'''
''' Copyright (C) 2000, 2001, 2002 Loic Dachary <loic@senga.org>
'''
''' This program is free software; you can redistribute it and/or modify
''' it under the terms of the GNU General Public License as published by
''' the Free Software Foundation; either version 2 of the License, or
''' (at your option) any later version.
'''
''' This program is distributed in the hope that it will be useful,
''' but WITHOUT ANY WARRANTY; without even the implied warranty of
''' MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
''' GNU General Public License for more details.
'''
''' You should have received a copy of the GNU General Public License
''' along with this program; if not, write to the Free Software
''' Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
'''
.TH unac 3 local
.SH NAME
unac \- remove accents from string or character

.SH SYNOPSIS
.nf
.ft CW
#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)

.ft R
.fi

.SH DESCRIPTION
.I 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 
.I iconv(3)
is able to convert it into UTF-16 (Unicode).

The
.B 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
.B 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 
.B unac_string.

.I unac
does a little more than removing accents:
every character that is made of two character such as 
.B æ 
(ISO-8859-15 octal code 346)
will be expanded in two characters
.B a
and
.B 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 
.I iconv(3).
The
.B 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.

.B 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 
.I unac.

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

.SH FUNCTIONS

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

Returns the unaccented equivalent of the string
.B 'in'
of length
.B 'in_length'.
The returned string is stored in the pointer pointed by the
.B 'out'
argument and the length of the string is stored in the integer pointed by
the
.B 'out_length '
argument. 
If the 
.B '*out'
pointer is not null, it must point to an area allocated by
.I malloc(3)
and the length of the array must be specified in the 
.B '*out_length'
argument. Both arguments 
.B '*out'
and
.B '*out_length'
will be replaced with the return values when the function returns
on success. The
.B '*out'
pointer may point to a memory location that has been reallocated 
by the
.B 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
.B '*out'
pointer is null, the 
.B unac_string 
function allocates a new memory block using 
.I malloc(3).
It is the responsibility of the caller to deallocate the area returned
in the
.B '*out'
pointer.

The return value of
.B 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
.I iconv(3)
manual page may also help.

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

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

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

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

.TP
.B const char* unac_version()

Return the version number of
.B unac.

.TP
.B void unac_debug(int level)
Set the debug level of the unac library to 
.B '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.

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

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

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

If
.B '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.
.B 
.SH ERRORS

.TP
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.

.SH EXAMPLES
Convert the 
.B été
string into
.B ete.
.nf
.ft CW
#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("%.*s\n", out_length, out);
   free(out);
}
.ft R
.fi

.SH IMPLEMENTATION NOTES

The endianess of the UTF-16 strings manipulated by 
.I unac
must always be big endian. When using 
.I 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,
.B 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 
.I unac
library uses the Unicode database to map accented letters to their unaccented
equivalent. Mapping tables are generated from the 
.I UnicodeData-3.2.0.txt
file (as found at http://www.unicode.org/Public/3.2-Update/) by the 
.I builder
perl script. The
.I builder
script inserts these tables in the 
.I unac.h
and
.I unac.c
files, replacing the existing ones. Looking for the
.B 'Generated by builder'
string in the
.I 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
.B unac_data_table
points to a set of
.B 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.
.nf
.ft CW


      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


.ft R
.fi
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 
.B æ
will
be replaced by two characters 
.B a
and
.B e
therefore unac_dataXX has a variable size.

The unaccented equivalent of an UTF-16 character is calculated by
applying a 
.I compatibility decomposition 
and then stripping all characters that belong to the 
.I 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 
.B UCD Terms of Use.

.SH 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
.B 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.

.SH SEE ALSO
unaccent(1), iconv(3)
.nf
.ft CW
http://www.unicode.org/
http://oss.software.ibm.com/icu/
http://www.gnu.org/manual/glibc-2.2.5/libc.html
.ft R
.fi

.SH AUTHOR
Loic Dachary loic@senga.org
.nf
.ft CW
http://www.senga.org/unac/
.ft R
.fi