dyaml/source/dyaml/tinyendian.d

//          Copyright Ferdinand Majerech 2014.
// Distributed under the Boost Software License, Version 1.0.
//    (See accompanying file LICENSE_1_0.txt or copy at
//          http://www.boost.org/LICENSE_1_0.txt)

module tinyendian;


import core.stdc.string;

import std.algorithm;
import std.system;
import std.utf;

// TODO tinyendian should be its own DUB package

/// Unicode UTF encodings.
enum UTFEncoding : ubyte
{
    UTF_8,
    UTF_16,
    UTF_32
}

@system pure nothrow @nogc:

/// Swap byte order of items in an array in place.
///
/// Params:
///
/// T     = Item type. Must be either 2 or 4 bytes long.
/// array = Buffer with values to fix byte order of.
void swapByteOrder(T)(T[] array)
    if([2, 4].canFind(T.sizeof))
{
    import core.bitop;
    // Swap the byte order of all read characters.
    foreach(ref item; array)
    {
        static if(T.sizeof == 2)
        {
            swap(*cast(ubyte*)&item, *(cast(ubyte*)&item + 1));
        }
        else static if(T.sizeof == 4)
        {
            item = bswap(cast(uint)item);
        }
        else static assert(false, "Unsupported T: " ~ T.stringof);
    }
}

/// Convert byte order of an array encoded in UTF(8/16/32) to system endianness in
/// place.
///
/// Uses the UTF byte-order-mark (BOM) to determine UTF encoding. If there is no BOM
/// at the beginning of array, UTF-8 is assumed (this is compatible with ASCII). The
/// BOM, if any, will be removed from the buffer.
///
/// If the encoding is determined to be UTF-16 or UTF-32 and there aren't enough bytes
/// for the last code unit (i.e. if array.length is odd for UTF-16 or not divisible by
/// 4 for UTF-32), the extra bytes (1 for UTF-16, 1-3 for UTF-32) are stripped.
///
/// Note that this function does $(B not) check if the array is a valid UTF string. It
/// only works with the BOM and 1,2 or 4-byte items.
///
/// Params:
///
/// array = The array with UTF-data.
///
/// Returns:
///
/// A struct with the following members:
///
/// $(D ubyte[] array)            A slice of the input array containing data in correct
///                               byte order, without BOM and in case of UTF-16/UTF-32,
///                               without stripped bytes, if any.
/// $(D UTFEncoding encoding)     Encoding of the result (UTF-8, UTF-16 or UTF-32)
/// $(D std.system.Endian endian) Endianness of the original array.
/// $(D uint bytesStripped)       Number of bytes stripped from a UTF-16/UTF-32 array,
///                               if any. This is non-zero only if array.length was not
///                               divisible by 2 or 4 for UTF-16 and UTF-32,
///                               respectively.
///
/// Complexity: (BIGOH array.length)
auto fixUTFByteOrder(ubyte[] array)
{
    // Enumerates UTF BOMs, matching indices to byteOrderMarks/bomEndian.
    enum BOM: ubyte
    {
        UTF_8     = 0,
        UTF_16_LE = 1,
        UTF_16_BE = 2,
        UTF_32_LE = 3,
        UTF_32_BE = 4,
        None      = ubyte.max
    }

    // From std.stream
    static immutable ubyte[][5] byteOrderMarks = [ [0xEF, 0xBB, 0xBF],
                                                   [0xFF, 0xFE],
                                                   [0xFE, 0xFF],
                                                   [0xFF, 0xFE, 0x00, 0x00],
                                                   [0x00, 0x00, 0xFE, 0xFF] ];

    // From std.stream
    static immutable Endian[5] bomEndian = [ std.system.endian,
                                             Endian.littleEndian,
                                             Endian.bigEndian,
                                             Endian.littleEndian, 
                                             Endian.bigEndian ];

    // Documented in function ddoc.
    struct Result
    {
        ubyte[] array;
        UTFEncoding encoding;
        Endian endian;
        uint bytesStripped = 0;
    }
    Result result;

    // Detect BOM, if any, in the bytes we've read. -1 means no BOM.
    // Need the last match: First 2 bytes of UTF-32LE BOM match the UTF-16LE BOM. If we
    // used the first match, UTF-16LE would be detected when we have a UTF-32LE BOM.
    BOM bomId = BOM.None;
    foreach(i, bom; byteOrderMarks) if(array.startsWith(bom))
    {
        bomId = cast(BOM)i;
    }

    result.endian = (bomId != BOM.None) ? bomEndian[bomId] : Endian.init;

    // Start of UTF data (after BOM, if any)
    size_t start = 0;
    // If we've read more than just the BOM, put the rest into the array.
    with(BOM) final switch(bomId)
    {
        case None: result.encoding = UTFEncoding.UTF_8; break;
        case UTF_8:
            start = 3;
            result.encoding = UTFEncoding.UTF_8;
            break;
        case UTF_16_LE, UTF_16_BE:
            result.bytesStripped = array.length % 2;
            start = 2;
            result.encoding = UTFEncoding.UTF_16;
            break;
        case UTF_32_LE, UTF_32_BE:
            result.bytesStripped = array.length % 4;
            start = 4;
            result.encoding = UTFEncoding.UTF_32;
            break;
    }

    array = array[0 .. $ - result.bytesStripped];
    // If there's a BOM, we need to move data back to ensure it starts at array[0]
    if(start != 0)
    {
        core.stdc.string.memmove(array.ptr, array.ptr + start, array.length - start);
        array = array[0 .. $ - start];
    }

    // We enforce above that array.length is divisible by 2/4 for UTF-16/32
    if(std.system.endian != result.endian)
    {
        if(result.encoding == UTFEncoding.UTF_16)      { swapByteOrder(cast(wchar[])array); }
        else if(result.encoding == UTFEncoding.UTF_32) { swapByteOrder(cast(dchar[])array); }
    }

    result.array = array;
    return result;
}