Enable Regex to use SearchValues<string> in compiled / source generator for IgnoreCase multi-strings

src/libraries/System.Text.RegularExpressions/gen/RegexGenerator.Emitter.cs

@@ -732,6 +732,11 @@ private static void EmitTryFindNextPossibleStartingPosition(IndentedTextWriter w
                             EmitIndexOfString_RightToLeft();
                             break;
 
+                        case FindNextStartingPositionMode.LeadingStrings_LeftToRight:
+                        case FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight:
+                            EmitIndexOfStrings_LeftToRight();
+                            break;
+
                         case FindNextStartingPositionMode.LeadingSet_LeftToRight:
                         case FindNextStartingPositionMode.FixedDistanceSets_LeftToRight:
                             EmitFixedSet_LeftToRight();
@@ -1041,6 +1046,37 @@ UnicodeCategory.NonSpacingMark or
                 }
             }
 
+            // Emits a case-sensitive left-to-right search for any one of multiple leading prefixes.
+            void EmitIndexOfStrings_LeftToRight()
+            {
+                RegexFindOptimizations opts = regexTree.FindOptimizations;
+                Debug.Assert(opts.FindMode is FindNextStartingPositionMode.LeadingStrings_LeftToRight or FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight);
+
+                string prefixes = string.Join(", ", opts.LeadingPrefixes.Select(prefix => Literal(prefix)));
+                StringComparison stringComparison = opts.FindMode is FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight ?
+                    StringComparison.OrdinalIgnoreCase :
+                    StringComparison.Ordinal;
+                string fieldName = GetSHA256FieldName($"s_indexOfAnyStrings_{stringComparison}_", prefixes);
+
+                if (!requiredHelpers.ContainsKey(fieldName))
+                {
+                    requiredHelpers.Add(fieldName,
+                    [
+                        $"/// <summary>Supports searching for the specified strings.</summary>",
+                        $"internal static readonly SearchValues<string> {fieldName} = SearchValues.Create([{prefixes}], StringComparison.{stringComparison});", // explicitly using an array in case prefixes is large
+                    ]);
+                }
+
+                writer.WriteLine($"// The pattern has multiple strings that could begin the match. Search for any of them.");
+                writer.WriteLine($"// If none can be found, there's no match.");
+                writer.WriteLine($"int i = inputSpan.Slice(pos).IndexOfAny({HelpersTypeName}.{fieldName});");
+                using (EmitBlock(writer, "if (i >= 0)"))
+                {
+                    writer.WriteLine("base.runtextpos = pos + i;");
+                    writer.WriteLine("return true;");
+                }
+            }
+
             // Emits a case-sensitive right-to-left search for a substring.
             void EmitIndexOfString_RightToLeft()
             {

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

@@ -1054,6 +1054,21 @@ public static bool IsAscii(ReadOnlySpan<char> s)
 #endif
         }
 
+        /// <summary>Gets whether the set description string is for two ASCII letters that case to each other under OrdinalIgnoreCase rules.</summary>
+        public static bool SetContainsAsciiOrdinalIgnoreCaseCharacter(string set, Span<char> twoChars)
+        {
+            Debug.Assert(twoChars.Length >= 2);
+            return
+                !IsNegated(set) &&
+                GetSetChars(set, twoChars) == 2 &&
+                twoChars[0] < 128 &&
+                twoChars[1] < 128 &&
+                twoChars[0] != twoChars[1] &&
+                char.IsLetter(twoChars[0]) &&
+                char.IsLetter(twoChars[1]) &&
+                (twoChars[0] | 0x20) == (twoChars[1] | 0x20);
+        }
+
         /// <summary>Gets whether we can iterate through the set list pairs in order to completely enumerate the set's contents.</summary>
         /// <remarks>This may enumerate negated characters if the set is negated.  This will return false if the set has subtraction.</remarks>
         private static bool CanEasilyEnumerateSetContents(string set) =>

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

@@ -460,6 +460,8 @@ protected void EmitTryFindNextPossibleStartingPosition()
             {
                 case FindNextStartingPositionMode.LeadingString_LeftToRight:
                 case FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight:
+                case FindNextStartingPositionMode.LeadingStrings_LeftToRight:
+                case FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight:
                 case FindNextStartingPositionMode.FixedDistanceString_LeftToRight:
                     EmitIndexOfString_LeftToRight();
                     break;
@@ -745,15 +747,19 @@ bool EmitAnchors()
                 return false;
             }
 
-            // Emits a case-sensitive left-to-right search for a substring.
+            // Emits a case-sensitive left-to-right search for a substring or substrings.
             void EmitIndexOfString_LeftToRight()
             {
                 RegexFindOptimizations opts = _regexTree.FindOptimizations;
-                Debug.Assert(opts.FindMode is FindNextStartingPositionMode.LeadingString_LeftToRight or FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight or FindNextStartingPositionMode.FixedDistanceString_LeftToRight);
+                Debug.Assert(opts.FindMode is FindNextStartingPositionMode.LeadingString_LeftToRight or
+                                              FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight or
+                                              FindNextStartingPositionMode.FixedDistanceString_LeftToRight or
+                                              FindNextStartingPositionMode.LeadingStrings_LeftToRight or
+                                              FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight);
 
                 using RentedLocalBuilder i = RentInt32Local();
 
-                // int i = inputSpan.Slice(pos).IndexOf(prefix);
+                // int i = inputSpan.Slice(pos)...
                 Ldloca(inputSpan);
                 Ldloc(pos);
                 if (opts.FindMode is FindNextStartingPositionMode.FixedDistanceString_LeftToRight &&
@@ -763,11 +769,21 @@ void EmitIndexOfString_LeftToRight()
                     Add();
                 }
                 Call(s_spanSliceIntMethod);
-                string literalString = opts.FindMode is FindNextStartingPositionMode.LeadingString_LeftToRight or FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight ?
-                    opts.LeadingPrefix :
-                    opts.FixedDistanceLiteral.String!;
-                LoadSearchValues([literalString], opts.FindMode is FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight ? StringComparison.OrdinalIgnoreCase : StringComparison.Ordinal);
-                Call(s_spanIndexOfAnySearchValuesString);
+
+                // ...IndexOf(prefix);
+                if (opts.FindMode is FindNextStartingPositionMode.LeadingStrings_LeftToRight or FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight)
+                {
+                    LoadSearchValues(opts.LeadingPrefixes, opts.FindMode is FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight ? StringComparison.OrdinalIgnoreCase : StringComparison.Ordinal);
+                    Call(s_spanIndexOfAnySearchValuesString);
+                }
+                else
+                {
+                    string literalString = opts.FindMode is FindNextStartingPositionMode.LeadingString_LeftToRight or FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight ?
+                        opts.LeadingPrefix :
+                        opts.FixedDistanceLiteral.String!;
+                    LoadSearchValues([literalString], opts.FindMode is FindNextStartingPositionMode.LeadingString_OrdinalIgnoreCase_LeftToRight ? StringComparison.OrdinalIgnoreCase : StringComparison.Ordinal);
+                    Call(s_spanIndexOfAnySearchValuesString);
+                }
                 Stloc(i);
 
                 // if (i < 0) goto ReturnFalse;

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

@@ -137,7 +137,28 @@ public RegexFindOptimizations(RegexNode root, RegexOptions options)
                 return;
             }
 
-            // We're now left-to-right only and looking for sets.
+            // We're now left-to-right only and looking for multiple prefixes and/or sets.
+
+            // If there are multiple leading strings, we can search for any of them.
+            if (compiled)
+            {
+                if (RegexPrefixAnalyzer.FindPrefixes(root, ignoreCase: true) is { Length: > 1 } caseInsensitivePrefixes)
+                {
+                    LeadingPrefixes = caseInsensitivePrefixes;
+                    FindMode = FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight;
+                    return;
+                }
+
+                // TODO: While some benchmarks benefit from this significantly, others regressed a bit (in particular those with few
+                //       matches). Before enabling this, we need to investigate the performance impact on real-world scenarios,
+                //       and see if there are ways to reduce the impact.
+                //if (RegexPrefixAnalyzer.FindPrefixes(root, ignoreCase: false) is { Length: > 1 } caseSensitivePrefixes)
+                //{
+                //    LeadingPrefixes = caseSensitivePrefixes;
+                //    FindMode = FindNextStartingPositionMode.LeadingStrings_LeftToRight;
+                //    return;
+                //}
+            }
 
             // Build up a list of all of the sets that are a fixed distance from the start of the expression.
             List<FixedDistanceSet>? fixedDistanceSets = RegexPrefixAnalyzer.FindFixedDistanceSets(root, thorough: !interpreter);
@@ -244,6 +265,9 @@ public RegexFindOptimizations(RegexNode root, RegexOptions options)
         /// <summary>Gets the leading prefix.  May be an empty string.</summary>
         public string LeadingPrefix { get; } = string.Empty;
 
+        /// <summary>Gets the leading prefixes.  May be an empty array.</summary>
+        public string[] LeadingPrefixes { get; } = Array.Empty<string>();
+
         /// <summary>When in fixed distance literal mode, gets the literal and how far it is from the start of the pattern.</summary>
         public (char Char, string? String, int Distance) FixedDistanceLiteral { get; }
 
@@ -767,10 +791,16 @@ public bool TryFindNextStartingPositionLeftToRight(ReadOnlySpan<char> textSpan,
                         return false;
                     }
 
+                // Not supported in the interpreter, but we could end up here for patterns so complex the compiler gave up on them.
+
+                case FindNextStartingPositionMode.LeadingStrings_LeftToRight:
+                case FindNextStartingPositionMode.LeadingStrings_OrdinalIgnoreCase_LeftToRight:
+                    return true;
+
                 // Nothing special to look for.  Just return true indicating this is a valid position to try to match.
 
                 default:
-                    Debug.Assert(FindMode == FindNextStartingPositionMode.NoSearch);
+                    Debug.Assert(FindMode == FindNextStartingPositionMode.NoSearch, $"Unexpected FindMode {FindMode}");
                     return true;
             }
         }
@@ -810,6 +840,11 @@ internal enum FindNextStartingPositionMode
         /// <summary>A multi-character ordinal case-insensitive substring at the beginning of the pattern.</summary>
         LeadingString_OrdinalIgnoreCase_LeftToRight,
 
+        /// <summary>Multiple leading prefix strings</summary>
+        LeadingStrings_LeftToRight,
+        /// <summary>Multiple leading ordinal case-insensitive prefix strings</summary>
+        LeadingStrings_OrdinalIgnoreCase_LeftToRight,
+
         /// <summary>A set starting the pattern.</summary>
         LeadingSet_LeftToRight,
         /// <summary>A set starting the right-to-left pattern.</summary>

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

@@ -2561,14 +2561,7 @@ public bool TryGetOrdinalCaseInsensitiveString(int childIndex, int exclusiveChil
                 {
                     // In particular we want to look for sets that contain only the upper and lowercase variant
                     // of the same ASCII letter.
-                    if (RegexCharClass.IsNegated(child.Str!) ||
-                        RegexCharClass.GetSetChars(child.Str!, twoChars) != 2 ||
-                        twoChars[0] >= 128 ||
-                        twoChars[1] >= 128 ||
-                        twoChars[0] == twoChars[1] ||
-                        !char.IsLetter(twoChars[0]) ||
-                        !char.IsLetter(twoChars[1]) ||
-                        ((twoChars[0] | 0x20) != (twoChars[1] | 0x20)))
+                    if (!RegexCharClass.SetContainsAsciiOrdinalIgnoreCaseCharacter(child.Str!, twoChars))
                     {
                         break;
                     }