Speed up KeyAnalyzer for substring based frozen collections

src/libraries/System.Collections.Immutable/src/System.Collections.Immutable.csproj

@@ -10,6 +10,8 @@ The System.Collections.Immutable library is built-in as part of the shared frame
 
   <ItemGroup>
     <Compile Include="Properties\InternalsVisibleTo.cs" />
+    <Compile Include="System\Collections\Frozen\String\SubstringEquality\SubstringEqualityComparers.cs" />
+    <Compile Include="System\Collections\Frozen\String\SubstringEquality\SubstringEqualityComparerBase.cs" />
 
     <Compile Include="System\Polyfills.cs" />
     <Compile Include="System\Collections\ThrowHelper.cs" />

src/libraries/System.Collections.Immutable/src/System/Collections/Frozen/String/KeyAnalyzer.cs

@@ -2,11 +2,10 @@
 // The .NET Foundation licenses this file to you under the MIT license.
 
 using System.Buffers;
+using System.Collections.Frozen.String.SubstringEquality;
 using System.Collections.Generic;
 using System.Diagnostics;
-#if !NET8_0_OR_GREATER
 using System.Runtime.CompilerServices;
-#endif
 
 namespace System.Collections.Frozen
 {
@@ -33,84 +32,83 @@ public static AnalysisResults Analyze(
         {
             Debug.Assert(!uniqueStrings.IsEmpty);
 
-            // Try to pick a substring comparer. If we can't find a good substring comparer, fallback to a full string comparer.
-            AnalysisResults results;
-            if (minLength == 0 || !TryUseSubstring(uniqueStrings, ignoreCase, minLength, maxLength, out results))
+            if (minLength > 0)
             {
-                results = CreateAnalysisResults(uniqueStrings, ignoreCase, minLength, maxLength, 0, 0, static (s, _, _) => s.AsSpan());
-            }
+                const int MaxSubstringLengthLimit = 8; // arbitrary small-ish limit...it's not worth the increase in algorithmic complexity to analyze longer substrings
+                int uniqueStringsLength = uniqueStrings.Length;
 
-            return results;
-        }
-
-        /// <summary>Try to find the minimal unique substring index/length to use for comparisons.</summary>
-        private static bool TryUseSubstring(ReadOnlySpan<string> uniqueStrings, bool ignoreCase, int minLength, int maxLength, out AnalysisResults results)
-        {
-            const int MaxSubstringLengthLimit = 8; // arbitrary small-ish limit... t's not worth the increase in algorithmic complexity to analyze longer substrings
+                // Sufficient uniqueness factor of 95% is good enough.
+                // Instead of ensuring that 95% of data is good, we stop when we know that at least 5% is bad.
+                int acceptableNonUniqueCount = uniqueStringsLength / 20;
 
-            SubstringComparer comparer = ignoreCase ? new JustifiedCaseInsensitiveSubstringComparer() : new JustifiedSubstringComparer();
-            HashSet<string> set = new HashSet<string>(
-#if NET6_0_OR_GREATER
-                uniqueStrings.Length,
-#endif
-                comparer);
+                ISubstringEqualityComparer leftComparer = ignoreCase ? new LeftSubstringCaseInsensitiveComparer() : new LeftSubstringOrdinalComparer();
+                HashSet<string> leftSet = MakeHashSet(uniqueStringsLength, leftComparer);
 
-            // For each substring length...
-            int maxSubstringLength = Math.Min(minLength, MaxSubstringLengthLimit);
-            for (int count = 1; count <= maxSubstringLength; count++)
-            {
-                comparer.IsLeft = true;
-                comparer.Count = count;
+                // we lazily spin up the right comparators when/if needed
+                ISubstringEqualityComparer? rightComparer = null;
+                HashSet<string>? rightSet = null;
 
-                // For each index, get a uniqueness factor for the left-justified substrings.
-                // If any is above our threshold, we're done.
-                for (int index = 0; index <= minLength - count; index++)
+                // For each substring length...preferring the shortest length that provides
+                // enough uniqueness
+                int maxSubstringLength = Math.Min(minLength, MaxSubstringLengthLimit);
+                for (int count = 1; count <= maxSubstringLength; count++)
                 {
-                    comparer.Index = index;
+                    leftComparer.Count = count;
 
-                    if (HasSufficientUniquenessFactor(set, uniqueStrings))
+                    // For each index, get a uniqueness factor for the left-justified substrings.
+                    // If any is above our threshold, we're done.
+                    for (int index = 0; index <= minLength - count; index++)
                     {
-                        results = CreateAnalysisResults(
-                            uniqueStrings, ignoreCase, minLength, maxLength, index, count,
-                            static (string s, int index, int count) => s.AsSpan(index, count));
-                        return true;
-                    }
-                }
+                        leftComparer.Index = index;
 
-                // There were no left-justified substrings of this length available.
-                // If all of the strings are of the same length, then just checking left-justification is sufficient.
-                // But if any strings are of different lengths, then we'll get different alignments for left- vs
-                // right-justified substrings, and so we also check right-justification.
-                if (minLength != maxLength)
-                {
-                    // toggle the direction and re-use the comparer and hashset (HasSufficientUniquenessFactor clears it)
-                    comparer.IsLeft = false;
+                        if (HasSufficientUniquenessFactor(leftSet, uniqueStrings, acceptableNonUniqueCount))
+                        {
+                            return CreateAnalysisResults(uniqueStrings, ignoreCase, minLength, maxLength, leftComparer);
+                        }
+                    }
 
-                    // For each index, get a uniqueness factor for the right-justified substrings.
-                    // If any is above our threshold, we're done.
-                    for (int index = 0; index <= minLength - count; index++)
+                    // There were no left-justified substrings of this length available.
+                    // If all of the strings are of the same length, then just checking left-justification is sufficient.
+                    // But if any strings are of different lengths, then we'll get different alignments for left- vs
+                    // right-justified substrings, and so we also check right-justification.
+                    if (minLength != maxLength)
                     {
-                        // Get a uniqueness factor for the right-justified substrings.
-                        // If it's above our threshold, we're done.
-                        comparer.Index = -index - count;
-                        if (HasSufficientUniquenessFactor(set, uniqueStrings))
+                        rightComparer ??= ignoreCase ? new RightSubstringCaseInsensitiveComparer() : new RightSubstringOrdinalComparer();
+                        rightSet ??= MakeHashSet(uniqueStringsLength, rightComparer);
+
+                        // when Index is negative, we're offsetting from the right, ensure we're at least
+                        // far enough from the right that we have count characters available
+                        rightComparer!.Count = count;
+                        rightComparer!.Index = -count;
+
+                        // For each index, get a uniqueness factor for the right-justified substrings.
+                        // If any is above our threshold, we're done.
+                        for (int offset = 0; offset <= minLength - count; offset++, rightComparer!.Index--)
                         {
-                            results = CreateAnalysisResults(
-                                uniqueStrings, ignoreCase, minLength, maxLength, comparer.Index, count,
-                                static (string s, int index, int count) => s.AsSpan(s.Length + index, count));
-                            return true;
+                            if (HasSufficientUniquenessFactor(rightSet!, uniqueStrings, acceptableNonUniqueCount))
+                            {
+                                return CreateAnalysisResults(uniqueStrings, ignoreCase, minLength, maxLength, rightComparer);
+                            }
                         }
                     }
                 }
             }
 
-            // Could not find a substring index/length that was good enough.
-            results = default;
-            return false;
+            // Could not find a substring index/length that was good enough, use the entire string.
+            return CreateAnalysisResults(uniqueStrings, ignoreCase, minLength, maxLength, s_FullComparer);
+        }
+
+        private static HashSet<string> MakeHashSet(int length, IEqualityComparer<string> comparer)
+        {
+            return new HashSet<string>(
+#if NET6_0_OR_GREATER
+                    length,
+#endif
+                    comparer);
         }
 
         private static AnalysisResults CreateAnalysisResults(
-            ReadOnlySpan<string> uniqueStrings, bool ignoreCase, int minLength, int maxLength, int index, int count, GetSpan getSubstringSpan)
+            ReadOnlySpan<string> uniqueStrings, bool ignoreCase, int minLength, int maxLength, ISubstringEqualityComparer comparer)
         {
             // Start off by assuming all strings are ASCII
             bool allAsciiIfIgnoreCase = true;
@@ -129,7 +127,7 @@ private static AnalysisResults CreateAnalysisResults(
                 foreach (string s in uniqueStrings)
                 {
                     // Get the span for the substring.
-                    ReadOnlySpan<char> substring = getSubstringSpan(s, index, count);
+                    ReadOnlySpan<char> substring = comparer.Slice(s);
 
                     // If the substring isn't ASCII, bail out to return the results.
                     if (!IsAllAscii(substring))
@@ -155,11 +153,9 @@ private static AnalysisResults CreateAnalysisResults(
             }
 
             // Return the analysis results.
-            return new AnalysisResults(ignoreCase, allAsciiIfIgnoreCase, index, count, minLength, maxLength);
+            return new AnalysisResults(ignoreCase, allAsciiIfIgnoreCase, comparer.Index, comparer.Count, minLength, maxLength);
         }
 
-        private delegate ReadOnlySpan<char> GetSpan(string s, int index, int count);
-
         internal static unsafe bool IsAllAscii(ReadOnlySpan<char> s)
         {
 #if NET8_0_OR_GREATER
@@ -202,7 +198,7 @@ internal static unsafe bool IsAllAscii(ReadOnlySpan<char> s)
 #if NET8_0_OR_GREATER
         private static readonly SearchValues<char> s_asciiLetters = SearchValues.Create("ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz");
 #endif
-        private static bool ContainsAnyLetters(ReadOnlySpan<char> s)
+        internal static bool ContainsAnyLetters(ReadOnlySpan<char> s)
         {
             Debug.Assert(IsAllAscii(s));
 
@@ -221,18 +217,13 @@ private static bool ContainsAnyLetters(ReadOnlySpan<char> s)
 #endif
         }
 
-        private static bool HasSufficientUniquenessFactor(HashSet<string> set, ReadOnlySpan<string> uniqueStrings)
+        internal static bool HasSufficientUniquenessFactor(HashSet<string> set, ReadOnlySpan<string> uniqueStrings, int acceptableNonUniqueCount)
         {
-            set.Clear();
-
-            // Sufficient uniqueness factor of 95% is good enough.
-            // Instead of ensuring that 95% of data is good, we stop when we know that at least 5% is bad.
-            int acceptableNonUniqueCount = uniqueStrings.Length / 20;
-
             foreach (string s in uniqueStrings)
             {
                 if (!set.Add(s) && --acceptableNonUniqueCount < 0)
                 {
+                    set.Clear();
                     return false;
                 }
             }
@@ -263,25 +254,6 @@ public AnalysisResults(bool ignoreCase, bool allAsciiIfIgnoreCase, int hashIndex
             public bool RightJustifiedSubstring => HashIndex < 0;
         }
 
-        private abstract class SubstringComparer : IEqualityComparer<string>
-        {
-            public int Index;
-            public int Count;
-            public bool IsLeft;
-            public abstract bool Equals(string? x, string? y);
-            public abstract int GetHashCode(string s);
-        }
-
-        private sealed class JustifiedSubstringComparer : SubstringComparer
-        {
-            public override bool Equals(string? x, string? y) => x.AsSpan(IsLeft ? Index : (x!.Length + Index), Count).SequenceEqual(y.AsSpan(IsLeft ? Index : (y!.Length + Index), Count));
-            public override int GetHashCode(string s) => Hashing.GetHashCodeOrdinal(s.AsSpan(IsLeft ? Index : (s.Length + Index), Count));
-        }
-
-        private sealed class JustifiedCaseInsensitiveSubstringComparer : SubstringComparer
-        {
-            public override bool Equals(string? x, string? y) => x.AsSpan(IsLeft ? Index : (x!.Length + Index), Count).Equals(y.AsSpan(IsLeft ? Index : (y!.Length + Index), Count), StringComparison.OrdinalIgnoreCase);
-            public override int GetHashCode(string s) => Hashing.GetHashCodeOrdinalIgnoreCase(s.AsSpan(IsLeft ? Index : (s.Length + Index), Count));
-        }
+        private static FullStringEqualityComparer s_FullComparer = new FullStringEqualityComparer();
     }
 }

src/libraries/System.Collections.Immutable/src/System/Collections/Frozen/String/SubstringEquality/SubstringEqualityComparerBase.cs

@@ -0,0 +1,76 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System.Collections.Generic;
+using System.Runtime.CompilerServices;
+
+namespace System.Collections.Frozen.String.SubstringEquality
+{
+    internal interface ISubstringEqualityComparer : IEqualityComparer<string>
+    {
+        /// <summary>
+        /// The index at which to begin this slice
+        /// </summary>
+        /// <remarks>Offset from the left side (if zero or positive) or right side (if negative)</remarks>
+        public int Index { get; set; }
+
+        /// <summary>
+        /// The desired length for the slice (exclusive).
+        /// </summary>
+        public int Count { get; set; }
+
+        /// <summary>
+        /// Creates a new readonly span over the portion of the target string.
+        /// </summary>
+        /// <param name="s">The target string.</param>
+        /// <exception cref="ArgumentNullException"><paramref name="s"/> is null.</exception>
+        /// <exception cref="ArgumentOutOfRangeException">
+        /// Thrown when the specified Index or Count is not in range.
+        /// </exception>
+        public abstract ReadOnlySpan<char> Slice(string s);
+    }
+
+    internal abstract class SubstringEqualityComparerBase<TThisWrapper> : ISubstringEqualityComparer
+    where TThisWrapper : struct, SubstringEqualityComparerBase<TThisWrapper>.IGenericSpecializedWrapper
+    {
+        /// <summary>A wrapper around this that enables access to important members without making virtual calls.</summary>
+        private readonly TThisWrapper _this;
+
+        protected SubstringEqualityComparerBase()
+        {
+            _this = default;
+            _this.Store(this);
+        }
+
+        /// <inheritdoc />
+        public int Index { get; set; }
+        /// <inheritdoc />
+        public int Count { get; set; }
+
+        /// <inheritdoc />
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public ReadOnlySpan<char> Slice(string s) => _this.Slice(s);
+
+        /// <inheritdoc />
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public bool Equals(string? x, string? y) => _this.Equals(x, y);
+
+        /// <inheritdoc />
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public int GetHashCode(string s) => _this.GetHashCode(s);
+
+        /// <summary>Used to enable generic specialization with reference types.</summary>
+        /// <remarks>
+        /// To avoid each of those incurring virtual dispatch to the derived type, the derived
+        /// type hands down a struct wrapper through which all calls are performed.  This base
+        /// class uses that generic struct wrapper to specialize and de-virtualize.
+        /// </remarks>
+        internal interface IGenericSpecializedWrapper
+        {
+            void Store(ISubstringEqualityComparer @this);
+            public ReadOnlySpan<char> Slice(string s);
+            public bool Equals(string? x, string? y);
+            public int GetHashCode(string s);
+        }
+    }
+}