// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.

using System.Buffers;
using System.Diagnostics;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using Internal.Runtime.CompilerServices;

namespace System.Text
{
    public partial class Encoding
    {
        /*
         * This file contains infrastructure code that supports a simplified way of writing
         * internally-implemented Encoding types. In this system, the individual Encoding types
         * are no longer responsible for handling anything related to the EncoderNLS / DecoderNLS
         * infrastructure, nor are they responsible for implementing anything related to fallback
         * buffers logic.
         * 
         * Instead, subclassed types are responsible only for transcoding of individual scalar values
         * to and from the encoding's byte representation (see the two methods immediately below).
         * They can optionally implement fast-path logic to perform bulk transcoding up until the
         * first segment of data that cannot be transcoded. They can special-case certain fallback
         * mechanisms if desired.
         * 
         * Most of the fast-path code is written using raw pointers as the exchange types, just as
         * in the standard Encoding infrastructure. Since the fallback logic is more complex, most
         * of it is written using type-safe constructs like Span<T>, with some amount of glue to
         * allow it to work correctly with pointer-based fast-path code.
         * 
         * A typical call graph for GetBytes is represented below, using ASCIIEncoding as an example.
         * 
         * ASCIIEncoding.GetBytes(...) [non-EncoderNLS path, public virtual override]
         * `- <parameter validation>
         *  - ASCIIEncoding.GetBytesCommon [private helper method per derived type, inlined]
         *    `- ASCIIEncoding.GetBytesFast [overridden fast-path implementation, inlined]
         *     - <if all data transcoded, return immediately>
         *     - <if all data not transcoded...>
         *       `- Encoding.GetBytesWithFallback [non-virtual stub method to call main GetBytesWithFallback worker]
         *          `- Encoding.GetBytesWithFallback [virtual method whose base implementation contains slow fallback logic]
         *             `- <may be overridden to provide optimized fallback logic>
         *              - <create EncodeFallbackBuffer instance>
         *              - <perform the following in a loop:>
         *                `- <invoke fast-path logic via virtual method dispatch on derived type>
         *                 - <read next "bad" scalar value from source>
         *                 - <run this bad value through the fallback buffer>
         *                 - <drain the fallback buffer to the destination>
         *                 - <loop until source is fully consumed or destination is full>
         *              - <signal full or partial success to EncoderNLS instance / throw if necessary>
         * 
         * The call graph for GetBytes(..., EncoderNLS) is similar:
         * 
         * Encoding.GetBytes(..., EncoderNLS) [base implementation]
         * `- <if no leftover data from previous invocation, invoke fast-path>
         *  - <if fast-path invocation above completed, return immediately>
         *  - <if not all data transcoded, or if there was leftover data from previous invocation...>
         *    `- Encoding.GetBytesWithFallback [non-virtual stub method]
         *       `- <drain any leftover data from previous invocation>
         *        - <invoke fast-path again>
         *        - <if all data transcoded, return immediately>
         *        - <if all data not transcoded...>
         *          `- Encoding.GetBytesWithFallback [virtual method as described above]
         *  
         * There are different considerations in each call graph for things like error handling,
         * since the error conditions will be different depending on whether or not an EncoderNLS
         * instance is available and what values its properties have.
         */

        /*
         * THESE TWO METHODS MUST BE OVERRIDDEN BY A SUBCLASSED TYPE
         */

        internal virtual OperationStatus DecodeFirstRune(ReadOnlySpan<byte> bytes, out Rune value, out int bytesConsumed)
        {
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
        }

        internal virtual OperationStatus EncodeRune(Rune value, Span<byte> bytes, out int bytesWritten)
        {
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
        }

        /*
         * ALL OTHER LOGIC CAN BE IMPLEMENTED IN TERMS OF THE TWO METHODS ABOVE.
         * FOR IMPROVED PERFORMANCE, SUBCLASSED TYPES MAY WANT TO OVERRIDE ONE OR MORE VIRTUAL METHODS BELOW.
         */

        /*
         * GETBYTECOUNT FAMILY OF FUNCTIONS
         */

        /// <summary>
        /// Given a <see cref="Rune"/>, determines its byte count under the current <see cref="Encoding"/>.
        /// Returns <see langword="false"/> if the <see cref="Rune"/> cannot be represented in the
        /// current <see cref="Encoding"/>.
        /// </summary>
        internal virtual bool TryGetByteCount(Rune value, out int byteCount)
        {
            // Any production-quality type would override this method and provide a real
            // implementation, so we won't provide a base implementation. However, a
            // non-shipping slow reference implementation is provided below for convenience.

#if false
            Span<byte> bytes = stackalloc byte[4]; // max 4 bytes per input scalar

            OperationStatus opStatus = EncodeRune(value, bytes, out byteCount);
            Debug.Assert(opStatus == OperationStatus.Done || opStatus == OperationStatus.InvalidData, "Unexpected return value.");

            return (opStatus == OperationStatus.Done);
#else
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
#endif
        }

        /// <summary>
        /// Entry point from <see cref="EncoderNLS.GetByteCount"/>.
        /// </summary>
        internal virtual unsafe int GetByteCount(char* pChars, int charCount, EncoderNLS? encoder)
        {
            Debug.Assert(encoder != null, "This code path should only be called from EncoderNLS.");
            Debug.Assert(charCount >= 0, "Caller should've checked this condition.");
            Debug.Assert(pChars != null || charCount == 0, "Cannot provide a null pointer and a non-zero count.");

            // We're going to try to stay on the fast-path as much as we can. That means that we have
            // no leftover data to drain and the entire source buffer can be consumed in a single
            // fast-path invocation. If either of these doesn't hold, we'll go down the slow path of
            // creating spans, draining the EncoderNLS instance, and falling back.

            int totalByteCount = 0;
            int charsConsumed = 0;

            if (!encoder.HasLeftoverData)
            {
                totalByteCount = GetByteCountFast(pChars, charCount, encoder.Fallback, out charsConsumed);
                if (charsConsumed == charCount)
                {
                    return totalByteCount;
                }
            }

            // We had leftover data, or we couldn't consume the entire input buffer.
            // Let's go down the draining + fallback mechanisms.

            totalByteCount += GetByteCountWithFallback(pChars, charCount, charsConsumed, encoder);
            if (totalByteCount < 0)
            {
                ThrowConversionOverflow();
            }

            return totalByteCount;
        }

        /// <summary>
        /// Counts the number of <see langword="byte"/>s that would result from transcoding the source
        /// data, exiting when the source buffer is consumed or when the first unreadable data is encountered.
        /// The implementation may inspect <paramref name="fallback"/> to short-circuit any counting
        /// operation, but it should not attempt to call <see cref="EncoderFallback.CreateFallbackBuffer"/>.
        /// </summary>
        /// <returns>
        /// Via <paramref name="charsConsumed"/>, the number of elements from <paramref name="pChars"/> which
        /// were consumed; and returns the transcoded byte count up to this point.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the byte count would be greater than <see cref="int.MaxValue"/>.
        /// (Implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        /// <remarks>
        /// The implementation should not attempt to perform any sort of fallback behavior.
        /// If custom fallback behavior is necessary, override <see cref="GetByteCountWithFallback"/>.
        /// </remarks>
        private protected virtual unsafe int GetByteCountFast(char* pChars, int charsLength, EncoderFallback? fallback, out int charsConsumed)
        {
            // Any production-quality type would override this method and provide a real
            // implementation, so we won't provide a base implementation. However, a
            // non-shipping slow reference implementation is provided below for convenience.

#if false
            ReadOnlySpan<char> chars = new ReadOnlySpan<char>(pChars, charsLength);
            int totalByteCount = 0;

            while (!chars.IsEmpty)
            {
                if (Rune.DecodeUtf16(chars, out Rune scalarValue, out int charsConsumedThisIteration) != OperationStatus.Done
                    || !TryGetByteCount(scalarValue, out int byteCountThisIteration))
                {
                    // Invalid UTF-16 data, or not convertible to target encoding

                    break;
                }

                chars = chars.Slice(charsConsumedThisIteration);

                totalByteCount += byteCountThisIteration;
                if (totalByteCount < 0)
                {
                    ThrowConversionOverflow();
                }
            }

            charsConsumed = charsLength - chars.Length; // number of chars consumed across all loop iterations above
            return totalByteCount;
#else
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
#endif
        }

        /// <summary>
        /// Counts the number of bytes that would result from transcoding the provided chars,
        /// with no associated <see cref="EncoderNLS"/>. The first two arguments are based on the
        /// original input before invoking this method; and <paramref name="charsConsumedSoFar"/>
        /// signals where in the provided buffer the fallback loop should begin operating.
        /// </summary>
        /// <returns>
        /// The byte count resulting from transcoding the input data.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the resulting byte count is greater than <see cref="int.MaxValue"/>.
        /// (Implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        [MethodImpl(MethodImplOptions.NoInlining)] // don't stack spill spans into our caller
        private protected unsafe int GetByteCountWithFallback(char* pCharsOriginal, int originalCharCount, int charsConsumedSoFar)
        {
            // This is a stub method that's marked "no-inlining" so that it we don't stack-spill spans
            // into our immediate caller. Doing so increases the method prolog in what's supposed to
            // be a very fast path.

            Debug.Assert(0 <= charsConsumedSoFar && charsConsumedSoFar < originalCharCount, "Invalid arguments provided to method.");

            return GetByteCountWithFallback(
                chars: new ReadOnlySpan<char>(pCharsOriginal, originalCharCount).Slice(charsConsumedSoFar),
                originalCharsLength: originalCharCount,
                encoder: null);
        }

        /// <summary>
        /// Gets the number of <see langword="byte"/>s that would result from transcoding the provided
        /// input data, with an associated <see cref="EncoderNLS"/>. The first two arguments are
        /// based on the original input before invoking this method; and <paramref name="charsConsumedSoFar"/>
        /// signals where in the provided source buffer the fallback loop should begin operating.
        /// The behavior of this method is to consume (non-destructively) any leftover data in the
        /// <see cref="EncoderNLS"/> instance, then to invoke the <see cref="GetByteCountFast"/> virtual method
        /// after data has been drained, then to call <see cref="GetByteCountWithFallback(ReadOnlySpan{char}, int, EncoderNLS)"/>.
        /// </summary>
        /// <returns>
        /// The total number of bytes that would result from transcoding the remaining portion of the source buffer.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the return value would exceed <see cref="int.MaxValue"/>.
        /// (The implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        private unsafe int GetByteCountWithFallback(char* pOriginalChars, int originalCharCount, int charsConsumedSoFar, EncoderNLS encoder)
        {
            Debug.Assert(encoder != null, "This code path should only be called from EncoderNLS.");
            Debug.Assert(0 <= charsConsumedSoFar && charsConsumedSoFar <= originalCharCount, "Caller should've checked this condition.");

            // First, try draining any data that already exists on the encoder instance. If we can't complete
            // that operation, there's no point to continuing down to the main workhorse methods.

            ReadOnlySpan<char> chars = new ReadOnlySpan<char>(pOriginalChars, originalCharCount).Slice(charsConsumedSoFar);

            int totalByteCount = encoder.DrainLeftoverDataForGetByteCount(chars, out int charsConsumedJustNow);
            chars = chars.Slice(charsConsumedJustNow);

            // Now try invoking the "fast path" (no fallback) implementation.
            // We can use Unsafe.AsPointer here since these spans are created from pinned data (raw pointers).

            totalByteCount += GetByteCountFast(
                pChars: (char*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(chars)),
                charsLength: chars.Length,
                fallback: encoder.Fallback,
                charsConsumed: out charsConsumedJustNow);

            if (totalByteCount < 0)
            {
                ThrowConversionOverflow();
            }

            chars = chars.Slice(charsConsumedJustNow);

            // If there's still data remaining in the source buffer, go down the fallback path.
            // Otherwise we're finished.

            if (!chars.IsEmpty)
            {
                totalByteCount += GetByteCountWithFallback(chars, originalCharCount, encoder);
                if (totalByteCount < 0)
                {
                    ThrowConversionOverflow();
                }
            }

            return totalByteCount;
        }

        /// <summary>
        /// Counts the number of bytes that would result from transcoding the provided chars,
        /// using the provided <see cref="EncoderFallbackBuffer"/> if necessary.
        /// </summary>
        /// <returns>
        /// The byte count resulting from transcoding the input data.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the resulting byte count is greater than <see cref="int.MaxValue"/>.
        /// (Implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        private protected virtual unsafe int GetByteCountWithFallback(ReadOnlySpan<char> chars, int originalCharsLength, EncoderNLS? encoder)
        {
            Debug.Assert(!chars.IsEmpty, "Caller shouldn't invoke this method with an empty input buffer.");
            Debug.Assert(originalCharsLength >= 0, "Caller provided invalid parameter.");

            // Since we're using Unsafe.AsPointer in our central loop, we want to ensure everything is pinned.

            fixed (char* _pChars_Unused = &MemoryMarshal.GetReference(chars))
            {
                EncoderFallbackBuffer fallbackBuffer = EncoderFallbackBuffer.CreateAndInitialize(this, encoder, originalCharsLength);
                int totalByteCount = 0;

                do
                {
                    // There's still data in the source buffer; why wasn't the previous fast-path able to consume it fully?
                    // There are two scenarios: (a) the source buffer contained invalid / incomplete UTF-16 data;
                    // or (b) the encoding can't translate this scalar value.

                    if (Rune.DecodeFromUtf16(chars, out Rune firstScalarValue, out int charsConsumedThisIteration) == OperationStatus.NeedMoreData
                           && encoder != null
                           && !encoder.MustFlush)
                    {
                        // We saw a standalone high surrogate at the end of the buffer, and the
                        // active EncoderNLS instance isn't asking us to flush. Since a call to
                        // GetBytes would've consumed this char by storing it in EncoderNLS._charLeftOver,
                        // we'll "consume" it by ignoring it. The next call to GetBytes will
                        // pick it up correctly.

                        goto Finish;
                    }

                    // We saw invalid UTF-16 data, or we saw a high surrogate that we need to flush (and
                    // thus treat as invalid), or we saw valid UTF-16 data that this encoder doesn't support.
                    // In any case we'll run it through the fallback mechanism.

                    int byteCountThisIteration = fallbackBuffer.InternalFallbackGetByteCount(chars, out charsConsumedThisIteration);

                    Debug.Assert(byteCountThisIteration >= 0, "Fallback shouldn't have returned a negative value.");
                    Debug.Assert(charsConsumedThisIteration >= 0, "Fallback shouldn't have returned a negative value.");

                    totalByteCount += byteCountThisIteration;
                    if (totalByteCount < 0)
                    {
                        ThrowConversionOverflow();
                    }

                    chars = chars.Slice(charsConsumedThisIteration);

                    if (!chars.IsEmpty)
                    {
                        // Still data remaining - run it through the fast-path to find the next data to fallback.
                        // While building up the tally we need to continually check for integer overflow
                        // since fallbacks can change the total byte count in unexpected ways.

                        byteCountThisIteration = GetByteCountFast(
                            pChars: (char*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(chars)),
                            charsLength: chars.Length,
                            fallback: null, // already tried this earlier and we still fell down the common path, so skip from now on
                            charsConsumed: out charsConsumedThisIteration);

                        Debug.Assert(byteCountThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");
                        Debug.Assert(charsConsumedThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");

                        totalByteCount += byteCountThisIteration;
                        if (totalByteCount < 0)
                        {
                            ThrowConversionOverflow();
                        }

                        chars = chars.Slice(charsConsumedThisIteration);
                    }
                } while (!chars.IsEmpty);

            Finish:

                Debug.Assert(fallbackBuffer.Remaining == 0, "There should be no data in the fallback buffer after GetByteCount.");

                return totalByteCount;
            }
        }

        /*
         * GETBYTES FAMILY OF FUNCTIONS
         */

        /// <summary>
        /// Entry point from <see cref="EncoderNLS.GetBytes"/> and <see cref="EncoderNLS.Convert"/>.
        /// </summary>
        internal virtual unsafe int GetBytes(char* pChars, int charCount, byte* pBytes, int byteCount, EncoderNLS? encoder)
        {
            Debug.Assert(encoder != null, "This code path should only be called from EncoderNLS.");
            Debug.Assert(charCount >= 0, "Caller should've checked this condition.");
            Debug.Assert(pChars != null || charCount == 0, "Cannot provide a null pointer and a non-zero count.");
            Debug.Assert(byteCount >= 0, "Caller should've checked this condition.");
            Debug.Assert(pBytes != null || byteCount == 0, "Cannot provide a null pointer and a non-zero count.");

            // We're going to try to stay on the fast-path as much as we can. That means that we have
            // no leftover data to drain and the entire source buffer can be transcoded in a single
            // fast-path invocation. If either of these doesn't hold, we'll go down the slow path of
            // creating spans, draining the EncoderNLS instance, and falling back.

            int bytesWritten = 0;
            int charsConsumed = 0;

            if (!encoder.HasLeftoverData)
            {
                bytesWritten = GetBytesFast(pChars, charCount, pBytes, byteCount, out charsConsumed);
                if (charsConsumed == charCount)
                {
                    encoder._charsUsed = charCount;
                    return bytesWritten;
                }
            }

            // We had leftover data, or we couldn't consume the entire input buffer.
            // Let's go down the draining + fallback mechanisms.

            return GetBytesWithFallback(pChars, charCount, pBytes, byteCount, charsConsumed, bytesWritten, encoder);
        }

        /// <summary>
        /// Transcodes <see langword="char"/>s to <see langword="byte"/>s, exiting when the source or destination
        /// buffer is consumed or when the first unreadable data is encountered.
        /// </summary>
        /// <returns>
        /// Via <paramref name="charsConsumed"/>, the number of elements from <paramref name="pChars"/> which
        /// were consumed; and returns the number of elements written to <paramref name="pBytes"/>.
        /// </returns>
        /// <remarks>
        /// The implementation should not attempt to perform any sort of fallback behavior.
        /// If custom fallback behavior is necessary, override <see cref="GetBytesWithFallback"/>.
        /// </remarks>
        private protected virtual unsafe int GetBytesFast(char* pChars, int charsLength, byte* pBytes, int bytesLength, out int charsConsumed)
        {
            // Any production-quality type would override this method and provide a real
            // implementation, so we won't provide a base implementation. However, a
            // non-shipping slow reference implementation is provided below for convenience.

#if false
            ReadOnlySpan<char> chars = new ReadOnlySpan<char>(pChars, charsLength);
            Span<byte> bytes = new Span<byte>(pBytes, bytesLength);

            while (!chars.IsEmpty)
            {
                if (Rune.DecodeUtf16(chars, out Rune scalarValue, out int charsConsumedJustNow) != OperationStatus.Done
                    || EncodeRune(scalarValue, bytes, out int bytesWrittenJustNow) != OperationStatus.Done)
                {
                    // Invalid UTF-16 data, or not convertible to target encoding, or destination buffer too small to contain encoded value

                    break;
                }

                chars = chars.Slice(charsConsumedJustNow);
                bytes = bytes.Slice(bytesWrittenJustNow);
            }

            charsConsumed = charsLength - chars.Length; // number of chars consumed across all loop iterations above
            return bytesLength - bytes.Length; // number of bytes written across all loop iterations above
#else
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
#endif
        }

        /// <summary>
        /// Transcodes chars to bytes, with no associated <see cref="EncoderNLS"/>. The first four arguments are
        /// based on the original input before invoking this method; and <paramref name="charsConsumedSoFar"/>
        /// and <paramref name="bytesWrittenSoFar"/> signal where in the provided buffers the fallback loop
        /// should begin operating. The behavior of this method is to call the <see cref="GetBytesWithFallback"/>
        /// virtual method as overridden by the specific type, and failing that go down the shared fallback path.
        /// </summary>
        /// <returns>
        /// The total number of bytes written to <paramref name="pOriginalBytes"/>, including <paramref name="bytesWrittenSoFar"/>.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the destination buffer is not large enough to hold the entirety of the transcoded data.
        /// </exception>
        [MethodImpl(MethodImplOptions.NoInlining)]
        private protected unsafe int GetBytesWithFallback(char* pOriginalChars, int originalCharCount, byte* pOriginalBytes, int originalByteCount, int charsConsumedSoFar, int bytesWrittenSoFar)
        {
            // This is a stub method that's marked "no-inlining" so that it we don't stack-spill spans
            // into our immediate caller. Doing so increases the method prolog in what's supposed to
            // be a very fast path.

            Debug.Assert(0 <= charsConsumedSoFar && charsConsumedSoFar < originalCharCount, "Invalid arguments provided to method.");
            Debug.Assert(0 <= bytesWrittenSoFar && bytesWrittenSoFar <= originalByteCount, "Invalid arguments provided to method.");

            return GetBytesWithFallback(
                chars: new ReadOnlySpan<char>(pOriginalChars, originalCharCount).Slice(charsConsumedSoFar),
                originalCharsLength: originalCharCount,
                bytes: new Span<byte>(pOriginalBytes, originalByteCount).Slice(bytesWrittenSoFar),
                originalBytesLength: originalByteCount,
                encoder: null);
        }

        /// <summary>
        /// Transcodes chars to bytes, with an associated <see cref="EncoderNLS"/>. The first four arguments are
        /// based on the original input before invoking this method; and <paramref name="charsConsumedSoFar"/>
        /// and <paramref name="bytesWrittenSoFar"/> signal where in the provided buffers the fallback loop
        /// should begin operating. The behavior of this method is to drain any leftover data in the
        /// <see cref="EncoderNLS"/> instance, then to invoke the <see cref="GetBytesFast"/> virtual method
        /// after data has been drained, then to call <see cref="GetBytesWithFallback(ReadOnlySpan{char}, int, Span{byte}, int, EncoderNLS)"/>.
        /// </summary>
        /// <returns>
        /// The total number of bytes written to <paramref name="pOriginalBytes"/>, including <paramref name="bytesWrittenSoFar"/>.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the destination buffer is too small to make any forward progress at all, or if the destination buffer is
        /// too small to contain the entirety of the transcoded data and the <see cref="EncoderNLS"/> instance disallows
        /// partial transcoding.
        /// </exception>
        private unsafe int GetBytesWithFallback(char* pOriginalChars, int originalCharCount, byte* pOriginalBytes, int originalByteCount, int charsConsumedSoFar, int bytesWrittenSoFar, EncoderNLS encoder)
        {
            Debug.Assert(encoder != null, "This code path should only be called from EncoderNLS.");
            Debug.Assert(0 <= charsConsumedSoFar && charsConsumedSoFar <= originalCharCount, "Caller should've checked this condition.");
            Debug.Assert(0 <= bytesWrittenSoFar && bytesWrittenSoFar <= originalByteCount, "Caller should've checked this condition.");

            // First, try draining any data that already exists on the encoder instance. If we can't complete
            // that operation, there's no point to continuing down to the main workhorse methods.

            ReadOnlySpan<char> chars = new ReadOnlySpan<char>(pOriginalChars, originalCharCount).Slice(charsConsumedSoFar);
            Span<byte> bytes = new Span<byte>(pOriginalBytes, originalByteCount).Slice(bytesWrittenSoFar);

            bool drainFinishedSuccessfully = encoder.TryDrainLeftoverDataForGetBytes(chars, bytes, out int charsConsumedJustNow, out int bytesWrittenJustNow);

            chars = chars.Slice(charsConsumedJustNow); // whether or not the drain finished, we may have made some progress
            bytes = bytes.Slice(bytesWrittenJustNow);

            if (!drainFinishedSuccessfully)
            {
                ThrowBytesOverflow(encoder, nothingEncoded: bytes.Length == originalByteCount); // might not throw if we wrote at least one byte
            }
            else
            {
                // Now try invoking the "fast path" (no fallback) implementation.
                // We can use Unsafe.AsPointer here since these spans are created from pinned data (raw pointers).

                bytesWrittenJustNow = GetBytesFast(
                    pChars: (char*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(chars)),
                    charsLength: chars.Length,
                    pBytes: (byte*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(bytes)),
                    bytesLength: bytes.Length,
                    charsConsumed: out charsConsumedJustNow);

                chars = chars.Slice(charsConsumedJustNow);
                bytes = bytes.Slice(bytesWrittenJustNow);

                // If there's still data remaining in the source buffer, go down the fallback path.
                // Otherwise we're finished.

                if (!chars.IsEmpty)
                {
                    // We'll optimistically tell the encoder that we're using everything; the
                    // GetBytesWithFallback method will overwrite this field if necessary.

                    encoder._charsUsed = originalCharCount;
                    return GetBytesWithFallback(chars, originalCharCount, bytes, originalByteCount, encoder);
                }
            }

            encoder._charsUsed = originalCharCount - chars.Length; // total number of characters consumed up until now
            return originalByteCount - bytes.Length; // total number of bytes written up until now
        }

        /// <summary>
        /// Transcodes chars to bytes, using <see cref="Encoding.EncoderFallback"/> or <see cref="Encoder.Fallback"/> if needed.
        /// </summary>
        /// <returns>
        /// The total number of bytes written to <paramref name="bytes"/> (based on <paramref name="originalBytesLength"/>).
        /// </returns>
        /// <remarks>
        /// The derived class should override this method if it might be able to provide a more optimized fallback
        /// implementation, deferring to the base implementation if needed. This method calls <see cref="ThrowBytesOverflow"/>
        /// if necessary.
        /// </remarks>
        private protected virtual unsafe int GetBytesWithFallback(ReadOnlySpan<char> chars, int originalCharsLength, Span<byte> bytes, int originalBytesLength, EncoderNLS? encoder)
        {
            Debug.Assert(!chars.IsEmpty, "Caller shouldn't invoke this method with an empty input buffer.");
            Debug.Assert(originalCharsLength >= 0, "Caller provided invalid parameter.");
            Debug.Assert(originalBytesLength >= 0, "Caller provided invalid parameter.");

            // Since we're using Unsafe.AsPointer in our central loop, we want to ensure everything is pinned.

            fixed (char* _pChars_Unused = &MemoryMarshal.GetReference(chars))
            fixed (byte* _pBytes_Unused = &MemoryMarshal.GetReference(bytes))
            {
                EncoderFallbackBuffer fallbackBuffer = EncoderFallbackBuffer.CreateAndInitialize(this, encoder, originalCharsLength);

                do
                {
                    // There's still data in the source buffer; why wasn't the previous fast-path able to consume it fully?
                    // There are two scenarios: (a) the source buffer contained invalid / incomplete UTF-16 data;
                    // or (b) the encoding can't translate this scalar value.

                    switch (Rune.DecodeFromUtf16(chars, out Rune firstScalarValue, out int charsConsumedThisIteration))
                    {
                        case OperationStatus.NeedMoreData:
                            Debug.Assert(charsConsumedThisIteration == chars.Length, "If returning NeedMoreData, should out the entire buffer length as chars consumed.");
                            if (encoder is null || encoder.MustFlush)
                            {
                                goto case OperationStatus.InvalidData; // see comment in GetByteCountWithFallback
                            }
                            else
                            {
                                encoder._charLeftOver = chars[0]; // squirrel away remaining high surrogate char and finish
                                chars = ReadOnlySpan<char>.Empty;
                                goto Finish;
                            }

                        case OperationStatus.InvalidData:
                            break;

                        default:
                            if (EncodeRune(firstScalarValue, bytes, out _) == OperationStatus.DestinationTooSmall)
                            {
                                goto Finish; // source buffer contained valid UTF-16 but encoder ran out of space in destination buffer
                            }
                            break; // source buffer contained valid UTF-16 but encoder doesn't support this scalar value
                    }

                    // Now we know the reason for failure was that the original input was invalid
                    // for the encoding in use. Run it through the fallback mechanism.

                    bool fallbackFinished = fallbackBuffer.TryInternalFallbackGetBytes(chars, bytes, out charsConsumedThisIteration, out int bytesWrittenThisIteration);

                    // Regardless of whether the fallback finished, it did consume some number of
                    // chars, and it may have written some number of bytes.

                    chars = chars.Slice(charsConsumedThisIteration);
                    bytes = bytes.Slice(bytesWrittenThisIteration);

                    if (!fallbackFinished)
                    {
                        goto Finish; // fallback has pending state - it'll get written out on the next GetBytes call
                    }

                    if (!chars.IsEmpty)
                    {
                        // Still data remaining - run it through the fast-path to find the next data to fallback.

                        bytesWrittenThisIteration = GetBytesFast(
                            pChars: (char*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(chars)),
                            charsLength: chars.Length,
                            pBytes: (byte*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(bytes)),
                            bytesLength: bytes.Length,
                            charsConsumed: out charsConsumedThisIteration);

                        Debug.Assert(bytesWrittenThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");
                        Debug.Assert(charsConsumedThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");

                        chars = chars.Slice(charsConsumedThisIteration);
                        bytes = bytes.Slice(bytesWrittenThisIteration);
                    }
                } while (!chars.IsEmpty);

            Finish:

                // We reach this point when we deplete the source or destination buffer. There are a few
                // cases to consider now. If the source buffer has been fully consumed and there's no
                // leftover data in the EncoderNLS or the fallback buffer, we've completed transcoding.
                // If the source buffer isn't empty or there's leftover data in the fallback buffer,
                // it means we ran out of space in the destintion buffer. This is an unrecoverable error
                // if no EncoderNLS is in use (because only EncoderNLS can handle partial success), and
                // even if an EncoderNLS is in use this is only recoverable if the EncoderNLS instance
                // allows partial completion. Let's check all of these conditions now.

                if (!chars.IsEmpty || fallbackBuffer.Remaining > 0)
                {
                    // The line below will also throw if the encoder couldn't make any progress at all
                    // because the output buffer wasn't large enough to contain the result of even
                    // a single scalar conversion or fallback.

                    ThrowBytesOverflow(encoder, nothingEncoded: bytes.Length == originalBytesLength);
                }

                // If an EncoderNLS instance is active, update its "total consumed character count" value.

                if (encoder != null)
                {
                    Debug.Assert(originalCharsLength >= chars.Length, "About to report a negative number of chars used?");
                    encoder._charsUsed = originalCharsLength - chars.Length; // number of chars consumed
                }

                Debug.Assert(fallbackBuffer.Remaining == 0 || encoder != null, "Shouldn't have any leftover data in fallback buffer unless an EncoderNLS is in use.");

                return originalBytesLength - bytes.Length;
            }
        }

        /*
         * GETCHARCOUNT FAMILY OF FUNCTIONS
         */

        /// <summary>
        /// Entry point from <see cref="DecoderNLS.GetCharCount"/>.
        /// </summary>
        internal virtual unsafe int GetCharCount(byte* pBytes, int byteCount, DecoderNLS? decoder)
        {
            Debug.Assert(decoder != null, "This code path should only be called from DecoderNLS.");
            Debug.Assert(byteCount >= 0, "Caller should've checked this condition.");
            Debug.Assert(pBytes != null || byteCount == 0, "Cannot provide a null pointer and a non-zero count.");

            // We're going to try to stay on the fast-path as much as we can. That means that we have
            // no leftover data to drain and the entire source buffer can be consumed in a single
            // fast-path invocation. If either of these doesn't hold, we'll go down the slow path of
            // creating spans, draining the DecoderNLS instance, and falling back.

            Debug.Assert(!decoder.InternalHasFallbackBuffer || decoder.FallbackBuffer.Remaining == 0, "Fallback buffer can't hold data between GetChars invocations.");

            int totalCharCount = 0;
            int bytesConsumed = 0;

            if (!decoder.HasLeftoverData)
            {
                totalCharCount = GetCharCountFast(pBytes, byteCount, decoder.Fallback, out bytesConsumed);
                if (bytesConsumed == byteCount)
                {
                    return totalCharCount;
                }
            }

            // We had leftover data, or we couldn't consume the entire input buffer.
            // Let's go down the draining + fallback mechanisms.

            totalCharCount += GetCharCountWithFallback(pBytes, byteCount, bytesConsumed, decoder);
            if (totalCharCount < 0)
            {
                ThrowConversionOverflow();
            }

            return totalCharCount;
        }

        /// <summary>
        /// Counts the number of <see langword="char"/>s that would result from transcoding the source
        /// data, exiting when the source buffer is consumed or when the first unreadable data is encountered.
        /// The implementation may inspect <paramref name="fallback"/> to short-circuit any counting
        /// operation, but it should not attempt to call <see cref="DecoderFallback.CreateFallbackBuffer"/>.
        /// </summary>
        /// <returns>
        /// Via <paramref name="bytesConsumed"/>, the number of elements from <paramref name="pBytes"/> which
        /// were consumed; and returns the transcoded char count up to this point.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the char count would be greater than <see cref="int.MaxValue"/>.
        /// (Implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        /// <remarks>
        /// The implementation should not attempt to perform any sort of fallback behavior.
        /// If custom fallback behavior is necessary, override <see cref="GetCharCountWithFallback"/>.
        /// </remarks>
        private protected virtual unsafe int GetCharCountFast(byte* pBytes, int bytesLength, DecoderFallback? fallback, out int bytesConsumed)
        {
            // Any production-quality type would override this method and provide a real
            // implementation, so we won't provide a base implementation. However, a
            // non-shipping slow reference implementation is provided below for convenience.

#if false
            ReadOnlySpan<byte> bytes = new ReadOnlySpan<byte>(pBytes, bytesLength);
            int totalCharCount = 0;

            while (!bytes.IsEmpty)
            {
                // We don't care about statuses other than Done. The fallback mechanism will handle those.

                if (DecodeFirstRune(bytes, out Rune value, out int bytesConsumedJustNow) != OperationStatus.Done)
                {
                    break;
                }

                totalCharCount += value.Utf16SequenceLength;
                if (totalCharCount < 0)
                {
                    ThrowConversionOverflow();
                }

                bytes = bytes.Slice(bytesConsumedJustNow);
            }

            bytesConsumed = bytesLength - bytes.Length; // number of bytes consumed across all loop iterations above
            return totalCharCount;
#else
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
#endif
        }

        /// <summary>
        /// Counts the number of chars that would result from transcoding the provided bytes,
        /// with no associated <see cref="DecoderNLS"/>. The first two arguments are based on the
        /// original input before invoking this method; and <paramref name="bytesConsumedSoFar"/>
        /// signals where in the provided buffer the fallback loop should begin operating.
        /// </summary>
        /// <returns>
        /// The char count resulting from transcoding the input data.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the resulting char count is greater than <see cref="int.MaxValue"/>.
        /// (Implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        [MethodImpl(MethodImplOptions.NoInlining)] // don't stack spill spans into our caller
        private protected unsafe int GetCharCountWithFallback(byte* pBytesOriginal, int originalByteCount, int bytesConsumedSoFar)
        {
            // This is a stub method that's marked "no-inlining" so that it we don't stack-spill spans
            // into our immediate caller. Doing so increases the method prolog in what's supposed to
            // be a very fast path.

            Debug.Assert(0 <= bytesConsumedSoFar && bytesConsumedSoFar < originalByteCount, "Invalid arguments provided to method.");

            return GetCharCountWithFallback(
                bytes: new ReadOnlySpan<byte>(pBytesOriginal, originalByteCount).Slice(bytesConsumedSoFar),
                originalBytesLength: originalByteCount,
                decoder: null);
        }

        /// <summary>
        /// Gets the number of <see langword="char"/>s that would result from transcoding the provided
        /// input data, with an associated <see cref="DecoderNLS"/>. The first two arguments are
        /// based on the original input before invoking this method; and <paramref name="bytesConsumedSoFar"/>
        /// signals where in the provided source buffer the fallback loop should begin operating.
        /// The behavior of this method is to consume (non-destructively) any leftover data in the
        /// <see cref="DecoderNLS"/> instance, then to invoke the <see cref="GetCharCountFast"/> virtual method
        /// after data has been drained, then to call <see cref="GetCharCountWithFallback(ReadOnlySpan{byte}, int, DecoderNLS)"/>.
        /// </summary>
        /// <returns>
        /// The total number of chars that would result from transcoding the remaining portion of the source buffer.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the return value would exceed <see cref="int.MaxValue"/>.
        /// (The implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        private unsafe int GetCharCountWithFallback(byte* pOriginalBytes, int originalByteCount, int bytesConsumedSoFar, DecoderNLS decoder)
        {
            Debug.Assert(decoder != null, "This code path should only be called from DecoderNLS.");
            Debug.Assert(0 <= bytesConsumedSoFar && bytesConsumedSoFar <= originalByteCount, "Caller should've checked this condition.");

            // First, try draining any data that already exists on the decoder instance. If we can't complete
            // that operation, there's no point to continuing down to the main workhorse methods.

            ReadOnlySpan<byte> bytes = new ReadOnlySpan<byte>(pOriginalBytes, originalByteCount).Slice(bytesConsumedSoFar);

            int bytesConsumedJustNow = 0;
            int totalCharCount = 0;

            if (decoder.HasLeftoverData)
            {
                totalCharCount = decoder.DrainLeftoverDataForGetCharCount(bytes, out bytesConsumedJustNow);
                bytes = bytes.Slice(bytesConsumedJustNow);
            }

            // Now try invoking the "fast path" (no fallback) implementation.
            // We can use Unsafe.AsPointer here since these spans are created from pinned data (raw pointers).

            totalCharCount += GetCharCountFast(
                pBytes: (byte*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(bytes)),
                bytesLength: bytes.Length,
                fallback: decoder.Fallback,
                bytesConsumed: out bytesConsumedJustNow);

            if (totalCharCount < 0)
            {
                ThrowConversionOverflow();
            }

            bytes = bytes.Slice(bytesConsumedJustNow);

            // If there's still data remaining in the source buffer, go down the fallback path.
            // Otherwise we're finished.

            if (!bytes.IsEmpty)
            {
                totalCharCount += GetCharCountWithFallback(bytes, originalByteCount, decoder);
                if (totalCharCount < 0)
                {
                    ThrowConversionOverflow();
                }
            }

            return totalCharCount;
        }

        /// <summary>
        /// Counts the number of chars that would result from transcoding the provided bytes,
        /// using the provided <see cref="DecoderFallbackBuffer"/> if necessary.
        /// </summary>
        /// <returns>
        /// The char count resulting from transcoding the input data.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the resulting char count is greater than <see cref="int.MaxValue"/>.
        /// (Implementation should call <see cref="ThrowConversionOverflow"/>.)
        /// </exception>
        private unsafe int GetCharCountWithFallback(ReadOnlySpan<byte> bytes, int originalBytesLength, DecoderNLS? decoder)
        {
            Debug.Assert(!bytes.IsEmpty, "Caller shouldn't invoke this method with an empty input buffer.");
            Debug.Assert(originalBytesLength >= 0, "Caller provided invalid parameter.");

            // Since we're using Unsafe.AsPointer in our central loop, we want to ensure everything is pinned.

            fixed (byte* _pBytes_Unused = &MemoryMarshal.GetReference(bytes))
            {
                DecoderFallbackBuffer fallbackBuffer = DecoderFallbackBuffer.CreateAndInitialize(this, decoder, originalBytesLength);
                int totalCharCount = 0;

                do
                {
                    // There's still data in the source buffer; why wasn't the previous fast-path able to consume it fully?
                    // There are two scenarios: (a) the source buffer contained invalid data, or it contained incomplete data.

                    if (DecodeFirstRune(bytes, out Rune firstScalarValue, out int bytesConsumedThisIteration) == OperationStatus.NeedMoreData
                          && decoder != null
                          && !decoder.MustFlush)
                    {
                        // We saw incomplete data at the end of the buffer, and the active DecoderNLS isntance
                        // isn't asking us to flush. Since a call to GetChars would've consumed this data by
                        // storing it in the DecoderNLS instance, we'll "consume" it by ignoring it.
                        // The next call to GetChars will pick it up correctly.

                        goto Finish;
                    }

                    // We saw invalid binary data, or we saw incomplete data that we need to flush (and thus
                    // treat as invalid). In any case we'll run through the fallback mechanism.

                    int charCountThisIteration = fallbackBuffer.InternalFallbackGetCharCount(bytes, bytesConsumedThisIteration);

                    Debug.Assert(charCountThisIteration >= 0, "Fallback shouldn't have returned a negative value.");

                    totalCharCount += charCountThisIteration;
                    if (totalCharCount < 0)
                    {
                        ThrowConversionOverflow();
                    }

                    bytes = bytes.Slice(bytesConsumedThisIteration);

                    if (!bytes.IsEmpty)
                    {
                        // Still data remaining - run it through the fast-path to find the next data to fallback.
                        // While building up the tally we need to continually check for integer overflow
                        // since fallbacks can change the total byte count in unexpected ways.

                        charCountThisIteration = GetCharCountFast(
                            pBytes: (byte*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(bytes)),
                            bytesLength: bytes.Length,
                            fallback: null, // wasn't able to be short-circuited by our caller; don't bother trying again
                            bytesConsumed: out bytesConsumedThisIteration);

                        Debug.Assert(charCountThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");
                        Debug.Assert(bytesConsumedThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");

                        totalCharCount += charCountThisIteration;
                        if (totalCharCount < 0)
                        {
                            ThrowConversionOverflow();
                        }

                        bytes = bytes.Slice(bytesConsumedThisIteration);
                    }
                } while (!bytes.IsEmpty);

            Finish:

                Debug.Assert(fallbackBuffer.Remaining == 0, "There should be no data in the fallback buffer after GetCharCount.");

                return totalCharCount;
            }
        }

        /*
         * GETCHARS FAMILY OF FUNCTIONS
         */

        /// <summary>
        /// Entry point from <see cref="DecoderNLS.GetChars"/> and <see cref="DecoderNLS.Convert"/>.
        /// </summary>
        internal virtual unsafe int GetChars(byte* pBytes, int byteCount, char* pChars, int charCount, DecoderNLS? decoder)
        {
            Debug.Assert(decoder != null, "This code path should only be called from DecoderNLS.");
            Debug.Assert(byteCount >= 0, "Caller should've checked this condition.");
            Debug.Assert(pBytes != null || byteCount == 0, "Cannot provide a null pointer and a non-zero count.");
            Debug.Assert(charCount >= 0, "Caller should've checked this condition.");
            Debug.Assert(pChars != null || charCount == 0, "Cannot provide a null pointer and a non-zero count.");

            // We're going to try to stay on the fast-path as much as we can. That means that we have
            // no leftover data to drain and the entire source buffer can be transcoded in a single
            // fast-path invocation. If either of these doesn't hold, we'll go down the slow path of
            // creating spans, draining the DecoderNLS instance, and falling back.

            int charsWritten = 0;
            int bytesConsumed = 0;

            if (!decoder.HasLeftoverData)
            {
                charsWritten = GetCharsFast(pBytes, byteCount, pChars, charCount, out bytesConsumed);
                if (bytesConsumed == byteCount)
                {
                    decoder._bytesUsed = byteCount;
                    return charsWritten;
                }
            }

            // We had leftover data, or we couldn't consume the entire input buffer.
            // Let's go down the draining + fallback mechanisms.

            return GetCharsWithFallback(pBytes, byteCount, pChars, charCount, bytesConsumed, charsWritten, decoder);
        }

        /// <summary>
        /// Transcodes <see langword="byte"/>s to <see langword="char"/>s, exiting when the source or destination
        /// buffer is consumed or when the first unreadable data is encountered.
        /// </summary>
        /// <returns>
        /// Via <paramref name="bytesConsumed"/>, the number of elements from <paramref name="pBytes"/> which
        /// were consumed; and returns the number of elements written to <paramref name="pChars"/>.
        /// </returns>
        /// <remarks>
        /// The implementation should not attempt to perform any sort of fallback behavior.
        /// If custom fallback behavior is necessary, override <see cref="GetCharsWithFallback"/>.
        /// </remarks>
        private protected virtual unsafe int GetCharsFast(byte* pBytes, int bytesLength, char* pChars, int charsLength, out int bytesConsumed)
        {
            // Any production-quality type would override this method and provide a real
            // implementation, so we won't provide a base implementation. However, a
            // non-shipping slow reference implementation is provided below for convenience.

#if false
            ReadOnlySpan<byte> bytes = new ReadOnlySpan<byte>(pBytes, bytesLength);
            Span<char> chars = new Span<char>(pChars, charsLength);

            while (!bytes.IsEmpty)
            {
                if ((DecodeFirstRune(bytes, out Rune firstScalarValue, out int bytesConsumedJustNow) != OperationStatus.Done)
                    || !firstScalarValue.TryEncode(chars, out int charsWrittenJustNow))
                {
                    // Invalid or incomplete binary data, or destination buffer too small to contain decoded value

                    break;
                }

                bytes = bytes.Slice(bytesConsumedJustNow);
                chars = chars.Slice(charsWrittenJustNow);
            }

            bytesConsumed = bytesLength - bytes.Length; // number of bytes consumed across all loop iterations above
            return charsLength - chars.Length; // number of chars written across all loop iterations above
#else
            Debug.Fail("This should be overridden by a subclassed type.");
            throw NotImplemented.ByDesign;
#endif
        }

        /// <summary>
        /// Transcodes bytes to chars, with no associated <see cref="DecoderNLS"/>. The first four arguments are
        /// based on the original input before invoking this method; and <paramref name="bytesConsumedSoFar"/>
        /// and <paramref name="charsWrittenSoFar"/> signal where in the provided buffers the fallback loop
        /// should begin operating. The behavior of this method is to call the <see cref="GetCharsWithFallback"/>
        /// virtual method as overridden by the specific type, and failing that go down the shared fallback path.
        /// </summary>
        /// <returns>
        /// The total number of chars written to <paramref name="pOriginalChars"/>, including <paramref name="charsWrittenSoFar"/>.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the destination buffer is not large enough to hold the entirety of the transcoded data.
        /// </exception>
        [MethodImpl(MethodImplOptions.NoInlining)]
        private protected unsafe int GetCharsWithFallback(byte* pOriginalBytes, int originalByteCount, char* pOriginalChars, int originalCharCount, int bytesConsumedSoFar, int charsWrittenSoFar)
        {
            // This is a stub method that's marked "no-inlining" so that it we don't stack-spill spans
            // into our immediate caller. Doing so increases the method prolog in what's supposed to
            // be a very fast path.

            Debug.Assert(0 <= bytesConsumedSoFar && bytesConsumedSoFar < originalByteCount, "Invalid arguments provided to method.");
            Debug.Assert(0 <= charsWrittenSoFar && charsWrittenSoFar <= originalCharCount, "Invalid arguments provided to method.");

            return GetCharsWithFallback(
                bytes: new ReadOnlySpan<byte>(pOriginalBytes, originalByteCount).Slice(bytesConsumedSoFar),
                originalBytesLength: originalByteCount,
                chars: new Span<char>(pOriginalChars, originalCharCount).Slice(charsWrittenSoFar),
                originalCharsLength: originalCharCount,
                decoder: null);
        }

        /// <summary>
        /// Transcodes bytes to chars, with an associated <see cref="DecoderNLS"/>. The first four arguments are
        /// based on the original input before invoking this method; and <paramref name="bytesConsumedSoFar"/>
        /// and <paramref name="charsWrittenSoFar"/> signal where in the provided buffers the fallback loop
        /// should begin operating. The behavior of this method is to drain any leftover data in the
        /// <see cref="DecoderNLS"/> instance, then to invoke the <see cref="GetCharsFast"/> virtual method
        /// after data has been drained, then to call <see cref="GetCharsWithFallback(ReadOnlySpan{byte}, int, Span{char}, int, DecoderNLS)"/>.
        /// </summary>
        /// <returns>
        /// The total number of chars written to <paramref name="pOriginalChars"/>, including <paramref name="charsWrittenSoFar"/>.
        /// </returns>
        /// <exception cref="ArgumentException">
        /// If the destination buffer is too small to make any forward progress at all, or if the destination buffer is
        /// too small to contain the entirety of the transcoded data and the <see cref="DecoderNLS"/> instance disallows
        /// partial transcoding.
        /// </exception>
        private protected unsafe int GetCharsWithFallback(byte* pOriginalBytes, int originalByteCount, char* pOriginalChars, int originalCharCount, int bytesConsumedSoFar, int charsWrittenSoFar, DecoderNLS decoder)
        {
            Debug.Assert(decoder != null, "This code path should only be called from DecoderNLS.");
            Debug.Assert(0 <= bytesConsumedSoFar && bytesConsumedSoFar <= originalByteCount, "Caller should've checked this condition.");
            Debug.Assert(0 <= charsWrittenSoFar && charsWrittenSoFar <= originalCharCount, "Caller should've checked this condition.");

            // First, try draining any data that already exists on the encoder instance. If we can't complete
            // that operation, there's no point to continuing down to the main workhorse methods.
            //
            // Like GetBytes, there may be leftover data in the DecoderNLS instance. But unlike GetBytes,
            // the bytes -> chars conversion doesn't allow leftover data in the fallback buffer. This means
            // that the drain operation below will either succeed fully or fail; there's no partial success
            // condition as with the chars -> bytes conversion. The drain method will throw if there's not
            // enough space in the destination buffer.

            ReadOnlySpan<byte> bytes = new ReadOnlySpan<byte>(pOriginalBytes, originalByteCount).Slice(bytesConsumedSoFar);
            Span<char> chars = new Span<char>(pOriginalChars, originalCharCount).Slice(charsWrittenSoFar);

            int bytesConsumedJustNow = 0;
            int charsWrittenJustNow = 0;

            if (decoder.HasLeftoverData)
            {
                charsWrittenJustNow = decoder.DrainLeftoverDataForGetChars(bytes, chars, out bytesConsumedJustNow);
                bytes = bytes.Slice(bytesConsumedJustNow);
                chars = chars.Slice(charsWrittenJustNow);
            }

            Debug.Assert(!decoder.InternalHasFallbackBuffer || decoder.FallbackBuffer.Remaining == 0, "Should be no remaining fallback data at this point.");

            // Now try invoking the "fast path" (no fallback buffer) implementation.
            // We can use Unsafe.AsPointer here since these spans are created from pinned data (raw pointers).

            charsWrittenJustNow = GetCharsFast(
                pBytes: (byte*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(bytes)),
                bytesLength: bytes.Length,
                pChars: (char*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(chars)),
                charsLength: chars.Length,
                bytesConsumed: out bytesConsumedJustNow);

            bytes = bytes.Slice(bytesConsumedJustNow);
            chars = chars.Slice(charsWrittenJustNow);

            // We'll optimistically tell the decoder that we're using everything; the
            // GetCharsWithFallback method will overwrite this field if necessary.

            decoder._bytesUsed = originalByteCount;

            if (bytes.IsEmpty)
            {
                return originalCharCount - chars.Length; // total number of chars written
            }
            else
            {
                return GetCharsWithFallback(bytes, originalByteCount, chars, originalCharCount, decoder);
            }
        }

        /// <summary>
        /// Transcodes bytes to chars, using <see cref="Encoding.DecoderFallback"/> or <see cref="Decoder.Fallback"/> if needed.
        /// </summary>
        /// <returns>
        /// The total number of chars written to <paramref name="chars"/> (based on <paramref name="originalCharsLength"/>).
        /// </returns>
        /// <remarks>
        /// The derived class should override this method if it might be able to provide a more optimized fallback
        /// implementation, deferring to the base implementation if needed. This method calls <see cref="ThrowCharsOverflow"/>
        /// if necessary.
        /// </remarks>
        private protected virtual unsafe int GetCharsWithFallback(ReadOnlySpan<byte> bytes, int originalBytesLength, Span<char> chars, int originalCharsLength, DecoderNLS? decoder)
        {
            Debug.Assert(!bytes.IsEmpty, "Caller shouldn't invoke this method with an empty input buffer.");
            Debug.Assert(originalBytesLength >= 0, "Caller provided invalid parameter.");
            Debug.Assert(originalCharsLength >= 0, "Caller provided invalid parameter.");

            // Since we're using Unsafe.AsPointer in our central loop, we want to ensure everything is pinned.

            fixed (byte* _pBytes_Unused = &MemoryMarshal.GetReference(bytes))
            fixed (char* _pChars_Unused = &MemoryMarshal.GetReference(chars))
            {
                DecoderFallbackBuffer fallbackBuffer = DecoderFallbackBuffer.CreateAndInitialize(this, decoder, originalBytesLength);

                do
                {
                    // There's still data in the source buffer; why wasn't the previous fast-path able to consume it fully?
                    // There are two scenarios: (a) the source buffer contained invalid data, or it contained incomplete data.

                    int charsWrittenThisIteration;

                    switch (DecodeFirstRune(bytes, out _, out int bytesConsumedThisIteration))
                    {
                        case OperationStatus.NeedMoreData:
                            Debug.Assert(bytesConsumedThisIteration == bytes.Length, "If returning NeedMoreData, should out the entire buffer length as bytes consumed.");
                            if (decoder is null || decoder.MustFlush)
                            {
                                goto case OperationStatus.InvalidData; // see comment in GetCharCountWithFallback
                            }
                            else
                            {
                                decoder.SetLeftoverData(bytes); // squirrel away remaining data and finish
                                bytes = ReadOnlySpan<byte>.Empty;
                                goto Finish;
                            }

                        case OperationStatus.InvalidData:
                            if (fallbackBuffer.TryInternalFallbackGetChars(bytes, bytesConsumedThisIteration, chars, out charsWrittenThisIteration))
                            {
                                // We successfully consumed some bytes, sent it through the fallback, and wrote some chars.

                                Debug.Assert(charsWrittenThisIteration >= 0, "Fallback shouldn't have returned a negative value.");
                                break;
                            }
                            else
                            {
                                // We generated fallback data, but the destination buffer wasn't large enough to hold it.
                                // Don't mark any of the bytes we ran through the fallback as consumed, and terminate
                                // the loop now and let our caller handle this condition.

                                goto Finish;
                            }

                        default:
                            goto Finish; // no error on input, so destination must have been too small
                    }

                    bytes = bytes.Slice(bytesConsumedThisIteration);
                    chars = chars.Slice(charsWrittenThisIteration);

                    if (!bytes.IsEmpty)
                    {
                        // Still data remaining - run it through the fast-path to find the next data to fallback.
                        // We need to figure out why we weren't able to make progress.

                        charsWrittenThisIteration = GetCharsFast(
                            pBytes: (byte*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(bytes)),
                            bytesLength: bytes.Length,
                            pChars: (char*)Unsafe.AsPointer(ref MemoryMarshal.GetReference(chars)),
                            charsLength: chars.Length,
                            bytesConsumed: out bytesConsumedThisIteration);

                        Debug.Assert(charsWrittenThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");
                        Debug.Assert(bytesConsumedThisIteration >= 0, "Workhorse shouldn't have returned a negative value.");

                        bytes = bytes.Slice(bytesConsumedThisIteration);
                        chars = chars.Slice(charsWrittenThisIteration);
                    }
                } while (!bytes.IsEmpty);

            Finish:

                // We reach this point when we deplete the source or destination buffer. See main comment
                // at the end of GetBytesWithFallback for how the below logic works; the primary difference
                // here is that GetChars disallows leftover data in the fallback buffer between calls.

                Debug.Assert(fallbackBuffer.Remaining == 0);

                if (!bytes.IsEmpty)
                {
                    // The line below will also throw if the decoder couldn't make any progress at all
                    // because the output buffer wasn't large enough to contain the result of even
                    // a single scalar conversion or fallback.

                    ThrowCharsOverflow(decoder, nothingDecoded: chars.Length == originalCharsLength);
                }

                // If a DecoderNLS instance is active, update its "total consumed byte count" value.

                if (decoder != null)
                {
                    Debug.Assert(originalBytesLength >= bytes.Length, "About to report a negative number of bytes used?");
                    decoder._bytesUsed = originalBytesLength - bytes.Length; // number of bytes consumed
                }

                return originalCharsLength - chars.Length; // total number of chars written
            }
        }
    }
}