Fix bug in word boundary caching (#769)

Author: milseman

Sources/_StringProcessing/Unicode/WordBreaking.swift

@@ -81,18 +81,18 @@ extension String {
     }
 
     if #available(SwiftStdlib 5.7, *) {
-      var indices: Set<String.Index> = []
+      if cache == nil {
+        cache = []
+      }
       var j = maxIndex ?? range.lowerBound
 
       while j < range.upperBound, j <= i {
-        indices.insert(j)
+        cache!.insert(j)
         j = _wordIndex(after: j)
       }
 
-      cache = indices
       maxIndex = j
-      
-      return indices.contains(i)
+      return cache!.contains(i)
     } else {
       return false
     }

Tests/RegexTests/MatchTests.swift

@@ -2381,7 +2381,18 @@ extension RegexTests {
     XCTAssertTrue("cafe".contains(caseInsensitiveRegex))
     XCTAssertTrue("CaFe".contains(caseInsensitiveRegex))
   }
-  
+
+  // https://github.com/swiftlang/swift-experimental-string-processing/issues/768
+  func testWordBoundaryCaching() throws {
+    // This will first find word boundaries up til the middle before failing,
+    // then it will find word boundaries til late in the string, then fail,
+    // and finally should succeed on a word boundary cached from the first
+    // attempt.
+    let input = "first second third fourth"
+    let regex = try Regex(#".*second\bX|.*third\bX|.*first\b"#)
+    XCTAssertTrue(input.contains(regex))
+  }
+
   // MARK: Character Semantics
 
   var eComposed: String { "é" }