From 13f2e5c125ad769b0275d7e7f4a98f392cbe10e3 Mon Sep 17 00:00:00 2001 From: Ferdinand Majerech Date: Mon, 21 Jul 2014 23:40:44 +0200 Subject: [PATCH] Added a self-contained module to handle endianness fixing. --- source/dyaml/tinyendian.d | 172 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 172 insertions(+) create mode 100644 source/dyaml/tinyendian.d diff --git a/source/dyaml/tinyendian.d b/source/dyaml/tinyendian.d new file mode 100644 index 0000000..cce397c --- /dev/null +++ b/source/dyaml/tinyendian.d @@ -0,0 +1,172 @@ +// 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; +}