Improve performance for comparing AttributedStrings with differing character counts (swiftlang#1224)

* Improve performance for comparing AttributedStrings with differing character counts

* Address feedback

* Fix typo

Author: jmschonfeld

Benchmarks/Benchmarks/AttributedString/BenchmarkAttributedString.swift

@@ -414,6 +414,11 @@ let benchmarks = {
 #endif
 
     let manyAttributesString2 = createManyAttributesString()
+    let manyAttributesString3 =  {
+        var str = createManyAttributesString()
+        str.characters.append("a")
+        return str
+    }()
     let manyAttributesStringRange = manyAttributesString.characters.index(manyAttributesString.startIndex, offsetBy: manyAttributesString.characters.count / 2)...
     let manyAttributesSubstring = manyAttributesString[manyAttributesStringRange]
     let manyAttributes2Substring = manyAttributesString2[manyAttributesStringRange]
@@ -422,6 +427,10 @@ let benchmarks = {
         blackHole(manyAttributesString == manyAttributesString2)
     }
 
+    Benchmark("equalityDifferingCharacters") { benchmark in
+        blackHole(manyAttributesString == manyAttributesString3)
+    }
+    
     Benchmark("substringEquality") { benchmark in
         blackHole(manyAttributesSubstring == manyAttributes2Substring)
     }

Sources/FoundationEssentials/AttributedString/AttributedString+Guts.swift

@@ -110,6 +110,27 @@ extension AttributedString.Guts {
 
         guard left.count == right.count else { return false }
         guard !left.isEmpty else { return true }
+        
+        if !left._isPartial && !right._isPartial {
+            // For a full BigString, we can get the grapheme cluster count in constant time since
+            // the grapheme cluster count is cached at the node level in the tree. It is not
+            // possible for two AttributedStrings with differing character counts to be equal,
+            // so bail early if we detect that
+            //
+            // Note: we should not perform this check for cases where we are not knowingly working
+            // with the full string. Since character counts are only cached at the node level,
+            // to get the character count of a substring you would need to run the grapheme
+            // breaking algorithm over the partial first and last chunks. While this is
+            // technically done in constant time as chunks have a max of 255 UTF-8 scalars, grapheme
+            // breaking up to 510 UTF-8 scalars would not be cheap. In the future we can
+            // investigate best effort short cuts by comparing the counts of just the "middle"
+            // chunks that we can determine cheaply along with the knowledge that the partial
+            // first and last chunks have a character count no more than their UTF-8 counts.
+            guard left._guts.string.count == right._guts.string.count else {
+                return false
+            }
+        }
+        
 
         guard var leftIndex = left._strBounds.ranges.first?.lowerBound, var rightIndex = right._strBounds.ranges.first?.lowerBound else { return false }
 

Sources/FoundationEssentials/AttributedString/AttributedString+Runs.swift

@@ -31,6 +31,16 @@ extension AttributedString {
         internal let _strBounds: RangeSet<BigString.Index>
         internal let _isDiscontiguous: Bool
 
+        internal var _isPartial: Bool {
+            guard !_isDiscontiguous else {
+                return true
+            }
+            guard let lower = _bounds.lowerBound._stringIndex, let upper = _bounds.upperBound._stringIndex else {
+                preconditionFailure("AttributedString.Runs created with bounds that have un-set string indices")
+            }
+            return _guts.string.startIndex != lower || _guts.string.endIndex != upper
+        }
+        
         internal init(_ guts: Guts, in bounds: Range<BigString.Index>) {
             self.init(guts, in: RangeSet(bounds))
         }