Merge pull request #23832 from lorentey/foundation-hashing

[Foundation] Modernize hashing in Foundation's Swift-only types

Author: lorentey

stdlib/private/StdlibUnittest/StdlibUnittest.swift

@@ -2362,17 +2362,22 @@ internal func _checkEquatableImpl<Instance : Equatable>(
       let isEqualXY = x == y
       expectEqual(
         predictedXY, isEqualXY,
-        (predictedXY
-           ? "expected equal, found not equal\n"
-           : "expected not equal, found equal\n") +
-        "lhs (at index \(i)): \(String(reflecting: x))\n" +
-        "rhs (at index \(j)): \(String(reflecting: y))",
+        """
+        \((predictedXY
+           ? "expected equal, found not equal"
+           : "expected not equal, found equal"))
+        lhs (at index \(i)): \(String(reflecting: x))
+        rhs (at index \(j)): \(String(reflecting: y))
+        """,
         stackTrace: stackTrace.pushIf(showFrame, file: file, line: line))
 
       // Not-equal is an inverse of equal.
       expectNotEqual(
         isEqualXY, x != y,
-        "lhs (at index \(i)): \(String(reflecting: x))\nrhs (at index \(j)): \(String(reflecting: y))",
+        """
+        lhs (at index \(i)): \(String(reflecting: x))
+        rhs (at index \(j)): \(String(reflecting: y))
+        """,
         stackTrace: stackTrace.pushIf(showFrame, file: file, line: line))
 
       if !allowBrokenTransitivity {
@@ -2414,6 +2419,10 @@ public func checkEquatable<T : Equatable>(
     showFrame: false)
 }
 
+/// Produce an integer hash value for `value` by feeding it to a dedicated
+/// `Hasher`. This is always done by calling the `hash(into:)` method.
+/// If a non-nil `seed` is given, it is used to perturb the hasher state;
+/// this is useful for resolving accidental hash collisions.
 internal func hash<H: Hashable>(_ value: H, seed: Int? = nil) -> Int {
   var hasher = Hasher()
   if let seed = seed {
@@ -2429,6 +2438,7 @@ internal func hash<H: Hashable>(_ value: H, seed: Int? = nil) -> Int {
 public func checkHashableGroups<Groups: Collection>(
   _ groups: Groups,
   _ message: @autoclosure () -> String = "",
+  allowIncompleteHashing: Bool = false,
   stackTrace: SourceLocStack = SourceLocStack(),
   showFrame: Bool = true,
   file: String = #file, line: UInt = #line
@@ -2446,6 +2456,7 @@ public func checkHashableGroups<Groups: Collection>(
     equalityOracle: equalityOracle,
     hashEqualityOracle: equalityOracle,
     allowBrokenTransitivity: false,
+    allowIncompleteHashing: allowIncompleteHashing,
     stackTrace: stackTrace.pushIf(showFrame, file: file, line: line),
     showFrame: false)
 }
@@ -2457,6 +2468,7 @@ public func checkHashable<Instances: Collection>(
   _ instances: Instances,
   equalityOracle: (Instances.Index, Instances.Index) -> Bool,
   allowBrokenTransitivity: Bool = false,
+  allowIncompleteHashing: Bool = false,
   _ message: @autoclosure () -> String = "",
   stackTrace: SourceLocStack = SourceLocStack(),
   showFrame: Bool = true,
@@ -2467,6 +2479,7 @@ public func checkHashable<Instances: Collection>(
     equalityOracle: equalityOracle,
     hashEqualityOracle: equalityOracle,
     allowBrokenTransitivity: allowBrokenTransitivity,
+    allowIncompleteHashing: allowIncompleteHashing,
     stackTrace: stackTrace.pushIf(showFrame, file: file, line: line),
     showFrame: false)
 }
@@ -2480,6 +2493,7 @@ public func checkHashable<Instances: Collection>(
   equalityOracle: (Instances.Index, Instances.Index) -> Bool,
   hashEqualityOracle: (Instances.Index, Instances.Index) -> Bool,
   allowBrokenTransitivity: Bool = false,
+  allowIncompleteHashing: Bool = false,
   _ message: @autoclosure () -> String = "",
   stackTrace: SourceLocStack = SourceLocStack(),
   showFrame: Bool = true,
@@ -2532,12 +2546,12 @@ public func checkHashable<Instances: Collection>(
         expectEqual(
           x._rawHashValue(seed: 0), y._rawHashValue(seed: 0),
           """
-          _rawHashValue expected to match, found to differ
+          _rawHashValue(seed:) expected to match, found to differ
           lhs (at index \(i)): \(x)
           rhs (at index \(j)): \(y)
           """,
           stackTrace: stackTrace.pushIf(showFrame, file: file, line: line))
-      } else {
+      } else if !allowIncompleteHashing {
         // Try a few different seeds; at least one of them should discriminate
         // between the hashes. It is extremely unlikely this check will fail
         // all ten attempts, unless the type's hash encoding is not unique,

stdlib/public/Darwin/Foundation/AffineTransform.swift

@@ -276,8 +276,13 @@ public struct AffineTransform : ReferenceConvertible, Hashable, CustomStringConv
         return newSize
     }
 
-    public var hashValue : Int {
-        return Int(m11 + m12 + m21 + m22 + tX + tY)
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(m11)
+        hasher.combine(m12)
+        hasher.combine(m21)
+        hasher.combine(m22)
+        hasher.combine(tX)
+        hasher.combine(tY)
     }
 
     public var description: String {

stdlib/public/Darwin/Foundation/Calendar.swift

@@ -899,15 +899,16 @@ public struct Calendar : Hashable, Equatable, ReferenceConvertible, _MutableBoxi
 
     // MARK: -
 
-    public var hashValue : Int {
-        // We implement hash ourselves, because we need to make sure autoupdating calendars have the same hash
+    public func hash(into hasher: inout Hasher) {
+        // We need to make sure autoupdating calendars have the same hash
         if _autoupdating {
-            return 1
+            hasher.combine(false)
         } else {
-            return _handle.map { $0.hash }
+            hasher.combine(true)
+            hasher.combine(_handle.map { $0 })
         }
     }
-    
+
     // MARK: -
     // MARK: Conversion Functions
 

stdlib/public/Darwin/Foundation/CharacterSet.swift

@@ -52,15 +52,15 @@ fileprivate final class __CharacterSetStorage : Hashable {
 
     // MARK: -
 
-    fileprivate var hashValue : Int {
+    fileprivate func hash(into hasher: inout Hasher) {
         switch _backing {
         case .immutable(let cs):
-            return Int(CFHash(cs))
+            hasher.combine(CFHash(cs))
         case .mutable(let cs):
-            return Int(CFHash(cs))
+            hasher.combine(CFHash(cs))
         }
     }
-    
+
     fileprivate static func ==(lhs : __CharacterSetStorage, rhs : __CharacterSetStorage) -> Bool {
         switch (lhs._backing, rhs._backing) {
         case (.immutable(let cs1), .immutable(let cs2)):
@@ -754,8 +754,8 @@ public struct CharacterSet : ReferenceConvertible, Equatable, Hashable, SetAlgeb
 
     // MARK: -
 
-    public var hashValue: Int {
-        return _storage.hashValue
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(_storage)
     }
 
     /// Returns true if the two `CharacterSet`s are equal.

stdlib/public/Darwin/Foundation/Date.swift

@@ -142,14 +142,10 @@ public struct Date : ReferenceConvertible, Comparable, Equatable {
     */
     public static let distantPast = Date(timeIntervalSinceReferenceDate: -63114076800.0)
 
-    public var hashValue: Int {
-        if #available(macOS 10.12, iOS 10.0, *) {
-            return Int(bitPattern: __CFHashDouble(_time))
-        } else { // 10.11 and previous behavior fallback; this must allocate a date to reference the hash value and then throw away the reference
-            return NSDate(timeIntervalSinceReferenceDate: _time).hash
-        }
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(_time)
     }
-    
+
     /// Compare two `Date` values.
     public func compare(_ other: Date) -> ComparisonResult {
         if _time < other.timeIntervalSinceReferenceDate {

stdlib/public/Darwin/Foundation/DateComponents.swift

@@ -263,10 +263,10 @@ public struct DateComponents : ReferenceConvertible, Hashable, Equatable, _Mutab
 
     // MARK: -
 
-    public var hashValue : Int {
-        return _handle.map { $0.hash }
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(_handle._uncopiedReference())
     }
-    
+
     // MARK: - Bridging Helpers
 
     fileprivate init(reference: __shared NSDateComponents) {

stdlib/public/Darwin/Foundation/DateInterval.swift

@@ -155,15 +155,11 @@ public struct DateInterval : ReferenceConvertible, Comparable, Hashable, Codable
         return false
     }
 
-    public var hashValue: Int {
-        var buf: (UInt, UInt) = (UInt(start.timeIntervalSinceReferenceDate), UInt(end.timeIntervalSinceReferenceDate))
-        return withUnsafeMutablePointer(to: &buf) {
-            $0.withMemoryRebound(to: UInt8.self, capacity: 2 * MemoryLayout<UInt>.size / MemoryLayout<UInt8>.size) {
-                return Int(bitPattern: CFHashBytes($0, CFIndex(MemoryLayout<UInt>.size * 2)))
-            }
-        }
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(start)
+        hasher.combine(duration)
     }
-    
+
     @available(macOS 10.12, iOS 10.0, watchOS 3.0, tvOS 10.0, *)
     public static func ==(lhs: DateInterval, rhs: DateInterval) -> Bool {
         return lhs.start == rhs.start && lhs.duration == rhs.duration

stdlib/public/Darwin/Foundation/Decimal.swift

@@ -177,8 +177,12 @@ extension Decimal : Hashable, Comparable {
         return _isNegative != 0 ? -d : d
     }
 
-    public var hashValue: Int {
-        return Int(bitPattern: __CFHashDouble(doubleValue))
+    public func hash(into hasher: inout Hasher) {
+        // FIXME: This is a weak hash.  We should rather normalize self to a
+        // canonical member of the exact same equivalence relation that
+        // NSDecimalCompare implements, then simply feed all components to the
+        // hasher.
+        hasher.combine(doubleValue)
     }
 
     public static func ==(lhs: Decimal, rhs: Decimal) -> Bool {

stdlib/public/Darwin/Foundation/IndexPath.swift

@@ -662,25 +662,31 @@ public struct IndexPath : ReferenceConvertible, Equatable, Hashable, MutableColl
         return .orderedSame
     }
 
-    public var hashValue: Int {
-        func hashIndexes(first: Int, last: Int, count: Int) -> Int {
-            let totalBits = MemoryLayout<Int>.size * 8
-            let lengthBits = 8
-            let firstIndexBits = (totalBits - lengthBits) / 2
-            return count &+ (first << lengthBits) &+ (last << (lengthBits + firstIndexBits))
-        }
-
+    public func hash(into hasher: inout Hasher) {
+        // Note: We compare all indices in ==, so for proper hashing, we must
+        // also feed them all to the hasher.
+        //
+        // To ensure we have unique hash encodings in nested hashing contexts,
+        // we combine the count of indices as well as the indices themselves.
+        // (This matches what Array does.)
         switch _indexes {
-            case .empty: return 0
-            case .single(let index): return index.hashValue
-            case .pair(let first, let second):
-                return hashIndexes(first: first, last: second, count: 2)
-            default:
-                let cnt = _indexes.count
-                return hashIndexes(first: _indexes[0], last: _indexes[cnt - 1], count: cnt)
+        case .empty:
+            hasher.combine(0)
+        case let .single(index):
+            hasher.combine(1)
+            hasher.combine(index)
+        case let .pair(first, second):
+            hasher.combine(2)
+            hasher.combine(first)
+            hasher.combine(second)
+        case let .array(indexes):
+            hasher.combine(indexes.count)
+            for index in indexes {
+                hasher.combine(index)
+            }
         }
     }
-    
+
     // MARK: - Bridging Helpers
 
     fileprivate init(nsIndexPath: __shared ReferenceType) {

stdlib/public/Darwin/Foundation/IndexSet.swift

@@ -137,8 +137,8 @@ public struct IndexSet : ReferenceConvertible, Equatable, BidirectionalCollectio
         _handle = _MutablePairHandle(NSIndexSet(), copying: false)
     }
 
-    public var hashValue: Int {
-        return _handle.map { $0.hash }
+    public func hash(into hasher: inout Hasher) {
+        _handle.map { hasher.combine($0) }
     }
 
     /// Returns the number of integers in `self`.

stdlib/public/Darwin/Foundation/Locale.swift

@@ -405,11 +405,12 @@ public struct Locale : Hashable, Equatable, ReferenceConvertible {
     // MARK: -
     //
 
-    public var hashValue : Int {
+    public func hash(into hasher: inout Hasher) {
         if _autoupdating {
-            return 1
+            hasher.combine(false)
         } else {
-            return _wrapped.hash
+            hasher.combine(true)
+            hasher.combine(_wrapped)
         }
     }
 

stdlib/public/Darwin/Foundation/Measurement.swift

@@ -36,8 +36,19 @@ public struct Measurement<UnitType : Unit> : ReferenceConvertible, Comparable, E
         self.unit = unit
     }
 
-    public var hashValue: Int {
-        return Int(bitPattern: __CFHashDouble(value))
+    public func hash(into hasher: inout Hasher) {
+        // Warning: The canonicalization performed here needs to be kept in
+        // perfect sync with the definition of == below. The floating point
+        // values that are compared there must match exactly with the values fed
+        // to the hasher here, or hashing would break.
+        if let dimension = unit as? Dimension {
+            // We don't need to feed the base unit to the hasher here; all
+            // dimensional measurements of the same type share the same unit.
+            hasher.combine(dimension.converter.baseUnitValue(fromValue: value))
+        } else {
+            hasher.combine(unit)
+            hasher.combine(value)
+        }
     }
 }
 
@@ -170,6 +181,10 @@ extension Measurement {
     /// If `lhs.unit == rhs.unit`, returns `lhs.value == rhs.value`. Otherwise, converts `rhs` to the same unit as `lhs` and then compares the resulting values.
     /// - returns: `true` if the measurements are equal.
     public static func ==<LeftHandSideType, RightHandSideType>(lhs: Measurement<LeftHandSideType>, rhs: Measurement<RightHandSideType>) -> Bool {
+        // Warning: This defines an equivalence relation that needs to be kept
+        // in perfect sync with the hash(into:) definition above. The floating
+        // point values that are fed to the hasher there must match exactly with
+        // the values compared here, or hashing would break.
         if lhs.unit == rhs.unit {
             return lhs.value == rhs.value
         } else {

stdlib/public/Darwin/Foundation/NSRange.swift

@@ -13,12 +13,9 @@
 @_exported import Foundation // Clang module
 
 extension NSRange : Hashable {
-    public var hashValue: Int {
-#if arch(i386) || arch(arm)
-        return Int(bitPattern: (UInt(bitPattern: location) | (UInt(bitPattern: length) << 16)))
-#elseif arch(x86_64) || arch(arm64)
-        return Int(bitPattern: (UInt(bitPattern: location) | (UInt(bitPattern: length) << 32)))
-#endif
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(location)
+        hasher.combine(length)
     }
 
     public static func==(lhs: NSRange, rhs: NSRange) -> Bool {
@@ -230,4 +227,4 @@ extension NSRange : Codable {
         try container.encode(self.location)
         try container.encode(self.length)
     }
-}
+}

stdlib/public/Darwin/Foundation/NSStringEncodings.swift

@@ -51,8 +51,8 @@ extension String {
 }
 
 extension String.Encoding : Hashable {
-    public var hashValue : Int {
-        return rawValue.hashValue
+    public func hash(into hasher: inout Hasher) {
+        hasher.combine(rawValue)
     }
 
     public static func ==(lhs: String.Encoding, rhs: String.Encoding) -> Bool {