Merge pull request #14913 from lorentey/resilient-hashing

Switch to a resilient hashing interface, currently implementing SipHash-1-3.
Compiler-synthesized Hashable conformances still use the old _combineHashValues interface for now.

Author: lorentey

lib/SILGen/SILGenExpr.cpp

@@ -3613,8 +3613,7 @@ getOrCreateKeyPathEqualsAndHash(SILGenModule &SGM,
 
     SILValue hashCode;
 
-    // TODO: Combine hashes of the indexes. There isn't a great hash combining
-    // interface in the standard library to do this yet.
+    // TODO: Combine hashes of the indexes using an inout _Hasher
     {
       auto &index = indexes[0];
 

stdlib/private/StdlibUnittest/StdlibUnittest.swift.gyb

@@ -1089,6 +1089,15 @@ struct PersistentState {
   static var runNoTestsWasCalled: Bool = false
   static var ranSomething: Bool = false
   static var complaintInstalled = false
+  static var hashingKeyOverridden = false
+
+  static func overrideHashingKey() {
+    if !hashingKeyOverridden {
+      // FIXME(hasher): This has to run before creating the first Set/Dictionary
+      _Hasher._secretKey = (0, 0)
+      hashingKeyOverridden = true
+    }
+  }
 
   static func complainIfNothingRuns() {
     if !complaintInstalled {
@@ -1200,6 +1209,7 @@ func stopTrackingObjects(_: UnsafePointer<CChar>) -> Int
 
 public final class TestSuite {
   public init(_ name: String) {
+    PersistentState.overrideHashingKey()
     self.name = name
     _precondition(
       _testNameToIndex[name] == nil,

stdlib/public/core/AnyHashable.swift

@@ -48,6 +48,7 @@ internal protocol _AnyHashableBox {
   ///   no comparison is possible. Otherwise, contains the result of `==`.
   func _isEqual(to: _AnyHashableBox) -> Bool?
   var _hashValue: Int { get }
+  func _hash(_into hasher: inout _Hasher)
 
   var _base: Any { get }
   func _downCastConditional<T>(into result: UnsafeMutablePointer<T>) -> Bool
@@ -93,6 +94,12 @@ internal struct _ConcreteHashableBox<Base : Hashable> : _AnyHashableBox {
     return _baseHashable.hashValue
   }
 
+  @_inlineable // FIXME(sil-serialize-all)
+  @_versioned // FIXME(sil-serialize-all)
+  func _hash(_into hasher: inout _Hasher) {
+    _baseHashable._hash(into: &hasher)
+  }
+
   @_inlineable // FIXME(sil-serialize-all)
   @_versioned // FIXME(sil-serialize-all)
   internal var _base: Any {
@@ -295,6 +302,11 @@ extension AnyHashable : Hashable {
   public var hashValue: Int {
     return _box._hashValue
   }
+
+  @_inlineable // FIXME(sil-serialize-all)
+  public func _hash(into hasher: inout _Hasher) {
+    _box._hash(_into: &hasher)
+  }
 }
 
 extension AnyHashable : CustomStringConvertible {

stdlib/public/core/Bool.swift

@@ -154,9 +154,13 @@ extension Bool : Equatable, Hashable {
   ///   invocations of the same program. Do not persist the hash value across
   ///   program runs.
   @_inlineable // FIXME(sil-serialize-all)
-  @_transparent
   public var hashValue: Int {
-    return self ? 1 : 0
+    return _hashValue(for: self)
+  }
+
+  @_inlineable // FIXME(sil-serialize-all)
+  public func _hash(into hasher: inout _Hasher) {
+    hasher.append((self ? 1 : 0) as UInt8)
   }
 
   @_inlineable // FIXME(sil-serialize-all)

stdlib/public/core/CTypes.swift

@@ -182,7 +182,12 @@ extension OpaquePointer: Hashable {
   /// program runs.
   @_inlineable // FIXME(sil-serialize-all)
   public var hashValue: Int {
-    return Int(Builtin.ptrtoint_Word(_rawValue))
+    return _hashValue(for: self)
+  }
+
+  @_inlineable // FIXME(sil-serialize-all)
+  public func _hash(into hasher: inout _Hasher) {
+    hasher.append(Int(Builtin.ptrtoint_Word(_rawValue)))
   }
 }
 

stdlib/public/core/DoubleWidth.swift.gyb

@@ -151,10 +151,13 @@ extension DoubleWidth : Comparable {
 extension DoubleWidth : Hashable {
   @_inlineable // FIXME(sil-serialize-all)
   public var hashValue: Int {
-    var result = 0
-    result = _combineHashValues(result, _storage.high.hashValue)
-    result = _combineHashValues(result, _storage.low.hashValue)
-    return result
+    return _hashValue(for: self)
+  }
+
+  @_inlineable // FIXME(sil-serialize-all)
+  public func _hash(into hasher: inout _Hasher) {
+    hasher.append(low)
+    hasher.append(high)
   }
 }
 

stdlib/public/core/DropWhile.swift

@@ -190,6 +190,10 @@ extension LazyDropWhileCollection.Index: Hashable where Base.Index: Hashable {
   public var hashValue: Int {
     return base.hashValue
   }
+
+  public func _hash(into hasher: inout _Hasher) {
+    hasher.append(base)
+  }
 }
 
 extension LazyDropWhileCollection: Collection {

stdlib/public/core/Flatten.swift

@@ -233,7 +233,14 @@ extension FlattenCollection.Index : Comparable {
 extension FlattenCollection.Index : Hashable
   where Base.Index : Hashable, Base.Element.Index : Hashable {
   public var hashValue: Int {
-    return _combineHashValues(_inner?.hashValue ?? 0, _outer.hashValue)
+    return _hashValue(for: self)
+  }
+
+  public func _hash(into hasher: inout _Hasher) {
+    hasher.append(_outer)
+    if let inner = _inner {
+      hasher.append(inner)
+    }
   }
 }
 

stdlib/public/core/FloatingPointTypes.swift.gyb

@@ -1524,26 +1524,25 @@ extension ${Self} : Hashable {
   /// your program. Do not save hash values to use during a future execution.
   @_inlineable // FIXME(sil-serialize-all)
   public var hashValue: Int {
+    return _hashValue(for: self)
+  }
+
+  @_inlineable // FIXME(sil-serialize-all)
+  public func _hash(into hasher: inout _Hasher) {
+    var v = self
     if isZero {
       // To satisfy the axiom that equality implies hash equality, we need to
       // finesse the hash value of -0.0 to match +0.0.
-      return 0
-    } else {
-    %if bits <= word_bits:
-      return Int(bitPattern: UInt(bitPattern))
-    %elif bits == 64: # Double -> 32-bit Int
-      return Int(truncatingIfNeeded: bitPattern &>> 32) ^
-             Int(truncatingIfNeeded: bitPattern)
-    %elif word_bits == 32: # Float80 -> 32-bit Int
-      return Int(truncatingIfNeeded: significandBitPattern &>> 32) ^
-             Int(truncatingIfNeeded: significandBitPattern) ^
-             Int(_representation.signAndExponent)
-    %else: # Float80 -> 64-bit Int
-      return Int(bitPattern: UInt(significandBitPattern)) ^
-             Int(_representation.signAndExponent)
-    %end
+      v = 0
     }
+  %if bits == 80:
+    hasher.append(v._representation.signAndExponent)
+    hasher.append(v.significandBitPattern)
+  %else:
+    hasher.append(v.bitPattern)
+  %end
   }
+
 }
 
 extension ${Self} {

stdlib/public/core/Hashable.swift

@@ -108,6 +108,24 @@ public protocol Hashable : Equatable {
   /// Hash values are not guaranteed to be equal across different executions of
   /// your program. Do not save hash values to use during a future execution.
   var hashValue: Int { get }
+
+  /// Feed bits to be hashed into the hash function represented by `hasher`.
+  func _hash(into hasher: inout _Hasher)
+}
+
+extension Hashable {
+  @inline(__always)
+  public func _hash(into hasher: inout _Hasher) {
+    hasher.append(self.hashValue)
+  }
+}
+
+// Called by synthesized `hashValue` implementations.
+@inline(__always)
+public func _hashValue<H: Hashable>(for value: H) -> Int {
+  var hasher = _Hasher()
+  hasher.append(value)
+  return hasher.finalize()
 }
 
 // Called by the SwiftValue implementation.
@@ -126,4 +144,3 @@ internal func Hashable_hashValue_indirect<T : Hashable>(
 ) -> Int {
   return value.pointee.hashValue
 }
-

stdlib/public/core/HashedCollections.swift.gyb

@@ -792,11 +792,16 @@ extension Set : Hashable {
   @_inlineable // FIXME(sil-serialize-all)
   public var hashValue: Int {
     // FIXME(ABI)#177: <rdar://problem/18915294> Cache Set<T> hashValue
-    var result: Int = _mixInt(0)
+    return _hashValue(for: self)
+  }
+
+  @_inlineable // FIXME(sil-serialize-all)
+  public func _hash(into hasher: inout _Hasher) {
+    var hash = 0
     for member in self {
-       result ^= _mixInt(member.hashValue)
+      hash ^= _hashValue(for: member)
     }
-    return result
+    hasher.append(hash)
   }
 }
 
@@ -3985,7 +3990,7 @@ extension _Native${Self}Buffer
   @_versioned
   @inline(__always) // For performance reasons.
   internal func _bucket(_ k: Key) -> Int {
-    return _squeezeHashValue(k.hashValue, bucketCount)
+    return _hashValue(for: k) & _bucketMask
   }
 
   @_inlineable // FIXME(sil-serialize-all)