Zero-copy SlicedString search (indexOf/startsWith/endsWith/includes)

Jint.Benchmark/StringSliceBenchmarks.cs

@@ -11,6 +11,9 @@ namespace Jint.Benchmark;
 /// previously fell below the zero-copy retention guard and copied on every call.
 /// SliceSmall guards the small-result path; SliceThenRead guards lazy-materialization cost
 /// when the result is actually consumed.
+/// SearchOnSlice exercises indexOf/startsWith/endsWith/includes on a fresh large slice each
+/// iteration — the case where the inherited base search methods materialize the whole substring
+/// on every call. With zero-copy span search overrides this drops to ~0 allocation.
 /// </summary>
 [MemoryDiagnoser]
 [HideColumns("Error", "Gen0", "Gen1", "Gen2")]
@@ -21,6 +24,8 @@ public class StringSliceBenchmarks
     private Prepared<Script> _substringLargeDiscard;
     private Prepared<Script> _sliceSmall;
     private Prepared<Script> _sliceThenRead;
+    private Prepared<Script> _searchOnSlice;
+    private Prepared<Script> _searchOnFlat;
 
     [GlobalSetup]
     public void Setup()
@@ -71,6 +76,39 @@ public void Setup()
             })();
             """, strict: true);
 
+        _searchOnSlice = Engine.PrepareScript("""
+            (function() {
+                var hits = 0;
+                for (var i = 0; i < 5000; i++) {
+                    var sub = str.slice(12000, -1);
+                    if (sub.indexOf("~absent~") !== -1) hits++;
+                    if (sub.startsWith("aB3$x")) hits++;
+                    if (sub.endsWith("_kEw")) hits++;
+                    if (sub.includes("Q9pLm")) hits++;
+                }
+                return hits;
+            })();
+            """, strict: true);
+
+        // Guard: searching a plain (non-view) JsString must not regress from making the base
+        // search methods virtual. str.slice(0) returns a flat JsString, not a SlicedString.
+        // Deliberately dispatch-bound (many cheap short searches that hit early / compare only the
+        // needle) so the measurement isolates per-call dispatch overhead rather than the highly
+        // thermal-sensitive throughput of one giant not-found scan.
+        _searchOnFlat = Engine.PrepareScript("""
+            (function() {
+                var hits = 0;
+                var flat = str.slice(0);
+                for (var i = 0; i < 100000; i++) {
+                    if (flat.indexOf("aB3$x") === 0) hits++;
+                    if (flat.startsWith("aB3$xQ9")) hits++;
+                    if (flat.endsWith("kEwZ")) hits++;
+                    if (flat.includes("aB3$x")) hits++;
+                }
+                return hits;
+            })();
+            """, strict: true);
+
         _engine = new Engine(static options => options.Strict());
         // Build the shared ~128K char base string once (dromaeo-style doubling).
         _engine.Execute("""
@@ -83,6 +121,8 @@ public void Setup()
         _engine.Evaluate(_substringLargeDiscard);
         _engine.Evaluate(_sliceSmall);
         _engine.Evaluate(_sliceThenRead);
+        _engine.Evaluate(_searchOnSlice);
+        _engine.Evaluate(_searchOnFlat);
     }
 
     [Benchmark]
@@ -96,4 +136,10 @@ public void Setup()
 
     [Benchmark]
     public JsValue SliceThenRead() => _engine.Evaluate(_sliceThenRead);
+
+    [Benchmark]
+    public JsValue SearchOnSlice() => _engine.Evaluate(_searchOnSlice);
+
+    [Benchmark]
+    public JsValue SearchOnFlat() => _engine.Evaluate(_searchOnFlat);
 }

Jint.Tests/Runtime/StringTests.cs

@@ -120,6 +120,63 @@ public void ShouldCompareWithLocale()
         Assert.Equal(-1, engine.Evaluate("'王五'.localeCompare('张三', 'zh-CN')").AsInteger());
     }
 
+    [Fact]
+    public void SlicedStringSearchMatchesFlatString()
+    {
+        // A large slice is represented internally as a zero-copy view (SlicedString) whose
+        // indexOf/includes/startsWith/endsWith search the backing span directly. This differential
+        // test compares those results against an identical *flat* string (built via a JSON round-trip,
+        // which routes through the base ToString-backed search) for many needles and positions. The
+        // sliced value is deliberately searched while still un-materialized, then again after being
+        // forced to materialize, so both AsSpan() sources are exercised.
+        const string script = @"
+var seed = 'aB3$xQ9pLm0_kEwZ';
+var s = seed;
+while (s.length < 4096) s += s;          // 4096 chars
+var sub = s.slice(100, 4000);            // length 3900 -> SlicedString (zero-copy view)
+var flat = JSON.parse(JSON.stringify(s.slice(100, 4000))); // identical content, plain string
+
+if (sub.length !== flat.length) throw new Error('length mismatch');
+
+var needles = ['aB3$x', 'Q9pLm', 'kEwZ', 'ZaB', '~none~', '', 'm0_', seed, seed + seed, 'Z'];
+var positions = [-1, 0, 1, 50, 100, 1000, 3899, 3900, 3901, 5000];
+
+function compareAll() {
+    for (var n = 0; n < needles.length; n++) {
+        var ndl = needles[n];
+        if (sub.indexOf(ndl) !== flat.indexOf(ndl)) return 'indexOf:' + n;
+        if (sub.includes(ndl) !== flat.includes(ndl)) return 'includes:' + n;
+        if (sub.startsWith(ndl) !== flat.startsWith(ndl)) return 'startsWith:' + n;
+        if (sub.endsWith(ndl) !== flat.endsWith(ndl)) return 'endsWith:' + n;
+        for (var p = 0; p < positions.length; p++) {
+            var pos = positions[p];
+            if (sub.indexOf(ndl, pos) !== flat.indexOf(ndl, pos)) return 'indexOf@' + n + ',' + pos;
+            if (sub.includes(ndl, pos) !== flat.includes(ndl, pos)) return 'includes@' + n + ',' + pos;
+            if (sub.startsWith(ndl, pos) !== flat.startsWith(ndl, pos)) return 'startsWith@' + n + ',' + pos;
+            if (sub.endsWith(ndl, pos) !== flat.endsWith(ndl, pos)) return 'endsWith@' + n + ',' + pos;
+        }
+    }
+    return 'ok';
+}
+
+var first = compareAll();              // sub is still an un-materialized view here
+var forced = sub + 'x';                // forces sub to materialize its backing substring
+var second = compareAll();             // now searches the materialized span
+
+// Absolute spot-checks (sub begins at s[100]; 100 % 16 == 4 -> 'x'), so a bug shared by both
+// the sliced and flat search paths cannot make the differential pass silently.
+var abs = sub.charAt(0) === 'x'
+    && sub.startsWith('xQ9pLm0_kEwZ')
+    && sub.indexOf('aB3$x') === 12
+    && sub.indexOf('x', 1) === 16
+    && sub.includes('kEwZaB3$')
+    && sub.endsWith(flat.slice(-7));
+
+first + '|' + second + '|' + (abs ? 'ok' : 'absfail');
+";
+        Assert.Equal("ok|ok|ok", _engine.Evaluate(script).AsString());
+    }
+
     public static TheoryData<string, string> GetLithuaniaTestsData()
     {
         return new StringTetsLithuaniaData().TestData();

Jint/Native/JsString.cs

@@ -289,7 +289,7 @@ internal sealed override bool ToBoolean()
 
     public override string ToString() => _value;
 
-    internal bool Contains(char c)
+    internal virtual bool Contains(char c)
     {
         if (c == 0)
         {
@@ -298,7 +298,7 @@ internal bool Contains(char c)
         return ToString().Contains(c);
     }
 
-    internal int IndexOf(string value, int startIndex = 0)
+    internal virtual int IndexOf(string value, int startIndex = 0)
     {
         if (Length - startIndex < value.Length)
         {
@@ -307,12 +307,12 @@ internal int IndexOf(string value, int startIndex = 0)
         return ToString().IndexOf(value, startIndex, StringComparison.Ordinal);
     }
 
-    internal bool StartsWith(string value, int start = 0)
+    internal virtual bool StartsWith(string value, int start = 0)
     {
         return value.Length + start <= Length && ToString().AsSpan(start).StartsWith(value.AsSpan(), StringComparison.Ordinal);
     }
 
-    internal bool EndsWith(string value, int end = 0)
+    internal virtual bool EndsWith(string value, int end = 0)
     {
         var start = end - value.Length;
         return start >= 0 && ToString().AsSpan(start, value.Length).EndsWith(value.AsSpan(), StringComparison.Ordinal);
@@ -528,6 +528,37 @@ public override string ToString()
 
         private ReadOnlySpan<char> AsSpan() => _value is not null ? _value.AsSpan() : _source.AsSpan(_start, _length);
 
+        // Search directly over the slice's span. The inherited base implementations route through
+        // ToString(), which materializes (and allocates) the whole substring on every search; these
+        // overrides keep a discarded slice zero-copy. Ordinal char comparison is binary/sequence
+        // equality, so the parameterless span overloads match the base StringComparison.Ordinal paths.
+        internal override int IndexOf(string value, int startIndex = 0)
+        {
+            if (_length - startIndex < value.Length)
+            {
+                return -1;
+            }
+
+            var index = AsSpan().Slice(startIndex).IndexOf(value.AsSpan());
+            return index < 0 ? -1 : index + startIndex;
+        }
+
+        internal override bool StartsWith(string value, int start = 0)
+        {
+            return value.Length + start <= _length && AsSpan().Slice(start).StartsWith(value.AsSpan());
+        }
+
+        internal override bool EndsWith(string value, int end = 0)
+        {
+            var start = end - value.Length;
+            return start >= 0 && AsSpan().Slice(start, value.Length).EndsWith(value.AsSpan());
+        }
+
+        internal override bool Contains(char c)
+        {
+            return c != 0 && AsSpan().IndexOf(c) >= 0;
+        }
+
         internal override JsString EnsureCapacity(int capacity)
         {
             // base implementation reads _value directly, which may not be materialized yet