Byte Order Mark

Unicode character stream may contain a byte order mark (BOM) at the beginning. Converters like IConv or ICU may possibly not eat the BOM. Also, if an engine runs in a genuine codec mode, i.e. without conversion, it does not know about BOMs. It is the user’s task to cut the BOM–but Quex helps.

Quex provides functions to handle the recognition and cutting of the BOM from a character stream. To access the following files must be included:

#include <quex/code_base/bom>

Note, that the corresponding implementation ‘bom.i’ file is only to be included manually, if no generated lexer is used. Otherwise, its implementation is part of the quex-core which is linked anyway. These files declare and implement BOM related functions in in the quex namespace. The list of available functions in C++ is

namespace quex {
    QUEX_TYPE_BOM      bom_snap(FILE* fh);
    template <class InStream>
    QUEX_TYPE_BOM      QUEX_NAME(bom_snap)(InStream* is_p);
    QUEX_TYPE_BOM      bom_identify(const uint8_t* const Buffer, size_t* n);
    const char*        bom_name(QUEX_TYPE_BOM BOM);

and in C

QUEX_TYPE_BOM  quex_bom_snap(FILE* fh);
QUEX_TYPE_BOM  quex_bom_identify(const uint8_t* const Buffer, size_t* n);
const char*    quex_bom_name(QUEX_TYPE_BOM BOM);

In the further discussion the functions will be referred to without their name or namespace prefix. The function bom_snap(...) checks whether the stream starts with a byte sequence that is identified as a byte order mark. If so, it steps over it so that the lexical analysis can start immediately after it. An exception is the BOM for UTF7; this coding is identified, but the stream does not step over it. If UTF7 is detected, it has to be considered with care because it may actually a normal character sequence. The type QUEX_TYPE_BOM defines constants that identify BOMs. It is defined as follows:


The BOM for UTF7 consists of the three bytes 0x2B, 0x2F, 0x76 plus one of 0x2B, 0x2F, 0x38, or 0x39. This corresponds to the characters “+”, “/”, “v” and one of “+”, “/”, “8”, or “9”. All of them are normal Unicode Characters. Thus a normal Unicode stream could wrongly be identified as an UTF7 stream. Also, The last two bits are the beginning of a new character. Thus, the BOM cannot easily be snapped from the stream. Instead, the whole byte stream would have to be bit shifted.

For the aforementioned reasons, a UTF7 BOM is not cut from the byte stream.

Basic characteristics can be identified by binary ‘and’ operations. For example

using namespace quex;

bom_type = bom_snap(file_handle);

if( bom_type & (QUEX_BOM_UTF32 | QUEX_BOM_NONE) ) {

Checks whether a BOM of codec UTF32 was found, or if there was no BOM. The statement also holds if a UTF7 BOM is found, since the QUEX_BOM_UTF7 has the QUEX_BOM_NONE bit raised. The exact information about the byte order can be detected by considering the whole value, e.g.

switch( bom_type ) {
/* Little ending BOM  => use the little endian converter. */
case QUEX_BOM_UTF_32_LE: codec_name = "UTF32LE"; break;
/* No BOM, or big endian bom => use big endian converter. */
case QUEX_BOM_UTF_32_BE: codec_name = "UTF32BE"; break;
/* Unkown BOM => break                                    */
     error_msg("Unknown BOM detected %s", bom_name(bom_type));

The example above, already, mentions another helper function that maps a BOM identifier to a human readable string

const char*     bom_name(QUEX_TYPE_BOM BOM);

If the user wishes to identify on some chunk of arbitrary memory the following function may be used

QUEX_TYPE_BOM   bom_identify(const uint8_t* const Buffer, size_t* n);

It receives a byte array in Buffer which must at least be of size four. It reports the found BOM as a return value and fills the number of bytes that the BOM occupies into what the second argument n points.

One important thing to notice is that the constructor does the first load from the data stream. Thus, if the BOM-cutting happens after the construction of the lexical analyzer object the ‘cut’ would not have any effect. Thus, the constructor call must be delayed after the call to BOM_snap(...). If the initial call to the constructor cannot be avoided, then the call to the BOM snap function must be followed by a call to the reset(...) function. Also, an attempt to cut the BOM, after the constructor has done its initial load must fail.


Do not use the file or stream handle that is used for BOM cutting in the lexical analyzer constructor before the BOM cutting. If this is desired, then the constructor call happen after the BOM cut.

An example of how to cut the BOM can be found in demo/*/003 in example-bom.c, respectively example-bom.c. The following code fragment shows an initialization in C language:

FILE*           fh = NULL;
EasyLexer       qlex;

fh = fopen(file_name, "rb");

/* Either there is no BOM, or if there is one, then it must be UTF8 */
QUEX_TYPE_BOM   bom_type = quex_bom_snap(fh);
if( (bom_type & (QUEX_BOM_UTF_8 | QUEX_BOM_NONE)) == 0 ) {
    printf("Found a non-UTF8 BOM. Exit\n");
    return 0;

/* The lexer **must** be constructed after the BOM-cut */
QUEX_NAME(from_FILE)(&qlex, fh, "UTF8", false);

/* Now, the qlex is ready for analysis. */

If a running lexer needs to set the bom dynamically, a pattern like the following may be followed:

 quex::my_lexer  qlex(...);
 switch( quex::bom_snap(fh) )
     case QUEX_BOM_UTF_8:      qlex.reset(fh, "utf-8");      break;
     case QUEX_BOM_UTF_1:      qlex.reset(fh, "iso-10646");  break;
     case QUEX_BOM_UTF_EBCDIC: qlex.reset(fh, "ebcdic-us");  break;
     case QUEX_BOM_GB_18030:   qlex.reset(fh, "gb18030");    break;
     case QUEX_BOM_UTF_7:      qlex.reset(fh, "utf-7");      break;
     case QUEX_BOM_UTF_16:     qlex.reset(fh, "utf-16");     break;
     case QUEX_BOM_UTF_16_LE:  qlex.reset(fh, "utf-16le");   break;
     case QUEX_BOM_UTF_16_BE:  qlex.reset(fh, "utf-16be");   break;
     case QUEX_BOM_UTF_32:     qlex.reset(fh, "utf-32");     break;
     case QUEX_BOM_UTF_32_LE:  qlex.reset(fh, "utf-32le");   break;
     case QUEX_BOM_UTF_32_BE:  qlex.reset(fh, "utf-32be");   break;
     default:                  qlex.reset(fh, get_file_encoding(fh));