
• Variations of combinations(ofCount:) and permutations(ofCount:) that take a range expression as a parameter, returning combinations and permutations of multiple lengths. (#51, #56)

Changes

• The LazyChunked type now precomputes its startIndex, making performance more predictable when using the collection.

Fixes

• randomSample(count:) no longer traps in rare circumstances.
• Index calculations have been improved in a variety of collection wrappers.
• A variety of documentation improvements and corrections.

Changes

• The rotate(at:) method has been updated to rotate(toStartAt:), with the old name deprecated.
• The chained(with:) method has been changed to the free function chain(_:_:), with the old version deprecated.
• Algorithms now uses RealModule from the swift-numerics package for its cross-platform elementary functions.
• Sequence/collection wrapper types, like Permutations and Indexed, now have conformance to the lazy protocols, so that any following operations maintain their laziness.

Fixes

• Swift Algorithms now builds under SwiftPM on Windows.
• A wide variety of errors, misspellings, and ommissions in the documentation and guides have been fixed.
• Index/distance calculations for the Product2 and Chain types have been corrected.
• Calling stablePartition(subrange:by:) now correctly uses the subrange's length instead of the whole collection.