Add backtracking loops, backreferences, and if-then-else constructs to Regex "simplified" code gen

src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexNode.cs

@@ -240,6 +240,8 @@ private void MakeLoopAtomic()
         [Conditional("DEBUG")]
         private void ValidateFinalTreeInvariants()
         {
+            Debug.Assert(Type == Capture, "Every generated tree should begin with a capture node");
+
             var toExamine = new Stack<RegexNode>();
             toExamine.Push(this);
             while (toExamine.Count > 0)
@@ -306,8 +308,11 @@ private void ValidateFinalTreeInvariants()
                         break;
 
                     case Testref:
+                        Debug.Assert(childCount is 1 or 2, $"Expected one or two children for {node.TypeName}, got {childCount}");
+                        break;
+
                     case Testgroup:
-                        Debug.Assert(childCount >= 1, $"Expected at least one child for {node.TypeName}, got {childCount}.");
+                        Debug.Assert(childCount is 2 or 3, $"Expected two or three children for {node.TypeName}, got {childCount}");
                         break;
 
                     case Concatenate:
@@ -2229,9 +2234,18 @@ internal bool SupportsSimplifiedCodeGenerationImplementation()
                     case Empty:
                     case Nothing:
                     case UpdateBumpalong:
+                    // Backreferences are supported
+                    case Ref:
                         supported = true;
                         break;
 
+                    // Conditional backreference tests are also supported, so long as both their yes/no branches are supported.
+                    case Testref:
+                        supported =
+                            Child(0).SupportsSimplifiedCodeGenerationImplementation() &&
+                            (childCount == 1 || Child(1).SupportsSimplifiedCodeGenerationImplementation());
+                        break;
+
                     // Single character greedy/lazy loops are supported if either they're actually a repeater
                     // or they're not contained in any construct other than simple nesting (e.g. concat, capture).
                     case Oneloop:
@@ -2244,17 +2258,10 @@ internal bool SupportsSimplifiedCodeGenerationImplementation()
                         supported = M == N || AncestorsAllowBacktracking(Next);
                         break;
 
-                    // Loop repeaters are the same, except their child also needs to be supported.
-                    // We also support such loops being atomic.
+                    // For greedy and lazy loops, they're supported if the node they wrap is supported
+                    // and either the node is actually a repeater, is atomic, or is in the tree in a
+                    // location where backtracking is allowed.
                     case Loop:
-                        supported =
-                            (M == N || (Next != null && Next.Type == Atomic)) &&
-                            Child(0).SupportsSimplifiedCodeGenerationImplementation();
-                        break;
-
-                    // Similarly, as long as the wrapped node supports simplified code gen,
-                    // Lazy is supported if it's a repeater or atomic, but also if it's in
-                    // a place where backtracking is allowed (e.g. it's top-level).
                     case Lazyloop:
                         supported =
                             (M == N || (Next != null && Next.Type == Atomic) || AncestorsAllowBacktracking(Next)) &&
@@ -2297,11 +2304,10 @@ internal bool SupportsSimplifiedCodeGenerationImplementation()
                         break;
 
                     case Capture:
-                        // Currently we only support capnums without uncapnums (for balancing groups)
-                        supported = N == -1;
+                        supported = Child(0).SupportsSimplifiedCodeGenerationImplementation();
                         if (supported)
                         {
-                            // And we only support them in certain places in the tree.
+                            // Captures are currently only supported in certain places in the tree.
                             RegexNode? parent = Next;
                             while (parent != null)
                             {
@@ -2322,25 +2328,31 @@ internal bool SupportsSimplifiedCodeGenerationImplementation()
                                 }
                             }
 
+                            // If we've found a supported capture, mark all of the nodes in its parent
+                            // hierarchy as containing a capture.
                             if (supported)
                             {
-                                // And we only support them if their children are supported.
-                                supported = Child(0).SupportsSimplifiedCodeGenerationImplementation();
-
-                                // If we've found a supported capture, mark all of the nodes in its parent
-                                // hierarchy as containing a capture.
-                                if (supported)
+                                parent = this;
+                                while (parent != null && ((parent.Options & HasCapturesFlag) == 0))
                                 {
-                                    parent = this;
-                                    while (parent != null && ((parent.Options & HasCapturesFlag) == 0))
-                                    {
-                                        parent.Options |= HasCapturesFlag;
-                                        parent = parent.Next;
-                                    }
+                                    parent.Options |= HasCapturesFlag;
+                                    parent = parent.Next;
                                 }
                             }
                         }
                         break;
+
+                    case Testgroup:
+                        supported =
+                            Child(0).SupportsSimplifiedCodeGenerationImplementation() &&
+                            Child(1).SupportsSimplifiedCodeGenerationImplementation() &&
+                            (childCount == 2 || Child(2).SupportsSimplifiedCodeGenerationImplementation());
+                        break;
+
+                    default:
+                        Debug.Fail($"Unknown type: {Type}");
+                        supported = false;
+                        break;
                 }
             }
 #if DEBUG

src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexParser.cs

@@ -2071,6 +2071,17 @@ private bool IsCaptureSlot(int i)
             return i >= 0 && i < _capsize;
         }
 
+        /// <summary>
+        /// When generating code on a regex that uses a sparse set
+        /// of capture slots, we hash them to a dense set of indices
+        /// for an array of capture slots. Instead of doing the hash
+        /// at match time, it's done at compile time, here.
+        /// </summary>
+        internal static int MapCaptureNumber(int capnum, Hashtable? caps) =>
+            capnum == -1 ? -1 :
+            caps != null ? (int)caps[capnum]! :
+            capnum;
+
         /// <summary>Looks up the slot number for a given name</summary>
         private bool IsCaptureName(string capname) => _capnames != null && _capnames.ContainsKey(capname);
 
@@ -2171,7 +2182,7 @@ private void AddConcatenate(int pos, int cch, bool isReplacement)
                     _concatenation!.AddChild(RegexNode.CreateOneWithCaseConversion(_pattern[pos], isReplacement ? _options & ~RegexOptions.IgnoreCase : _options, _culture));
                     break;
 
-                case > 1 when !UseOptionI() || isReplacement:
+                case > 1 when !UseOptionI() || isReplacement || !RegexCharClass.ParticipatesInCaseConversion(_pattern.AsSpan(pos, cch)):
                     _concatenation!.AddChild(new RegexNode(RegexNode.Multi, _options & ~RegexOptions.IgnoreCase, _pattern.Substring(pos, cch)));
                     break;
 

src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/RegexWriter.cs

@@ -214,17 +214,6 @@ private int StringCode(string str)
             return i;
         }
 
-        /// <summary>
-        /// When generating code on a regex that uses a sparse set
-        /// of capture slots, we hash them to a dense set of indices
-        /// for an array of capture slots. Instead of doing the hash
-        /// at match time, it's done at compile time, here.
-        /// </summary>
-        private int MapCapnum(int capnum) =>
-            capnum == -1 ? -1 :
-            _caps != null ? (int)_caps[capnum]! :
-            capnum;
-
         /// <summary>
         /// The main RegexCode generator. It does a depth-first walk
         /// through the tree and calls EmitFragment to emits code before
@@ -283,7 +272,7 @@ private void EmitFragment(int nodetype, RegexNode node, int curIndex)
                             Emit(RegexCode.Setjump);
                             _intStack.Append(_emitted.Length);
                             Emit(RegexCode.Lazybranch, 0);
-                            Emit(RegexCode.Testref, MapCapnum(node.M));
+                            Emit(RegexCode.Testref, RegexParser.MapCaptureNumber(node.M, _caps));
                             Emit(RegexCode.Forejump);
                             break;
                     }
@@ -391,7 +380,7 @@ private void EmitFragment(int nodetype, RegexNode node, int curIndex)
                     break;
 
                 case RegexNode.Capture | AfterChild:
-                    Emit(RegexCode.Capturemark, MapCapnum(node.M), MapCapnum(node.N));
+                    Emit(RegexCode.Capturemark, RegexParser.MapCaptureNumber(node.M, _caps), RegexParser.MapCaptureNumber(node.N, _caps));
                     break;
 
                 case RegexNode.Require | BeforeChild:
@@ -471,7 +460,7 @@ private void EmitFragment(int nodetype, RegexNode node, int curIndex)
                     break;
 
                 case RegexNode.Ref:
-                    Emit(node.Type | bits, MapCapnum(node.M));
+                    Emit(node.Type | bits, RegexParser.MapCaptureNumber(node.M, _caps));
                     break;
 
                 case RegexNode.Nothing:

src/libraries/System.Text.RegularExpressions/tests/MonoRegexTests.cs

@@ -70,7 +70,10 @@ public static IEnumerable<object[]> ValidateRegex_MemberData()
 
                     (string Pattern, RegexOptions Options, string Input, string Expected) testCase = allEngineCases[i];
                     yield return new object[] { engine, testCase.Pattern, testCase.Options, results[i], testCase.Input, expected };
-                    yield return new object[] { engine, testCase.Pattern, testCase.Options | RegexOptions.CultureInvariant, results[i], testCase.Input, expected };
+                    if ((testCase.Options & RegexOptions.IgnoreCase) != 0)
+                    {
+                        yield return new object[] { engine, testCase.Pattern, testCase.Options | RegexOptions.CultureInvariant, results[i], testCase.Input, expected };
+                    }
                 }
             }
 

src/libraries/System.Text.RegularExpressions/tests/Regex.Groups.Tests.cs

@@ -619,6 +619,15 @@ public static IEnumerable<object[]> Groups_Basic_TestData()
                 yield return new object[] { engine, null, @"(.*)/(.+).aspx", "/pages/homepage.aspx/index.aspx", RegexOptions.None, new string[] { "/pages/homepage.aspx/index.aspx", "/pages/homepage.aspx", "index" } };
                 yield return new object[] { engine, null, @"(.*)/(.+)/(.+).aspx", "/pages/homepage.aspx/index.aspx", RegexOptions.None, new string[] { "/pages/homepage.aspx/index.aspx", "/pages", "homepage.aspx", "index" } };
 
+                // Captures inside varying constructs with backtracking needing to uncapture
+                yield return new object[] { engine, null, @"a(bc)d|abc(e)", "abce", RegexOptions.None, new string[] { "abce", "", "e" } }; // alternation
+                yield return new object[] { engine, null, @"((ab){2}cd)*", "ababcdababcdababc", RegexOptions.None, new string[] { "ababcdababcd", "ababcd", "ab" } }; // loop
+                yield return new object[] { engine, null, @"(ab(?=(\w)\w))*a", "aba", RegexOptions.None, new string[] { "a", "", "" } }; // positive lookahead in a loop
+                yield return new object[] { engine, null, @"(ab(?=(\w)\w))*a", "ababa", RegexOptions.None, new string[] { "aba", "ab", "a" } }; // positive lookahead in a loop
+                yield return new object[] { engine, null, @"(ab(?=(\w)\w))*a", "abababa", RegexOptions.None, new string[] { "ababa", "ab", "a" } }; // positive lookahead in a loop
+                yield return new object[] { engine, null, @"\w\w(?!(\d)\d)", "aa..", RegexOptions.None, new string[] { "aa", "" } }; // negative lookahead
+                yield return new object[] { engine, null, @"\w\w(?!(\d)\d)", "aa.3", RegexOptions.None, new string[] { "aa", "" } }; // negative lookahead
+
                 // Quantifiers
                 yield return new object[] { engine, null, @"a*", "", RegexOptions.None, new string[] { "" } };
                 yield return new object[] { engine, null, @"a*", "a", RegexOptions.None, new string[] { "a" } };

src/libraries/System.Text.RegularExpressions/tests/Regex.KnownPattern.Tests.cs

@@ -504,8 +504,6 @@ public async Task Docs_GroupingConstructs_NonbacktrackingSubexpressions(RegexEng
 
             Regex rBack = await RegexHelpers.GetRegexAsync(engine, @"(\w)\1+.\b");
             Regex rNoBack = await RegexHelpers.GetRegexAsync(engine, @"(?>(\w)\1+).\b");
-            string[] inputs = { "aaad", "aaaa" };
-
             Match back, noback;
 
             back = rBack.Match("cccd.");
@@ -1117,6 +1115,95 @@ public async Task Docs_Anchors_ContiguousMatches(RegexEngine engine)
                 Regex.Replace(Input, Pattern, m => string.Concat(m.Value.Reverse())));
         }
 
+        //
+        // Based on examples from https://blog.stevenlevithan.com/archives/balancing-groups
+        //
+
+        [Theory]
+        [MemberData(nameof(RegexHelpers.AvailableEngines_MemberData), MemberType = typeof(RegexHelpers))]
+        public async Task Blog_Levithan_BalancingGroups_Palindromes(RegexEngine engine)
+        {
+            if (RegexHelpers.IsNonBacktracking(engine))
+            {
+                // balancing groups not supported
+                return;
+            }
+
+            Regex r = await RegexHelpers.GetRegexAsync(engine, @"(?<N>.)+.?(?<-N>\k<N>)+(?(N)(?!))");
+
+            // Palindromes
+            Assert.All(new[]
+            {
+                "kayak",
+                "racecar",
+                "never odd or even",
+                "madam im adam"
+            }, p => Assert.True(r.IsMatch(p)));
+
+            // Non-Palindromes
+            Assert.All(new[]
+            {
+                "canoe",
+                "raceboat"
+            }, p => Assert.False(r.IsMatch(p)));
+        }
+
+        [Theory]
+        [MemberData(nameof(RegexHelpers.AvailableEngines_MemberData), MemberType = typeof(RegexHelpers))]
+        public async Task Blog_Levithan_BalancingGroups_MatchingParentheses(RegexEngine engine)
+        {
+            if (RegexHelpers.IsNonBacktracking(engine))
+            {
+                // balancing groups not supported
+                return;
+            }
+
+            Regex r = await RegexHelpers.GetRegexAsync(engine, @"^\(
+                                                                     (?>
+                                                                         [^()]+
+                                                                     |
+                                                                         \( (?<Depth>)
+                                                                     |
+                                                                         \) (?<-Depth>)
+                                                                     )*
+                                                                     (?(Depth)(?!))
+                                                                 \)$", RegexOptions.IgnorePatternWhitespace);
+
+            Assert.True(r.IsMatch("()"));
+            Assert.True(r.IsMatch("(a(b c(de(f(g)hijkl))mn))"));
+
+            Assert.False(r.IsMatch("("));
+            Assert.False(r.IsMatch(")"));
+            Assert.False(r.IsMatch("())"));
+            Assert.False(r.IsMatch("(()"));
+            Assert.False(r.IsMatch("(ab(cd)ef"));
+        }
+
+        [Theory]
+        [MemberData(nameof(RegexHelpers.AvailableEngines_MemberData), MemberType = typeof(RegexHelpers))]
+        public async Task Blog_Levithan_BalancingGroups_WordLengthIncreases(RegexEngine engine)
+        {
+            if (RegexHelpers.IsNonBacktracking(engine))
+            {
+                // balancing groups not supported
+                return;
+            }
+
+            Regex r = await RegexHelpers.GetRegexAsync(engine, @"^(?:
+                                                                     (?(A)\s|)
+                                                                     (?<B>)
+                                                                     (?<C-B>\w)+ (?(B)(?!))
+                                                                     (?:
+                                                                         \s
+                                                                         (?<C>)
+                                                                         (?<B-C>\w)+ (?(C)(?!))
+                                                                         (?<A>)
+                                                                     )?
+                                                                 )+ \b$", RegexOptions.IgnorePatternWhitespace);
+
+            Assert.True(r.IsMatch("a bc def ghij klmni"));
+            Assert.False(r.IsMatch("a bc def ghi klmn"));
+        }
 
         //
         // These patterns come from real-world customer usages

src/libraries/System.Text.RegularExpressions/tests/RegexGeneratorHelper.netcoreapp.cs

@@ -111,8 +111,8 @@ internal static async Task<Regex[]> SourceGenRegexAsync(
             if (generatorResults.Diagnostics.Length != 0)
             {
                 throw new ArgumentException(
-                    string.Join(Environment.NewLine, generatorResults.Diagnostics) + Environment.NewLine +
-                    string.Join(Environment.NewLine, generatorResults.GeneratedTrees.Select(t => NumberLines(t.ToString()))));
+                    string.Join(Environment.NewLine, generatorResults.GeneratedTrees.Select(t => NumberLines(t.ToString()))) + Environment.NewLine +
+                    string.Join(Environment.NewLine, generatorResults.Diagnostics));
             }
 
             // Compile the assembly to a stream
@@ -122,8 +122,8 @@ internal static async Task<Regex[]> SourceGenRegexAsync(
             if (!results.Success || results.Diagnostics.Length != 0)
             {
                 throw new ArgumentException(
-                    string.Join(Environment.NewLine, results.Diagnostics.Concat(generatorResults.Diagnostics)) + Environment.NewLine +
-                    string.Join(Environment.NewLine, generatorResults.GeneratedTrees.Select(t => NumberLines(t.ToString()))));
+                    string.Join(Environment.NewLine, generatorResults.GeneratedTrees.Select(t => NumberLines(t.ToString()))) + Environment.NewLine +
+                    string.Join(Environment.NewLine, results.Diagnostics.Concat(generatorResults.Diagnostics)));
             }
             dll.Position = 0;