unac

NAME
SYNOPSIS
DESCRIPTION
FUNCTIONS
ERRORS
EXAMPLES
IMPLEMENTATION NOTES
BUGS
SEE ALSO
AUTHOR

---

NAME

unac - remove accents from string or character

SYNOPSIS

#include <unac.h> const char* unac_version(); int unac_string(const char* charset, const char* in, int in_length, char** out, int* out_length); int unac_string_utf16(const char* in, int in_length, char** out, int* out_length); /* MACRO: side effect on unaccented and length */ unac_char_utf16(unsigned short c, unsigned short* unaccented, int length); const char* unac_version() /* * level: UNAC_DEBUG_NONE UNAC_DEBUG_LOW UNAC_DEBUG_HIGH */ void unac_debug(int level) /* typedef void (*unac_debug_print_t)(const char* message, void* data); */ void unac_debug_callback(int level, unac_debug_print_t function, void* data)

DESCRIPTION

unac is a C library that removes accents from characters, regardless of the character set (ISO-8859-15, ISO-CELTIC, KOI8-RU...) as long as iconv(3) is able to convert it into UTF-16 (Unicode).

The unac_string function is given a charset (ISO-8859-15 for instance) and a string. It converts the string into UTF-16 and calls the unac_string_utf16 function to remove all accents from the UTF-16 version. The unaccented string is then converted into the original charset (ISO-8859-15 for instance) and sent back to the caller of unac_string.

unac does a little more than removing accents: every character that is made of two character such as æ (ISO-8859-15 octal code 346) will be expanded in two characters a and e. Should a character be made of three characters, it would be decomposed in the same way.

The conversion from and to UTF-16 is done with iconv(3). The iconv -l will list all available charsets. Using UTF-16 as a pivot implies an overhead but ensures that accents can be removed from all character for which there is an equivalent character in Unicode.

unac_char_utf16 is a macro that returns a pointer to the unaccented equivalent of any UTF-16 character. It is the basic building block of unac.

unac_string_utf16 repeatidly applies the unac_char_utf16 function on each character of an UTF-16 string.

FUNCTIONS

int unac_string(const char* charset, const char* in, int in_length, char** out, int* out_length)

Returns the unaccented equivalent of the string 'in' of length 'in_length'. The returned string is stored in the pointer pointed by the 'out' argument and the length of the string is stored in the integer pointed by the 'out_length ' argument. If the '*out' pointer is not null, it must point to an area allocated by malloc(3) and the length of the array must be specified in the '*out_length' argument. Both arguments '*out' and '*out_length' will be replaced with the return values when the function returns on success. The '*out' pointer may point to a memory location that has been reallocated by the unac_string function. There is no guarantee that '*out' is identical to the value given by the caller and the malloc'ed memory location given by the caller may not be useable when the function returns (either error or success). If the '*out' pointer is null, the unac_string function allocates a new memory block using malloc(3). It is the responsibility of the caller to deallocate the area returned in the '*out' pointer.

The return value of unac_string is 0 on success and -1 on error, in which case the errno variable is set to the corresponding error code. See ERROR section below for more information, the iconv(3) manual page may also help.

int unac_string_utf16(const char* in, int in_length, char** out, int* out_length)

Has the same effect as unac_string("UTF-16", in, in_length, out, out_length). Since the unac_string_utf16 is the backend function of the unac_string function it is more efficient because no charset conversion of the input string (from and to UTF-16) is necessary.

unac_char_utf16(const unsigned short c, unsigned short* p, int l)

Warning: this is a macro, each argument may be evaluated more than once. Returns the unaccented equivalent of the UTF-16 character 'c' in the pointer 'p'. The length of the unsigned short array pointed by 'p' is returned in the 'l' argument.

const char* unac_version()

Return the version number of unac.

void unac_debug(int level)

Set the debug level of the unac library to 'level'. Possible values are: UNAC_DEBUG_NONE for no debug at all, UNAC_DEBUG_LOW for terse human readable information, UNAC_DEBUG_HIGH for very detailed information only usable when translating a few strings.

unac_debug_callback with anything but UNAC_DEBUG_NONE is not thread safe.

void unac_debug_callback(int level, unac_debug_print_t function, void* data)

Set the debug level and define a printing function callback. The 'level' is the same as in unac_debug. The 'function' is in charge of dealing with the debug messages, presumably to print them to the user. The 'data' is an opaque pointer that is passed along to function, should it need to manage a persistent context.

The prototype of 'function' accepts two arguments. The first is the debug message (const char*), the second is the opaque pointer given as 'data' argument to unac_debug_callback.

If 'function' is NULL, messages are printed on the standard error output using fprintf(stderr...).

unac_debug_callback with anything but UNAC_DEBUG_NONE is not thread safe.

ERRORS

EINVAL

the requested conversion pair is not available. For instance, when specifying the ISO-0000 charset (imaginary), it means it is not possible to convert from ISO-0000 to UTF-16.

EXAMPLES

Convert the été string into ete.

#include <unac.h> char* out = 0; int out_length = 0; if(unac_string("ISO-8859-1", "été", strlen("été"), &out, &out_length)) { perror("unac_string"); } else { printf("%.*s0, out_length, out); free(out); }

IMPLEMENTATION NOTES

The endianess of the UTF-16 strings manipulated by unac must always be big endian. When using iconv(3) to translate strings, UTF-16BE should be used instead of UTF-16 to make sure it is big endian (BE). On some systems where UTF-16BE is not available, unac will rely on UTF-16 and hope it is properly big endian encoded. For more information check RFC2781 (http://www.faqs.org/rfcs/rfc2781.html: UTF-16, an encoding of ISO 10646).

The unac library uses the Unicode database to map accented letters to their unaccented equivalent. Mapping tables are generated from the UnicodeData-3.2.0.txt file (as found at http://www.unicode.org/Public/3.2-Update/) by the builder perl script. The builder script inserts these tables in the unac.h and unac.c files, replacing the existing ones. Looking for the 'Generated by builder' string in the unac.[ch] files allows to spot the various parts handled by the builder script.

The library data occupies 25KB where a simple minded table would occupy around 512Kbytes. The idea used to compress the tables is that many Unicode characters do not have unaccented equivalent. Instead of relying on a table mapping each Unicode character to the corresponding unaccented character, an intermediate array of pointers is created. In the drawing below, the range of UTF-16 character is not accurate but illustrates the method. The unac_data_table points to a set of unac_dataXX arrays. Each pointer covers a range of UTF-16 characters (4 in the example below). When a range of character does not contain any accented character, unac_data_table always points to the same array : unac_data0. Since there are many characters without accents, this is enough to achieve a good compression.

unac_data15 unac_data16 [ NULL, NULL, NULL, e ] <---- /------> [ a, NULL, NULL, NULL ] | | | | ^ ^ |-----| |-----| |-----| |-----| |-----| |-----| [ ... a b c d e f g h i j k é à 0 1 2 3 4 5 6 7 8 9 A... ] unac_data_table |-----| |-----| |-----| |-----| |-----| |-----| v v v v | | | | | | | | --------------------------------------/ | V [ NULL, NULL, NULL, NULL ] unac_data0 Beside this simple optimization, a table (unac_positions) listing the actual position of the unaccented replacement within a block (unac_dataXX) is necessary because they are not of fixed length. Some characters such as æ will be replaced by two characters a and e therefore unac_dataXX has a variable size.

The unaccented equivalent of an UTF-16 character is calculated by applying a compatibility decomposition and then stripping all characters that belong to the mark category. For a precise definition see the Unicode-3.2 normalization forms at http://www.unicode.org/unicode/reports/tr28/.

All original Unicode data files were taken from http://www.unicode.org/Public and prepended with the UCD Terms of Use.

BUGS

The input string must not contain partially formed characters, there is no support for this case.

UTF-16 surrogates are not handled.

Unicode may contain bugs in the decomposition of characters. When you suspect such a bug on a given string, add a test case with the faulty string in the t_unac.in test script in the source distribution and run make check. It will describe, in a very verbose way, how the string was unaccented. You may then fix the UnicodeData-3.2.0.txt file and run make check again to make sure the problem is solved. Please send such fixes to the author and to the Unicode consortium.

SEE ALSO

unaccent(1), iconv(3)

http://www.unicode.org/ http://oss.software.ibm.com/icu/ http://www.gnu.org/manual/glibc-2.2.5/libc.html

AUTHOR

Loic Dachary loic@senga.org

http://www.senga.org/unac/

---