[5.7] Implement atomic non-capturing groups

Sources/_RegexParser/Regex/Parse/Sema.swift

@@ -325,7 +325,8 @@ extension RegexValidator {
   func validateGroup(_ group: AST.Group) throws {
     let kind = group.kind
     switch kind.value {
-    case .capture, .namedCapture, .nonCapture, .lookahead, .negativeLookahead:
+    case .capture, .namedCapture, .nonCapture, .lookahead, .negativeLookahead,
+        .atomicNonCapturing:
       break
 
     case .balancedCapture:
@@ -336,9 +337,6 @@ extension RegexValidator {
       // We need to figure out how these interact with typed captures.
       throw error(.unsupported("branch reset group"), at: kind.location)
 
-    case .atomicNonCapturing:
-      throw error(.unsupported("atomic group"), at: kind.location)
-
     case .nonAtomicLookahead:
       throw error(.unsupported("non-atomic lookahead"), at: kind.location)
 

Sources/_StringProcessing/ByteCodeGen.swift

@@ -306,7 +306,7 @@ fileprivate extension Compiler.ByteCodeGen {
       save(restoringAt: success)
       save(restoringAt: intercept)
       <sub-pattern>    // failure restores at intercept
-      clearSavePoint   // remove intercept
+      clearThrough(intercept) // remove intercept and any leftovers from <sub-pattern>
       <if negative>:
         clearSavePoint // remove success
       fail             // positive->success, negative propagates
@@ -324,7 +324,7 @@ fileprivate extension Compiler.ByteCodeGen {
     builder.buildSave(success)
     builder.buildSave(intercept)
     try emitNode(child)
-    builder.buildClear()
+    builder.buildClearThrough(intercept)
     if !positive {
       builder.buildClear()
     }
@@ -339,6 +339,38 @@ fileprivate extension Compiler.ByteCodeGen {
     builder.label(success)
   }
 
+  mutating func emitAtomicNoncapturingGroup(
+    _ child: DSLTree.Node
+  ) throws {
+    /*
+      save(continuingAt: success)
+      save(restoringAt: intercept)
+      <sub-pattern>    // failure restores at intercept
+      clearThrough(intercept) // remove intercept and any leftovers from <sub-pattern>
+      fail             // ->success
+    intercept:
+      clearSavePoint   // remove success
+      fail             // propagate failure
+    success:
+      ...
+    */
+
+    let intercept = builder.makeAddress()
+    let success = builder.makeAddress()
+
+    builder.buildSaveAddress(success)
+    builder.buildSave(intercept)
+    try emitNode(child)
+    builder.buildClearThrough(intercept)
+    builder.buildFail()
+
+    builder.label(intercept)
+    builder.buildClear()
+    builder.buildFail()
+
+    builder.label(success)
+  }
+
   mutating func emitMatcher(
     _ matcher: @escaping _MatcherInterface
   ) -> ValueRegister {
@@ -384,6 +416,9 @@ fileprivate extension Compiler.ByteCodeGen {
       }
       options.apply(optionSequence)
       try emitNode(child)
+
+    case .atomicNonCapturing:
+      try emitAtomicNoncapturingGroup(child)
 
     default:
       // FIXME: Other kinds...

Sources/_StringProcessing/Engine/Instruction.swift

@@ -228,6 +228,13 @@ extension Instruction {
     /// Precondition: There is a save point to remove
     case clear
 
+    /// Remove save points up to and including the operand
+    ///
+    /// Operand: instruction address to look for
+    ///
+    /// Precondition: The operand is in the save point list
+    case clearThrough
+
     /// View the most recently saved point
     ///
     /// UNIMPLEMENTED

Sources/_StringProcessing/Engine/MEBuilder.swift

@@ -153,6 +153,10 @@ extension MEProgram.Builder {
   mutating func buildClear() {
     instructions.append(.init(.clear))
   }
+  mutating func buildClearThrough(_ t: AddressToken) {
+    instructions.append(.init(.clearThrough))
+    fixup(to: t)
+  }
   mutating func buildRestore() {
     instructions.append(.init(.restore))
   }
@@ -317,7 +321,7 @@ extension MEProgram.Builder {
       case .condBranchZeroElseDecrement:
         payload = .init(addr: addr, int: inst.payload.int)
 
-      case .branch, .save, .saveAddress, .call:
+      case .branch, .save, .saveAddress, .call, .clearThrough:
         payload = .init(addr: addr)
 
       case .splitSaving:

Sources/_StringProcessing/Engine/Processor.swift

@@ -239,6 +239,17 @@ extension Processor {
     }
   }
 
+  mutating func clearThrough(_ address: InstructionAddress) {
+    while let sp = savePoints.popLast() {
+      if sp.pc == address {
+        controller.step()
+        return
+      }
+    }
+    // TODO: What should we do here?
+    fatalError("Invalid code: Tried to clear save points when empty")
+  }
+
   mutating func cycle() {
     _checkInvariants()
     assert(state == .inProgress)
@@ -323,9 +334,13 @@ extension Processor {
       if let _ = savePoints.popLast() {
         controller.step()
       } else {
-        fatalError("TODO: What should we do here?")
+        // TODO: What should we do here?
+        fatalError("Invalid code: Tried to clear save points when empty")
       }
 
+    case .clearThrough:
+      clearThrough(payload.addr)
+
     case .peek:
       fatalError()
 

Tests/RegexBuilderTests/RegexDSLTests.swift

@@ -467,6 +467,29 @@ class RegexDSLTests: XCTestCase {
     XCTAssertEqual("ab12".firstMatch(of: octoDecimalRegex)!.output.1, 61904)
   }
 
+  func testLocal() throws {
+    try _testDSLCaptures(
+      ("aaaaa", nil),
+      matchType: Substring.self, ==)
+    {
+      Local {
+        OneOrMore("a")
+      }
+      "a"
+    }
+
+    try _testDSLCaptures(
+      ("aa", "aa"),
+      ("aaa", nil),
+      matchType: Substring.self, ==)
+    {
+      Local {
+        OneOrMore("a", .reluctant)
+      }
+      "a"
+    }
+  }
+
   func testAssertions() throws {
     try _testDSLCaptures(
       ("aaaaab", "aaaaab"),

Tests/RegexTests/MatchTests.swift

@@ -891,8 +891,7 @@ extension RegexTests {
       input: "Price: 100 dollars", match: nil)
     firstMatchTest(
       #"(?=\d+ dollars)\d+"#,
-      input: "Price: 100 dollars", match: "100",
-      xfail: true) // TODO
+      input: "Price: 100 dollars", match: "100")
 
     firstMatchTest(
       #"\d+(*pla: dollars)"#,
@@ -917,6 +916,14 @@ extension RegexTests {
       #"\d+(*negative_lookahead: dollars)"#,
       input: "Price: 100 pesos", match: "100")
 
+    // More complex lookaheads
+    firstMatchTests(
+      #"(?=.*e)(?=.*o)(?!.*z)."#,
+      (input: "hello", match: "h"),
+      (input: "hzello", match: "e"),
+      (input: "hezllo", match: nil),
+      (input: "helloz", match: nil))
+
     firstMatchTest(
       #"(?<=USD)\d+"#, input: "Price: USD100", match: "100", xfail: true)
     firstMatchTest(
@@ -1050,14 +1057,93 @@ extension RegexTests {
     firstMatchTest(
       #"(?:a|.b)c"#, input: "123abcacxyz", match: "abc")
     firstMatchTest(
-      #"(?>a|.b)c"#, input: "123abcacxyz", match: "ac", xfail: true)
+      #"(?>a|.b)c"#, input: "123abcacxyz", match: "ac")
     firstMatchTest(
-      "(*atomic:a|.b)c", input: "123abcacxyz", match: "ac", xfail: true)
+      "(*atomic:a|.b)c", input: "123abcacxyz", match: "ac")
     firstMatchTest(
       #"(?:a+)[a-z]c"#, input: "123aacacxyz", match: "aac")
     firstMatchTest(
-      #"(?>a+)[a-z]c"#, input: "123aacacxyz", match: "ac", xfail: true)
+      #"(?>a+)[a-z]c"#, input: "123aacacxyz", match: nil)
+
+    // Atomicity should stay in the atomic group
+    firstMatchTest(
+      #"(?:(?>a)|.b)c"#, input: "123abcacxyz", match: "abc")
+
+    // Quantifier behavior inside atomic groups
+
+    // (?:a+?) matches as few 'a's as possible, after matching the first
+    // (?>a+?) always matches exactly one 'a'
+    firstMatchTests(
+      #"^(?:a+?)a$"#,
+      (input: "a", match: nil),
+      (input: "aa", match: "aa"),
+      (input: "aaa", match:  "aaa"))
+    firstMatchTests(
+      #"^(?>a+?)a$"#,
+      (input: "a", match: nil),
+      (input: "aa", match: "aa"),
+      (input: "aaa", match:  nil))
+
+    // (?:a?+) and (?>a?+) are equivalent: they match one 'a' if available
+    firstMatchTests(
+      #"^(?:a?+)a$"#,
+      (input: "a", match: nil),
+      xfail: true)
+    firstMatchTests(
+      #"^(?:a?+)a$"#,
+      (input: "aa", match: "aa"),
+      (input: "aaa", match: nil))
+    firstMatchTests(
+      #"^(?>a?+)a$"#,
+      (input: "a", match: nil),
+      (input: "aa", match: "aa"),
+      (input: "aaa", match: nil))
 
+    // Capture behavior in non-atomic vs atomic groups
+    firstMatchTests(
+      #"(\d+)\w+\1"#,
+      (input: "123x12", match: "123x12"), // `\w+` matches "3x" in this case
+      (input: "23x23", match: "23x23"),
+      (input: "123x23", match: "23x23"))
+    firstMatchTests(
+      #"(?>(\d+))\w+\1"#,
+      (input: "123x12", match: nil))
+    firstMatchTests(
+      #"(?>(\d+))\w+\1"#,
+      (input: "23x23", match: "23x23"),
+      (input: "123x23", match: "23x23"),
+      xfail: true)
+
+    // Backreferences in lookaheads
+    firstMatchTests(
+      #"^(?=.*(.)(.)\2\1).+$"#,
+      (input: "abbba", match: nil),
+      (input: "ABBA", match: "ABBA"),
+      (input: "defABBAdef", match: "defABBAdef"))
+    firstMatchTests(
+      #"^(?=.*(.)(.)\2\1).+\2$"#,
+      (input: "abbba", match: nil),
+      (input: "ABBA", match: nil),
+      (input: "defABBAdef", match: nil))
+    // FIXME: Backreferences don't escape positive lookaheads
+    firstMatchTests(
+      #"^(?=.*(.)(.)\2\1).+\2$"#,
+      (input: "ABBAB", match: "ABBAB"),
+      (input: "defABBAdefB", match: "defABBAdefB"),
+      xfail: true)
+
+    firstMatchTests(
+      #"^(?!.*(.)(.)\2\1).+$"#,
+      (input: "abbba", match: "abbba"),
+      (input: "ABBA", match: nil),
+      (input: "defABBAdef", match: nil))
+    // Backreferences don't escape negative lookaheads;
+    // matching only proceeds when the lookahead fails
+    firstMatchTests(
+      #"^(?!.*(.)(.)\2\1).+\2$"#,
+      (input: "abbba", match: nil),
+      (input: "abbbab", match: nil),
+      (input: "ABBAB", match: nil))
 
     // TODO: Test example where non-atomic is significant
     firstMatchTest(

Tests/RegexTests/ParseTests.swift

@@ -979,10 +979,10 @@ extension RegexTests {
       concat("a", nonCaptureReset("b"), "c"), throwsError: .unsupported)
     parseTest(
       #"a(?>b)c"#,
-      concat("a", atomicNonCapturing("b"), "c"), throwsError: .unsupported)
+      concat("a", atomicNonCapturing("b"), "c"))
     parseTest(
       "a(*atomic:b)c",
-      concat("a", atomicNonCapturing("b"), "c"), throwsError: .unsupported)
+      concat("a", atomicNonCapturing("b"), "c"))
 
     parseTest("a(?=b)c", concat("a", lookahead("b"), "c"))
     parseTest("a(*pla:b)c", concat("a", lookahead("b"), "c"))