Improve search validator and in-memory evaluator. (#443)

* WIP. First draft.

* Refactored SearchMemoryEvaluator.

* Added allocation free SearchValidator implementation.

* Improved the nullability in search expressions.

* Cleanup.

* Minor fixes.

Author: fiseni

Directory.Packages.props

@@ -1,4 +1,5 @@
 <Project>
+
   <PropertyGroup>
     <ManagePackageVersionsCentrally>true</ManagePackageVersionsCentrally>
     <CentralPackageTransitivePinningEnabled>true</CentralPackageTransitivePinningEnabled>
@@ -10,8 +11,10 @@
     <PackageVersion Include="EntityFramework" Version="6.5.1" />
     <PackageVersion Include="Microsoft.CSharp" Version="4.7.0" />
     <PackageVersion Include="System.Data.DataSetExtensions" Version="4.5.0" />
+    <PackageVersion Include="System.Buffers" Version="4.6.0" />
+    <PackageVersion Include="System.Memory" Version="4.6.0" />
   </ItemGroup>
-  
+
   <ItemGroup Label="Test projects dependencies">
     <PackageVersion Include="ManagedObjectSize.ObjectPool" Version="0.0.7-gd53ba9da59" />
     <PackageVersion Include="MartinCostello.SqlLocalDb" Version="3.4.0" />

src/Ardalis.Specification/Ardalis.Specification.csproj

@@ -17,4 +17,9 @@
     </PackageReleaseNotes>    
   </PropertyGroup>
 
+  <ItemGroup Condition=" '$(TargetFramework)' == 'netstandard2.0' ">
+    <PackageReference Include="System.Buffers" />
+    <PackageReference Include="System.Memory" />
+  </ItemGroup>
+
 </Project>

src/Ardalis.Specification/Builders/Builder_Search.cs

@@ -16,7 +16,7 @@ public static partial class SpecificationBuilderExtensions
     /// <returns>The updated specification builder.</returns>
     public static ISpecificationBuilder<T, TResult> Search<T, TResult>(
         this ISpecificationBuilder<T, TResult> builder,
-        Expression<Func<T, string>> keySelector,
+        Expression<Func<T, string?>> keySelector,
         string pattern,
         int group = 1) where T : class
         => Search(builder, keySelector, pattern, true, group);
@@ -34,7 +34,7 @@ public static ISpecificationBuilder<T, TResult> Search<T, TResult>(
     /// <returns>The updated specification builder.</returns>
     public static ISpecificationBuilder<T, TResult> Search<T, TResult>(
         this ISpecificationBuilder<T, TResult> builder,
-        Expression<Func<T, string>> keySelector,
+        Expression<Func<T, string?>> keySelector,
         string pattern,
         bool condition,
         int group = 1) where T : class
@@ -59,7 +59,7 @@ public static ISpecificationBuilder<T, TResult> Search<T, TResult>(
     /// <returns>The updated specification builder.</returns>
     public static ISpecificationBuilder<T> Search<T>(
         this ISpecificationBuilder<T> builder,
-        Expression<Func<T, string>> keySelector,
+        Expression<Func<T, string?>> keySelector,
         string pattern,
         int group = 1) where T : class
         => Search(builder, keySelector, pattern, true, group);
@@ -76,7 +76,7 @@ public static ISpecificationBuilder<T> Search<T>(
     /// <returns>The updated specification builder.</returns>
     public static ISpecificationBuilder<T> Search<T>(
         this ISpecificationBuilder<T> builder,
-        Expression<Func<T, string>> keySelector,
+        Expression<Func<T, string?>> keySelector,
         string pattern,
         bool condition,
         int group = 1) where T : class

src/Ardalis.Specification/Evaluators/InMemorySpecificationEvaluator.cs

@@ -12,7 +12,7 @@ public InMemorySpecificationEvaluator()
         Evaluators.AddRange(new IInMemoryEvaluator[]
         {
             WhereEvaluator.Instance,
-            SearchEvaluator.Instance,
+            SearchMemoryEvaluator.Instance,
             OrderEvaluator.Instance,
             PaginationEvaluator.Instance
         });

src/Ardalis.Specification/Evaluators/Iterator.cs

@@ -0,0 +1,36 @@
+using System.Collections;
+using System.Diagnostics.CodeAnalysis;
+
+namespace Ardalis.Specification;
+
+internal abstract class Iterator<TSource> : IEnumerable<TSource>, IEnumerator<TSource>
+{
+    private readonly int _threadId = Environment.CurrentManagedThreadId;
+
+    private protected int _state;
+    private protected TSource _current = default!;
+
+    public Iterator<TSource> GetEnumerator()
+    {
+        var enumerator = _state == 0 && _threadId == Environment.CurrentManagedThreadId ? this : Clone();
+        enumerator._state = 1;
+        return enumerator;
+    }
+
+    public abstract Iterator<TSource> Clone();
+    public abstract bool MoveNext();
+
+    public TSource Current => _current;
+    object? IEnumerator.Current => Current;
+    IEnumerator<TSource> IEnumerable<TSource>.GetEnumerator() => GetEnumerator();
+    IEnumerator IEnumerable.GetEnumerator() => GetEnumerator();
+
+    [ExcludeFromCodeCoverage]
+    void IEnumerator.Reset() => throw new NotSupportedException();
+
+    public virtual void Dispose()
+    {
+        _current = default!;
+        _state = -1;
+    }
+}

src/Ardalis.Specification/Evaluators/SearchEvaluator.cs

@@ -1,42 +1,81 @@
-using System.Text.RegularExpressions;
+using System.Collections.Concurrent;
+using System.Diagnostics.CodeAnalysis;
+using System.Text.RegularExpressions;
 
 namespace Ardalis.Specification;
 
 public static class SearchExtension
 {
+    private static readonly RegexCache _regexCache = new();
+
+    private static Regex BuildRegex(string pattern)
+    {
+        // Escape special regex characters, excluding those handled separately
+        var regexPattern = Regex
+            .Escape(pattern)
+            .Replace("%", ".*")     // Translate SQL LIKE wildcard '%' to regex '.*'
+            .Replace("_", ".")      // Translate SQL LIKE wildcard '_' to regex '.'
+            .Replace(@"\[", "[")    // Unescape '[' as it's used for character classes/ranges
+            .Replace(@"\^", "^");   // Unescape '^' as it can be used for negation in character classes
+
+        // Ensure the pattern matches the entire string
+        regexPattern = "^" + regexPattern + "$";
+        var regex = new Regex(regexPattern, RegexOptions.IgnoreCase | RegexOptions.Compiled);
+        return regex;
+    }
+
     public static bool Like(this string input, string pattern)
     {
         try
         {
-            return SqlLike(input, pattern);
+            // The pattern is dynamic and arbitrary, the consumer might even compose it by an end-user input.
+            // We can not cache all Regex objects, but at least we can try to reuse the most "recent" ones. We'll cache 10 of them.
+            // This might improve the performance within the same closed loop for the in-memory evaluator and validator.
+
+            var regex = _regexCache.GetOrAdd(pattern, BuildRegex);
+            return regex.IsMatch(input);
         }
         catch (Exception ex)
         {
             throw new InvalidSearchPatternException(pattern, ex);
         }
     }
 
-    private static bool SqlLike(this string input, string pattern)
+    private class RegexCache
     {
-        // Escape special regex characters, excluding those handled separately
-        var regexPattern = Regex.Escape(pattern)
-            .Replace("%", ".*")     // Translate SQL LIKE wildcard '%' to regex '.*'
-            .Replace("_", ".")      // Translate SQL LIKE wildcard '_' to regex '.'
-            .Replace(@"\[", "[")    // Unescape '[' as it's used for character classes/ranges
-            .Replace(@"\^", "^");   // Unescape '^' as it can be used for negation in character classes
+        private const int MAX_SIZE = 10;
+        private readonly ConcurrentDictionary<string, Regex> _dictionary = new();
 
-        // Ensure the pattern matches the entire string
-        regexPattern = "^" + regexPattern + "$";
-        var regex = new Regex(regexPattern, RegexOptions.IgnoreCase);
+        public Regex GetOrAdd(string key, Func<string, Regex> valueFactory)
+        {
+            if (_dictionary.TryGetValue(key, out var regex))
+                return regex;
 
-        return regex.IsMatch(input);
+            // It might happen we end up with more items than max (concurrency), but we won't be too strict.
+            // We're just trying to avoid indefinite growth.
+            for (int i = _dictionary.Count - MAX_SIZE; i >= 0; i--)
+            {
+                // Avoid being smart, just remove sequentially from the start.
+                var firstKey = _dictionary.Keys.FirstOrDefault();
+                if (firstKey is not null)
+                {
+                    _dictionary.TryRemove(firstKey, out _);
+                }
+
+            }
+
+            var newRegex = valueFactory(key);
+            _dictionary.TryAdd(key, newRegex);
+            return newRegex;
+        }
     }
 
+#pragma warning disable IDE0051 // Remove unused private members
     // This C# implementation of SQL Like operator is based on the following SO post https://stackoverflow.com/a/8583383/10577116
     // It covers almost all of the scenarios, and it's faster than regex based implementations.
     // It may fail/throw in some very specific and edge cases, hence, wrap it in try/catch.
     // UPDATE: it returns incorrect results for some obvious cases.
-    // More details in this issue https://github.com/ardalis/Specification/issues/390
+    [ExcludeFromCodeCoverage] // Dead code. Keeping it just as a reference
     private static bool SqlLikeOption2(string str, string pattern)
     {
         var isMatch = true;
@@ -169,4 +208,5 @@ private static bool SqlLikeOption2(string str, string pattern)
         }
         return isMatch && endOfPattern;
     }
+#pragma warning restore IDE0051
 }

src/Ardalis.Specification/Evaluators/SearchExtension.cs

@@ -0,0 +1,109 @@
+using System.Diagnostics;
+
+namespace Ardalis.Specification;
+
+public class SearchMemoryEvaluator : IInMemoryEvaluator
+{
+    private SearchMemoryEvaluator() { }
+    public static SearchMemoryEvaluator Instance { get; } = new SearchMemoryEvaluator();
+
+    public IEnumerable<T> Evaluate<T>(IEnumerable<T> query, ISpecification<T> specification)
+    {
+        if (specification.SearchCriterias is List<SearchExpressionInfo<T>> { Count: > 0 } list)
+        {
+            // The search expressions are already sorted by SearchGroup.
+            return new SpecLikeIterator<T>(query, list);
+        }
+
+        return query;
+    }
+
+    private sealed class SpecLikeIterator<TSource> : Iterator<TSource>
+    {
+        private readonly IEnumerable<TSource> _source;
+        private readonly List<SearchExpressionInfo<TSource>> _searchExpressions;
+
+        private IEnumerator<TSource>? _enumerator;
+
+        public SpecLikeIterator(IEnumerable<TSource> source, List<SearchExpressionInfo<TSource>> searchExpressions)
+        {
+            _source = source;
+            _searchExpressions = searchExpressions;
+        }
+
+        public override Iterator<TSource> Clone()
+            => new SpecLikeIterator<TSource>(_source, _searchExpressions);
+
+        public override void Dispose()
+        {
+            if (_enumerator is not null)
+            {
+                _enumerator.Dispose();
+                _enumerator = null;
+            }
+            base.Dispose();
+        }
+
+        public override bool MoveNext()
+        {
+            switch (_state)
+            {
+                case 1:
+                    _enumerator = _source.GetEnumerator();
+                    _state = 2;
+                    goto case 2;
+                case 2:
+                    Debug.Assert(_enumerator is not null);
+                    var searchExpressions = _searchExpressions;
+                    while (_enumerator!.MoveNext())
+                    {
+                        TSource sourceItem = _enumerator.Current;
+                        if (IsValid(sourceItem, searchExpressions))
+                        {
+                            _current = sourceItem;
+                            return true;
+                        }
+                    }
+
+                    Dispose();
+                    break;
+            }
+
+            return false;
+        }
+
+        // This would be simpler using Span<SearchExpressionInfo<TSource>>
+        // but CollectionsMarshal.AsSpan is not available in .NET Standard 2.0
+        private static bool IsValid<T>(T sourceItem, List<SearchExpressionInfo<T>> list)
+        {
+            var groupStart = 0;
+            for (var i = 1; i <= list.Count; i++)
+            {
+                // If we reached the end of the list or the group has changed, we slice and process the group.
+                if (i == list.Count || list[i].SearchGroup != list[groupStart].SearchGroup)
+                {
+                    if (IsValidInOrGroup(sourceItem, list, groupStart, i) is false)
+                    {
+                        return false;
+                    }
+                    groupStart = i;
+                }
+            }
+            return true;
+
+            static bool IsValidInOrGroup(T sourceItem, List<SearchExpressionInfo<T>> list, int from, int to)
+            {
+                var validOrGroup = false;
+                for (int i = from; i < to; i++)
+                {
+                    if (list[i].SelectorFunc(sourceItem)?.Like(list[i].SearchTerm) ?? false)
+                    {
+                        validOrGroup = true;
+                        break;
+                    }
+                }
+                return validOrGroup;
+            }
+        }
+    }
+}

src/Ardalis.Specification/Evaluators/SearchMemoryEvaluator.cs

@@ -6,7 +6,7 @@
 /// <typeparam name="T">Type of the source from which search target should be selected.</typeparam>
 public class SearchExpressionInfo<T>
 {
-    private readonly Lazy<Func<T, string>> _selectorFunc;
+    private readonly Lazy<Func<T, string?>> _selectorFunc;
 
     /// <summary>
     /// Creates instance of <see cref="SearchExpressionInfo{T}" />.
@@ -16,7 +16,7 @@ public class SearchExpressionInfo<T>
     /// <param name="searchGroup">The index used to group sets of Selectors and SearchTerms together.</param>
     /// <exception cref="ArgumentNullException">If <paramref name="selector"/> is null.</exception>
     /// <exception cref="ArgumentNullException">If <paramref name="searchTerm"/> is null or empty.</exception>
-    public SearchExpressionInfo(Expression<Func<T, string>> selector, string searchTerm, int searchGroup = 1)
+    public SearchExpressionInfo(Expression<Func<T, string?>> selector, string searchTerm, int searchGroup = 1)
     {
         _ = selector ?? throw new ArgumentNullException(nameof(selector));
         if (string.IsNullOrEmpty(searchTerm)) throw new ArgumentException("The search term can not be null or empty.");
@@ -25,13 +25,13 @@ public SearchExpressionInfo(Expression<Func<T, string>> selector, string searchT
         SearchTerm = searchTerm;
         SearchGroup = searchGroup;
 
-        _selectorFunc = new Lazy<Func<T, string>>(Selector.Compile);
+        _selectorFunc = new Lazy<Func<T, string?>>(Selector.Compile);
     }
 
     /// <summary>
     /// The property to apply the SQL LIKE against.
     /// </summary>
-    public Expression<Func<T, string>> Selector { get; }
+    public Expression<Func<T, string?>> Selector { get; }
 
     /// <summary>
     /// The value to use for the SQL LIKE.
@@ -46,5 +46,5 @@ public SearchExpressionInfo(Expression<Func<T, string>> selector, string searchT
     /// <summary>
     /// Compiled <see cref="Selector" />.
     /// </summary>
-    public Func<T, string> SelectorFunc => _selectorFunc.Value;
+    public Func<T, string?> SelectorFunc => _selectorFunc.Value;
 }