AyCode.Core/AyCode.Core/Serializers/Binaries/AcBinarySerializer.BinarySerializationContext.cs

using System;
using System.Buffers;
using System.Collections.Concurrent;
using System.Collections.Generic;
using System.Numerics;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Runtime.Intrinsics;
using System.Text;
using System.Threading;
using static AyCode.Core.Helpers.JsonUtilities;

namespace AyCode.Core.Serializers.Binaries;

public static partial class AcBinarySerializer
{
    private static class BinarySerializationContextPool
    {
        private static readonly ConcurrentQueue<BinarySerializationContext> Pool = new();

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public static BinarySerializationContext Get(AcBinarySerializerOptions options)
        {
            if (Pool.TryDequeue(out var context))
            {
                context.Reset(options);
                return context;
            }

            return new BinarySerializationContext(options);
        }

        public static void ReturnAsync(BinarySerializationContext context)
        {
            // 🔥 FIRE-AND-FORGET: cleanup háttérben
            ThreadPool.UnsafeQueueUserWorkItem(Return, context, preferLocal: true);
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public static void Return(BinarySerializationContext context)
        {
            if (Pool.Count < context.Options.MaxContextPoolSize)
            {
                context.Clear();
                Pool.Enqueue(context);
            }
            else
            {
                context.Dispose();
            }
        }
    }

    public static int GrowBufferCount =>
#if DEBUG
        BinarySerializationContext.GrowBufferCount;
#else
            -1;
#endif

    public static long GrowBufferTotalBytes =>
#if DEBUG
        BinarySerializationContext.GrowBufferTotalBytes;
#else
        -1;
#endif

    /// <summary>
    /// Binary serialization context. Public for generated serializers.
    /// </summary>
    internal sealed class BinarySerializationContext : SerializationContextBase<BinarySerializeTypeMetadata, AcBinarySerializerOptions>, IDisposable
    {
        private const int MinBufferSize = 512;
        private const int PropertyIndexBufferMaxCache = 512;
        private const int PropertyStateBufferMaxCache = 512;
        private const int InitialInternCapacity = 32;
        private byte[] _buffer;
        private int _position;
        private int _initialBufferSize;

#if DEBUG
        /// <summary>
        /// Counts how many times GrowBuffer was called during serialization.
        /// Used for benchmarking buffer allocation efficiency.
        /// </summary>
        public static int GrowBufferCount { get; set; }

        /// <summary>
        /// Total bytes allocated by GrowBuffer during serialization.
        /// Used for benchmarking buffer allocation efficiency.
        /// </summary>
        public static long GrowBufferTotalBytes { get; set; }
#endif

        // Use shared reference tracker from AcSerializerCommon
        //private readonly AcSerializerCommon.SerializationReferenceTracker _refTracker = new();

        private IdentityMap<string, InternEntry>? _stringInternMap;
        private int _nextCacheIndex;    // Next dense cache index to assign (starts at 0, uses ++_nextCacheIndex)
        private int _nextFirstIndex;    // Next first occurrence index to assign (scan pass)

        /// <summary>
        /// Next cache index reference for scan pass. Direct ref access for TryTrack methods.
        /// </summary>
        public ref int NextCacheIndexRef
        {
            [MethodImpl(MethodImplOptions.AggressiveInlining)]
            get => ref _nextCacheIndex;
        }

        /// <summary>
        /// Next first occurrence index for scan pass. Direct access for performance.
        /// </summary>
        public int NextFirstIndex
        {
            [MethodImpl(MethodImplOptions.AggressiveInlining)]
            get => _nextFirstIndex;
            [MethodImpl(MethodImplOptions.AggressiveInlining)]
            set => _nextFirstIndex = value;
        }

        private int[]? _propertyIndexBuffer;
        private byte[]? _propertyStateBuffer;

#if DEBUG
        /// <summary>
        /// DEBUG ONLY: Current property path being serialized (e.g., "Order.Status").
        /// Used for string interning analysis.
        /// </summary>
        internal string? CurrentPropertyPath;

        /// <summary>
        /// DEBUG ONLY: Callback invoked when a string is registered for interning.
        /// Parameters: (propertyPath, stringValue)
        /// Use this to analyze which properties have repeated string values.
        /// </summary>
        internal Action<string?, string>? OnStringInterned;
#endif

        // These properties delegate to Options for convenience
        public bool UseStringInterning => Options.UseStringInterning != StringInterningMode.None;

        public bool IsValidForInterningString(int strLength)
        {
            return strLength >= MinStringInternLength && (MaxStringInternLength == 0 || strLength <= MaxStringInternLength);
        }

        /// <summary>
        /// True if we have interning/ref tracking (cache count needed in header).
        /// </summary>
        public bool HasCaching => UseStringInterning || ReferenceHandling != ReferenceHandlingMode.None;
        public bool UseMetadata => Options.UseMetadata;
        public byte MinStringInternLength => Options.MinStringInternLength;
        public byte MaxStringInternLength => Options.MaxStringInternLength;
        public BinaryPropertyFilter? PropertyFilter => Options.PropertyFilter;

        /// <summary>
        /// Cached check for PropertyFilter != null. Set in Reset() to avoid property getter in hot loop.
        /// </summary>
        public bool HasPropertyFilter { get; private set; }

        public int Position => _position;

        public BinarySerializationContext(AcBinarySerializerOptions options)
        {
            _initialBufferSize = Math.Max(options.InitialBufferCapacity, MinBufferSize);
            _buffer = ArrayPool<byte>.Shared.Rent(_initialBufferSize);
            Reset(options);
        }

        /// <summary>
        /// Factory for creating BinarySerializeTypeMetadata instances.
        /// </summary>
        protected override Func<Type, BinarySerializeTypeMetadata> MetadataFactory
            => static t => new BinarySerializeTypeMetadata(t, HasJsonIgnoreAttribute);

        public override void Reset(AcBinarySerializerOptions options)
        {
            // IMPORTANT: base.Reset sets Options first, so derived code can use Options-derived properties
            base.Reset(options);

            _position = 0;
            _initialBufferSize = Math.Max(Options.InitialBufferCapacity, MinBufferSize);
            HasPropertyFilter = Options.PropertyFilter != null;

            // NOTE: GrowBufferCount és GrowBufferTotalBytes NEM nullázódik itt!
            // Kumulatívan gyűjtjük a benchmark során.

            if (_buffer.Length < _initialBufferSize)
            {
                ArrayPool<byte>.Shared.Return(_buffer);
                _buffer = ArrayPool<byte>.Shared.Rent(_initialBufferSize);
            }
        }

        public override void Clear()
        {
            _position = 0;

            //_refTracker.Reset();
            _stringInternMap?.Reset();
            _nextCacheIndex = 0;
            _nextFirstIndex = 0;

            if (_propertyIndexBuffer != null && _propertyIndexBuffer.Length > PropertyIndexBufferMaxCache)
            {
                ArrayPool<int>.Shared.Return(_propertyIndexBuffer);
                _propertyIndexBuffer = null;
            }

            if (_propertyStateBuffer != null && _propertyStateBuffer.Length > PropertyStateBufferMaxCache)
            {
                ArrayPool<byte>.Shared.Return(_propertyStateBuffer);
                _propertyStateBuffer = null;
            }

            // Clear wrapper tracking - returns IdentityMap arrays to pool
            base.Clear();
        }


        public void Dispose()
        {
            if (_buffer != null)
            {
                ArrayPool<byte>.Shared.Return(_buffer);
                _buffer = null!;
            }

            if (_propertyIndexBuffer != null)
            {
                ArrayPool<int>.Shared.Return(_propertyIndexBuffer);
                _propertyIndexBuffer = null;
            }

            if (_propertyStateBuffer != null)
            {
                ArrayPool<byte>.Shared.Return(_propertyStateBuffer);
                _propertyStateBuffer = null;
            }

        }

        #region String Interning

        /// <summary>
        /// Serialize pass: looks up interned string state.
        /// Returns the entry ref for caller to check IsFirstWrite and update it.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public ref InternEntry GetInternedStringEntry(string value, out bool found)
        {
            if (_stringInternMap == null)
            {
                found = false;
                return ref System.Runtime.CompilerServices.Unsafe.NullRef<InternEntry>();
            }

            if (_stringInternMap.TryAdd(value, out var slotIndex))
            {
                // Not in map (shouldn't happen after scan pass for cached strings)
                found = false;
                return ref _stringInternMap.GetValueRef(slotIndex);
            }

            found = true;
            return ref _stringInternMap.GetValueRef(slotIndex);
        }

        /// <summary>
        /// Scan pass: tracks a string for interning. Assigns CacheIndex immediately on 2nd occurrence.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void ScanInternString(string value)
        {
            _stringInternMap ??= new IdentityMap<string, InternEntry>();

            if (!_stringInternMap.TryAdd(value, out var slotIndex))
            {
                // 2+ occurrence: assign CacheIndex immediately
                ref var entry = ref _stringInternMap.GetValueRef(slotIndex);
                if (entry.CacheIndex == -1)
                {
                    entry.CacheIndex = ++_nextCacheIndex;
                    entry.IsFirstWrite = true;
                }
                return;
            }

            // 1st occurrence: store FirstIndex for validation, CacheIndex = -1 (not cached yet)
            ref var newEntry = ref _stringInternMap.GetValueRef(slotIndex);
            newEntry.FirstIndex = _nextFirstIndex++;
            newEntry.CacheIndex = -1;
        }

        /// <summary>
        /// Returns true if there are any interned strings that occurred more than once.
        /// </summary>
        public bool HasInternedStrings => _stringInternMap != null && _stringInternMap.Count > 0;

        /// <summary>
        /// Gets the count of cached values (string intern + object ref that occurred more than once).
        /// </summary>
        public int GetCacheCount() => _nextCacheIndex;


        #endregion

        #region UseMetadata Type Tracking

        /// <summary>
        /// Regisztrálja a típust UseMetadata módban.
        /// Visszaadja true-t ha ez az első előfordulás (inline hash-eket kell írni),
        /// false-t ha ismételt (csak propNameHash kell).
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public static bool RegisterMetadataType(TypeMetadataWrapper<BinarySerializeTypeMetadata> wrapper)
        {
            if (wrapper.MetadataFooterIndex >= 0)
                return false; // ismételt

            wrapper.MetadataFooterIndex = 0; // jelöljük hogy már regisztrálva
            return true; // első előfordulás
        }

        /// <summary>
        /// Inline metadata kiírása az ObjectWithMetadata marker után.
        /// Első előfordulás: [propNameHash (4b)][propCount (VarUInt)][hash0 (4b)][hash1 (4b)]...
        /// Ismételt:         [propNameHash (4b)]
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteInlineMetadata(BinarySerializeTypeMetadata metadata, bool isFirstOccurrence)
        {
            WriteRaw(metadata.PropNameHash);

            if (isFirstOccurrence)
            {
                var hashes = metadata.MetadataPropertyHashes;
                WriteVarUInt((uint)hashes.Length);
                for (var i = 0; i < hashes.Length; i++)
                {
                    WriteRaw(hashes[i]);
                }
            }
        }

        #endregion

        #region Property State Buffer

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public byte[] RentPropertyStateBuffer(int size)
        {
            if (_propertyStateBuffer != null && _propertyStateBuffer.Length >= size)
            {
                return _propertyStateBuffer;
            }

            if (_propertyStateBuffer != null)
            {
                ArrayPool<byte>.Shared.Return(_propertyStateBuffer);
            }

            _propertyStateBuffer = ArrayPool<byte>.Shared.Rent(size);
            return _propertyStateBuffer;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void ReturnPropertyStateBuffer(byte[] buffer)
        {
            // Buffer stays cached for reuse.
        }

        #endregion

        #region Output

        /// <summary>
        /// Returns the serialized data as a ReadOnlySpan without allocation.
        /// Use this for compression or other processing before final ToArray().
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public ReadOnlySpan<byte> AsSpan() => _buffer.AsSpan(0, _position);

        public byte[] ToArray()
        {
            var result = GC.AllocateUninitializedArray<byte>(_position);
            _buffer.AsSpan(0, _position).CopyTo(result);
            return result;
        }

        public void WriteTo(IBufferWriter<byte> writer)
        {
            var span = writer.GetSpan(_position);
            _buffer.AsSpan(0, _position).CopyTo(span);
            writer.Advance(_position);
        }

        public BinarySerializationResult DetachResult()
        {
            var resultBuffer = _buffer;
            var resultLength = _position;

            _buffer = ArrayPool<byte>.Shared.Rent(_initialBufferSize);
            _position = 0;

            return new BinarySerializationResult(resultBuffer, resultLength, pooled: true);
        }

        #endregion

        #region Property Filtering

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public bool ShouldSerializeProperty(object instance, BinaryPropertyAccessor property)
        {
            if (PropertyFilter == null)
            {
                return true;
            }

            var context = new BinaryPropertyFilterContext(
                instance,
                property.DeclaringType,
                property.Name,
                property.PropertyType,
                property.DynamicGetter);
            return PropertyFilter(context);
        }

        public bool CheckDuplicatePropName => Options.CheckDuplicatePropName;

        #endregion

        #region Buffer Helpers

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private void EnsureCapacity(int additionalBytes)
        {
            var required = _position + additionalBytes;
            if (required <= _buffer.Length)
            {
                return;
            }

            GrowBuffer(required);
        }

        [MethodImpl(MethodImplOptions.NoInlining)]
        private void GrowBuffer(int required)
        {
            var newSize = Math.Max(_buffer.Length * 2, required);
            var newBuffer = ArrayPool<byte>.Shared.Rent(newSize);
            _buffer.AsSpan(0, _position).CopyTo(newBuffer);
            ArrayPool<byte>.Shared.Return(_buffer);
            _buffer = newBuffer;

            #if DEBUG
            GrowBufferCount++;
            GrowBufferTotalBytes += newSize;
            #endif
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteByte(byte value)
        {
            if (_position >= _buffer.Length)
            {
                GrowBuffer(_position + 1);
            }
            _buffer[_position++] = value;
        }

        /// <summary>
        /// Write type code byte followed by a raw value. Batches EnsureCapacity call.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteTypeCodeAndRaw<T>(byte typeCode, T value) where T : unmanaged
        {
            var size = 1 + Unsafe.SizeOf<T>();
            EnsureCapacity(size);
            _buffer[_position++] = typeCode;
            Unsafe.WriteUnaligned(ref _buffer[_position], value);
            _position += Unsafe.SizeOf<T>();
        }

        /// <summary>
        /// Write two bytes efficiently.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteTwoBytes(byte b1, byte b2)
        {
            EnsureCapacity(2);
            _buffer[_position++] = b1;
            _buffer[_position++] = b2;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteBytes(ReadOnlySpan<byte> data)
        {
            EnsureCapacity(data.Length);
            data.CopyTo(_buffer.AsSpan(_position));
            _position += data.Length;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteRaw<T>(T value) where T : unmanaged
        {
            var size = Unsafe.SizeOf<T>();
            EnsureCapacity(size);
            Unsafe.WriteUnaligned(ref _buffer[_position], value);
            _position += size;
        }

        #endregion

        #region Specialized Writers

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteDecimalBits(decimal value)
        {
            EnsureCapacity(16);
            Span<int> bits = stackalloc int[4];
            decimal.TryGetBits(value, bits, out _);
            MemoryMarshal.AsBytes(bits).CopyTo(_buffer.AsSpan(_position, 16));
            _position += 16;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteDateTimeBits(DateTime value)
        {
            EnsureCapacity(9);
            Unsafe.WriteUnaligned(ref _buffer[_position], value.Ticks);
            _buffer[_position + 8] = (byte)value.Kind;
            _position += 9;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteGuidBits(Guid value)
        {
            EnsureCapacity(16);
            value.TryWriteBytes(_buffer.AsSpan(_position, 16));
            _position += 16;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteDateTimeOffsetBits(DateTimeOffset value)
        {
            EnsureCapacity(10);
            Unsafe.WriteUnaligned(ref _buffer[_position], value.UtcTicks);
            Unsafe.WriteUnaligned(ref _buffer[_position + 8], (short)value.Offset.TotalMinutes);
            _position += 10;
        }


        public void WriteVarInt(int value)
        {
            var encoded = (uint)((value << 1) ^ (value >> 31));
            // Fast path for small positive values (0-63 when ZigZag encoded)
            if (encoded < 0x80)
            {
                EnsureCapacity(1);
                _buffer[_position++] = (byte)encoded;
                return;
            }
            EnsureCapacity(5);
            WriteVarUIntInternal(encoded);
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteVarUInt(uint value)
        {
            // Fast path for small values (0-127)
            if (value < 0x80)
            {
                EnsureCapacity(1);
                _buffer[_position++] = (byte)value;
                return;
            }
            EnsureCapacity(5);
            WriteVarUIntInternal(value);
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private void WriteVarUIntInternal(uint value)
        {
            while (value >= 0x80)
            {
                _buffer[_position++] = (byte)(value | 0x80);
                value >>= 7;
            }

            _buffer[_position++] = (byte)value;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteVarLong(long value)
        {
            var encoded = (ulong)((value << 1) ^ (value >> 63));
            // Fast path for small values
            if (encoded < 0x80)
            {
                EnsureCapacity(1);
                _buffer[_position++] = (byte)encoded;
                return;
            }
            EnsureCapacity(10);
            WriteVarULongInternal(encoded);
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteVarULong(ulong value)
        {
            // Fast path for small values (0-127)
            if (value < 0x80)
            {
                EnsureCapacity(1);
                _buffer[_position++] = (byte)value;
                return;
            }
            EnsureCapacity(10);
            WriteVarULongInternal(value);
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private void WriteVarULongInternal(ulong value)
        {
            while (value >= 0x80)
            {
                _buffer[_position++] = (byte)(value | 0x80);
                value >>= 7;
            }

            _buffer[_position++] = (byte)value;
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteStringUtf8(string value)
        {
            // Fast path for ASCII-only strings using SIMD-optimized check
            if (Ascii.IsValid(value))
            {
                WriteVarUInt((uint)value.Length);
                EnsureCapacity(value.Length);
                // Use System.Text.Ascii for SIMD-optimized ASCII to bytes conversion
                Ascii.FromUtf16(value.AsSpan(), _buffer.AsSpan(_position, value.Length), out _);
                _position += value.Length;
                return;
            }

            // Standard path for multi-byte UTF8
            var byteCount = Utf8NoBom.GetByteCount(value);
            WriteVarUInt((uint)byteCount);
            EnsureCapacity(byteCount);
            Utf8NoBom.GetBytes(value.AsSpan(), _buffer.AsSpan(_position, byteCount));
            _position += byteCount;
        }

        /// <summary>
        /// Checks if string contains only ASCII characters (0-127).
        /// Optimized loop with early exit.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private static bool IsAscii(string value)
        {
            var span = value.AsSpan();
            for (var i = 0; i < span.Length; i++)
            {
                if (span[i] > 127)
                    return false;
            }
            return true;
        }

        /// <summary>
        /// Writes ASCII string directly to byte buffer (char to byte, no encoding needed).
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private static void WriteAsciiDirect(ReadOnlySpan<char> source, Span<byte> destination)
        {
            for (var i = 0; i < source.Length; i++)
            {
                destination[i] = (byte)source[i];
            }
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WritePreencodedPropertyName(ReadOnlySpan<byte> utf8Name)
        {
            WriteByte(BinaryTypeCode.String);
            WriteVarUInt((uint)utf8Name.Length);
            WriteBytes(utf8Name);
        }

        public void WriteInt32ArrayOptimized(int[] array)
        {
            for (var i = 0; i < array.Length; i++)
            {
                var value = array[i];
                if (BinaryTypeCode.TryEncodeTinyInt(value, out var tiny))
                {
                    WriteByte(tiny);
                }
                else
                {
                    WriteByte(BinaryTypeCode.Int32);
                    WriteVarInt(value);
                }
            }
        }

        public void WriteLongArrayOptimized(long[] array)
        {
            for (var i = 0; i < array.Length; i++)
            {
                var value = array[i];
                if (value >= int.MinValue && value <= int.MaxValue)
                {
                    var intValue = (int)value;
                    if (BinaryTypeCode.TryEncodeTinyInt(intValue, out var tiny))
                    {
                        WriteByte(tiny);
                    }
                    else
                    {
                        WriteByte(BinaryTypeCode.Int32);
                        WriteVarInt(intValue);
                    }
                }
                else
                {
                    WriteByte(BinaryTypeCode.Int64);
                    WriteVarLong(value);
                }
            }
        }

        public void WriteDoubleArrayBulk(double[] array)
        {
            EnsureCapacity(array.Length * 9);
            for (var i = 0; i < array.Length; i++)
            {
                _buffer[_position++] = BinaryTypeCode.Float64;
                Unsafe.WriteUnaligned(ref _buffer[_position], array[i]);
                _position += 8;
            }
        }

        public void WriteFloatArrayBulk(float[] array)
        {
            EnsureCapacity(array.Length * 5);
            for (var i = 0; i < array.Length; i++)
            {
                _buffer[_position++] = BinaryTypeCode.Float32;
                Unsafe.WriteUnaligned(ref _buffer[_position], array[i]);
                _position += 4;
            }
        }

        public void WriteGuidArrayBulk(Guid[] array)
        {
            EnsureCapacity(array.Length * 17);
            for (var i = 0; i < array.Length; i++)
            {
                _buffer[_position++] = BinaryTypeCode.Guid;
                array[i].TryWriteBytes(_buffer.AsSpan(_position, 16));
                _position += 16;
            }
        }

        #endregion

        #region SIMD Bulk Copy

        /// <summary>
        /// Copy bytes using SIMD when available, otherwise fall back to standard copy.
        /// Optimized for Blazor WASM where Vector operations are supported.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteBytesSimd(ReadOnlySpan<byte> source)
        {
            EnsureCapacity(source.Length);
            var destination = _buffer.AsSpan(_position, source.Length);

            if (Vector.IsHardwareAccelerated && source.Length >= Vector<byte>.Count * 2)
            {
                CopyWithSimd(source, destination);
            }
            else
            {
                source.CopyTo(destination);
            }

            _position += source.Length;
        }

        /// <summary>
        /// SIMD-optimized memory copy for large buffers.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private static void CopyWithSimd(ReadOnlySpan<byte> source, Span<byte> destination)
        {
            var vectorSize = Vector<byte>.Count;
            var i = 0;
            var length = source.Length;

            // Process full vectors
            var vectorCount = length / vectorSize;
            for (var v = 0; v < vectorCount; v++)
            {
                var vec = new Vector<byte>(source.Slice(i, vectorSize));
                vec.CopyTo(destination.Slice(i, vectorSize));
                i += vectorSize;
            }

            // Copy remaining bytes
            if (i < length)
            {
                source.Slice(i).CopyTo(destination.Slice(i));
            }
        }

        /// <summary>
        /// Write double array using SIMD bulk copy (no per-element type codes).
        /// For use when caller handles type codes separately.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteDoubleBulkRaw(ReadOnlySpan<double> values)
        {
            var byteSpan = MemoryMarshal.AsBytes(values);
            WriteBytesSimd(byteSpan);
        }

        /// <summary>
        /// Write float array using SIMD bulk copy.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteFloatBulkRaw(ReadOnlySpan<float> values)
        {
            var byteSpan = MemoryMarshal.AsBytes(values);
            WriteBytesSimd(byteSpan);
        }

        /// <summary>
        /// Write Guid array using SIMD bulk copy.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteGuidBulkRaw(ReadOnlySpan<Guid> values)
        {
            // Guid is 16 bytes, perfect for SIMD
            var byteLength = values.Length * 16;
            EnsureCapacity(byteLength);

            for (var i = 0; i < values.Length; i++)
            {
                values[i].TryWriteBytes(_buffer.AsSpan(_position, 16));
                _position += 16;
            }
        }

        #endregion

        #region Header

        // Marker-based interning: no footer needed
        // Header: [version][flags][cacheCount (VarUInt, if caching enabled)]
        // Body: data with markers (StringInternFirst, ObjectRefFirst, etc.)

        /// <summary>
        /// Writes the binary header directly. Call AFTER ScanForDuplicates (cacheCount is known).
        /// No placeholder, no shift — single forward write.
        /// Layout: [version (1b)][flags (1b)][cacheCount (VarUInt, if caching)]
        /// </summary>
        public void WriteHeader()
        {
            var flags = BinaryTypeCode.HeaderFlagsBase;
            if (UseMetadata)
                flags |= BinaryTypeCode.HeaderFlag_Metadata;
            if (ReferenceHandling == ReferenceHandlingMode.OnlyId)
                flags |= BinaryTypeCode.HeaderFlag_RefHandling_OnlyId;
            else if (ReferenceHandling == ReferenceHandlingMode.All)
                flags |= (byte)(BinaryTypeCode.HeaderFlag_RefHandling_OnlyId | BinaryTypeCode.HeaderFlag_RefHandling_All);
            if (HasCaching)
                flags |= BinaryTypeCode.HeaderFlag_HasCacheCount;

            WriteByte(AcBinarySerializerOptions.FormatVersion);
            WriteByte(flags);

            if (HasCaching)
            {
                WriteVarUInt((uint)GetCacheCount());
            }
        }

        #endregion

        #region Reference Handling

        //[MethodImpl(MethodImplOptions.AggressiveInlining)]
        //public bool TrackForScanning(object obj) => _refTracker.TrackForScanning(obj);

        /// <summary>
        /// IId-aware tracking for the scan phase.
        /// First checks IId match (different instance, same Id), then falls back to ReferenceEquals.
        /// </summary>
        //[MethodImpl(MethodImplOptions.AggressiveInlining)]
        //public bool TrackForScanningWithIId(object obj, BinarySerializeTypeMetadata metadata, out int existingRefId)
        //{
        //    if (!ReferenceHandling)
        //    {
        //        existingRefId = 0;
        //        return true;  // No tracking needed
        //    }
        //    return _refTracker.TrackForScanningWithIId(obj, metadata, out existingRefId);
        //}

        //[MethodImpl(MethodImplOptions.AggressiveInlining)]
        //public bool ShouldWriteRef(object obj, out int refId) => _refTracker.ShouldWriteId(obj, out refId);

        //[MethodImpl(MethodImplOptions.AggressiveInlining)]
        //public void MarkAsWritten(object obj, int refId) => _refTracker.MarkAsWritten(obj, refId);

        //[MethodImpl(MethodImplOptions.AggressiveInlining)]
        //public bool TryGetExistingRef(object obj, out int refId) => _refTracker.TryGetExistingRef(obj, out refId);

        #endregion

        #region Helpers

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private static void ClearAndTrimIfNeeded<TKey, TValue>(Dictionary<TKey, TValue>? dict, int maxCapacity)
            where TKey : notnull
        {
            if (dict == null)
            {
                return;
            }

            dict.Clear();
            if (dict.EnsureCapacity(0) > maxCapacity)
            {
                dict.TrimExcess();
            }
        }

        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private static void ClearAndTrimIfNeeded<T>(HashSet<T>? set, int maxCapacity)
        {
            if (set == null)
            {
                return;
            }

            set.Clear();
            if (set.EnsureCapacity(0) > maxCapacity)
            {
                set.TrimExcess();
            }
        }

        #endregion

        #region FixStr Methods

        /// <summary>
        /// Write short ASCII string using FixStr encoding (type+length in single byte).
        /// Only call when string is ASCII and length <= 31.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteFixStr(string value)
        {
            var length = value.Length;
            EnsureCapacity(1 + length);
            _buffer[_position++] = BinaryTypeCode.EncodeFixStr(length);
            Ascii.FromUtf16(value.AsSpan(), _buffer.AsSpan(_position, length), out _);
            _position += length;
        }

        /// <summary>
        /// Optimized FixStr write: tries SIMD ASCII conversion, falls back to UTF8.
        /// Single-pass: uses Ascii.FromUtf16 which does validation + copy.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteFixStrDirect(string value)
        {
            var length = value.Length;
            EnsureCapacity(1 + length);

            // Ascii.FromUtf16: SIMD-optimized ASCII conversion
            // Returns actual bytes written - if less than input length, there was a non-ASCII char
            var destSpan = _buffer.AsSpan(_position + 1, length);
            var status = Ascii.FromUtf16(value.AsSpan(), destSpan, out var bytesWritten);

            if (status == System.Buffers.OperationStatus.Done && bytesWritten == length)
            {
                // Success - write FixStr header
                _buffer[_position] = BinaryTypeCode.EncodeFixStr(length);
                _position += 1 + length;
            }
            else
            {
                // Non-ASCII or partial - use standard string encoding
                _buffer[_position++] = BinaryTypeCode.String;
                WriteStringUtf8Internal(value);
            }
        }

        /// <summary>
        /// Internal string write (after String type code already written).
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        private void WriteStringUtf8Internal(string value)
        {
            var byteCount = Utf8NoBom.GetByteCount(value);
            WriteVarUInt((uint)byteCount);
            EnsureCapacity(byteCount);
            Utf8NoBom.GetBytes(value.AsSpan(), _buffer.AsSpan(_position, byteCount));
            _position += byteCount;
        }

        /// <summary>
        /// Write short UTF8 bytes using FixStr encoding.
        /// Only call when byteLength <= 31.
        /// </summary>
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public void WriteFixStrBytes(ReadOnlySpan<byte> utf8Bytes)
        {
            var length = utf8Bytes.Length;
            EnsureCapacity(1 + length);
            _buffer[_position++] = BinaryTypeCode.EncodeFixStr(length);
            utf8Bytes.CopyTo(_buffer.AsSpan(_position, length));
            _position += length;
        }

        #endregion
    }
}