c# – StreamReader and seeking

I realize this is really belated, but I just stumbled onto this incredible flaw in StreamReader myself; the fact that you can’t reliably seek when using StreamReader. Personally, my specific need is to have the ability to read characters, but then “back up” if a certain condition is met; it’s a side effect of one of the file formats I’m parsing.

Using ReadLine() isn’t an option because it’s only useful in really trivial parsing jobs. I have to support configurable record/line delimiter sequences and support escape delimiter sequences. Also, I don’t want to implement my own buffer so I can support “backing up” and escape sequences; that should be the StreamReader‘s job.

This method calculates the actual position in the underlying stream of bytes on-demand. It works for UTF8, UTF-16LE, UTF-16BE, UTF-32LE, UTF-32BE, and any single-byte encoding (e.g. code pages 1252, 437, 28591, etc.), regardless the presence of a preamble/BOM. This version will not work for UTF-7, Shift-JIS, or other variable-byte encodings.

When I need to seek to an arbitrary position in the underlying stream, I directly set BaseStream.Position and then call DiscardBufferedData() to get StreamReader back in sync for the next Read()/Peek() call.

And a friendly reminder: don’t arbitrarily set BaseStream.Position. If you bisect a character, you’ll invalidate the next Read() and, for UTF-16/-32, you’ll also invalidate the result of this method.

public static long GetActualPosition(StreamReader reader)
{
    System.Reflection.BindingFlags flags = System.Reflection.BindingFlags.DeclaredOnly | System.Reflection.BindingFlags.NonPublic | System.Reflection.BindingFlags.Instance | System.Reflection.BindingFlags.GetField;

    // The current buffer of decoded characters
    char[] charBuffer = (char[])reader.GetType().InvokeMember("charBuffer", flags, null, reader, null);

    // The index of the next char to be read from charBuffer
    int charPos = (int)reader.GetType().InvokeMember("charPos", flags, null, reader, null);

    // The number of decoded chars presently used in charBuffer
    int charLen = (int)reader.GetType().InvokeMember("charLen", flags, null, reader, null);

    // The current buffer of read bytes (byteBuffer.Length = 1024; this is critical).
    byte[] byteBuffer = (byte[])reader.GetType().InvokeMember("byteBuffer", flags, null, reader, null);

    // The number of bytes read while advancing reader.BaseStream.Position to (re)fill charBuffer
    int byteLen = (int)reader.GetType().InvokeMember("byteLen", flags, null, reader, null);

    // The number of bytes the remaining chars use in the original encoding.
    int numBytesLeft = reader.CurrentEncoding.GetByteCount(charBuffer, charPos, charLen - charPos);

    // For variable-byte encodings, deal with partial chars at the end of the buffer
    int numFragments = 0;
    if (byteLen > 0 && !reader.CurrentEncoding.IsSingleByte)
    {
        if (reader.CurrentEncoding.CodePage == 65001) // UTF-8
        {
            byte byteCountMask = 0;
            while ((byteBuffer[byteLen - numFragments - 1] >> 6) == 2) // if the byte is "10xx xxxx", it's a continuation-byte
                byteCountMask |= (byte)(1 << ++numFragments); // count bytes & build the "complete char" mask
            if ((byteBuffer[byteLen - numFragments - 1] >> 6) == 3) // if the byte is "11xx xxxx", it starts a multi-byte char.
                byteCountMask |= (byte)(1 << ++numFragments); // count bytes & build the "complete char" mask
            // see if we found as many bytes as the leading-byte says to expect
            if (numFragments > 1 && ((byteBuffer[byteLen - numFragments] >> 7 - numFragments) == byteCountMask))
                numFragments = 0; // no partial-char in the byte-buffer to account for
        }
        else if (reader.CurrentEncoding.CodePage == 1200) // UTF-16LE
        {
            if (byteBuffer[byteLen - 1] >= 0xd8) // high-surrogate
                numFragments = 2; // account for the partial character
        }
        else if (reader.CurrentEncoding.CodePage == 1201) // UTF-16BE
        {
            if (byteBuffer[byteLen - 2] >= 0xd8) // high-surrogate
                numFragments = 2; // account for the partial character
        }
    }
    return reader.BaseStream.Position - numBytesLeft - numFragments;
}

Of course, this uses Reflection to get at private variables, so there is risk involved. However, this method works with .Net 2.0, 3.0, 3.5, 4.0, 4.0.3, 4.5, 4.5.1, 4.5.2, 4.6, and 4.6.1. Beyond that risk, the only other critical assumption is that the underlying byte-buffer is a byte[1024]; if Microsoft changes it the wrong way, the method breaks for UTF-16/-32.

This has been tested against a UTF-8 file filled with Ažテ𣘺 (10 bytes: 0x41 C5 BE E3 83 86 F0 A3 98 BA) and a UTF-16 file filled with A𐐷 (6 bytes: 0x41 00 01 D8 37 DC). The point being to force-fragment characters along the byte[1024] boundaries, all the different ways they could be.

UPDATE (2013-07-03): I fixed the method, which originally used the broken code from that other answer. This version has been tested against data containing a characters requiring use of surrogate pairs. The data was put into 3 files, each with a different encoding; one UTF-8, one UTF-16LE, and one UTF-16BE.

UPDATE (2016-02): The only correct way to handle bisected characters is to directly interpret the underlying bytes. UTF-8 is properly handled, and UTF-16/-32 work (given the length of byteBuffer).