/*
******************************************************************************
*
* Copyright (C) 2000-2009, International Business Machines
* Corporation and others. All Rights Reserved.
*
******************************************************************************
* file name: ushape.c
* encoding: US-ASCII
* tab size: 8 (not used)
* indentation:4
*
* created on: 2000jun29
* created by: Markus W. Scherer
*
* Arabic letter shaping implemented by Ayman Roshdy
*/
#include "unicode/utypes.h"
#include "unicode/uchar.h"
#include "unicode/ustring.h"
#include "unicode/ushape.h"
#include "cmemory.h"
#include "putilimp.h"
#include "ustr_imp.h"
#include "ubidi_props.h"
#if UTF_SIZE<16
/*
* This implementation assumes that the internal encoding is UTF-16
* or UTF-32, not UTF-8.
* The main assumption is that the Arabic characters and their
* presentation forms each fit into a single UChar.
* With UTF-8, they occupy 2 or 3 bytes, and more than the ASCII
* characters.
*/
# error This implementation assumes UTF-16 or UTF-32 (check UTF_SIZE)
#endif
/*
* ### TODO in general for letter shaping:
* - the letter shaping code is UTF-16-unaware; needs update
* + especially invertBuffer()?!
* - needs to handle the "Arabic Tail" that is used in some legacy codepages
* as a glyph fragment of wide-glyph letters
* + IBM Unicode conversion tables map it to U+200B (ZWSP)
* + IBM Egypt has proposed to encode the tail in Unicode among Arabic Presentation Forms
*/
/* definitions for Arabic letter shaping ------------------------------------ */
#define IRRELEVANT 4
#define LAMTYPE 16
#define ALEFTYPE 32
#define LINKR 1
#define LINKL 2
#define APRESENT 8
#define SHADDA 64
#define CSHADDA 128
#define COMBINE (SHADDA+CSHADDA)
#define HAMZAFE_CHAR 0xfe80
#define HAMZA06_CHAR 0x0621
#define YEH_HAMZA_CHAR 0x0626
#define YEH_HAMZAFE_CHAR 0xFE89
#define LAMALEF_SPACE_SUB 0xFFFF
#define TASHKEEL_SPACE_SUB 0xFFFE
#define NEW_TAIL_CHAR 0xFE73
#define OLD_TAIL_CHAR 0x200B
#define LAM_CHAR 0x0644
#define SPACE_CHAR 0x0020
#define SHADDA_CHAR 0xFE7C
#define TATWEEL_CHAR 0x0640
#define SHADDA_TATWEEL_CHAR 0xFE7D
#define SHAPE_MODE 0
#define DESHAPE_MODE 1
UChar tailChar = OLD_TAIL_CHAR;
uint32_t uShapeLamalefBegin = U_SHAPE_LAMALEF_BEGIN;
uint32_t uShapeLamalefEnd = U_SHAPE_LAMALEF_END;
uint32_t uShapeTashkeelBegin = U_SHAPE_TASHKEEL_BEGIN;
uint32_t uShapeTashkeelEnd = U_SHAPE_TASHKEEL_END;
int spacesRelativeToTextBeginEnd = 0;
static const uint8_t tailFamilyIsolatedFinal[] = {
/* FEB1 */ 1,
/* FEB2 */ 1,
/* FEB3 */ 0,
/* FEB4 */ 0,
/* FEB5 */ 1,
/* FEB6 */ 1,
/* FEB7 */ 0,
/* FEB8 */ 0,
/* FEB9 */ 1,
/* FEBA */ 1,
/* FEBB */ 0,
/* FEBC */ 0,
/* FEBD */ 1,
/* FEBE */ 1
};
static const uint8_t tashkeelMedial[] = {
/* FE70 */ 0,
/* FE71 */ 1,
/* FE72 */ 0,
/* FE73 */ 0,
/* FE74 */ 0,
/* FE75 */ 0,
/* FE76 */ 0,
/* FE77 */ 1,
/* FE78 */ 0,
/* FE79 */ 1,
/* FE7A */ 0,
/* FE7B */ 1,
/* FE7C */ 0,
/* FE7D */ 1,
/* FE7E */ 0,
/* FE7F */ 1
};
static const UChar yehHamzaToYeh[] =
{
/* isolated*/ 0xFEEF,
/* final */ 0xFEF0
};
static const uint8_t IrrelevantPos[] = {
0x0, 0x2, 0x4, 0x6,
0x8, 0xA, 0xC, 0xE
};
static const UChar convertLamAlef[] =
{
/*FEF5*/ 0x0622,
/*FEF6*/ 0x0622,
/*FEF7*/ 0x0623,
/*FEF8*/ 0x0623,
/*FEF9*/ 0x0625,
/*FEFA*/ 0x0625,
/*FEFB*/ 0x0627,
/*FEFC*/ 0x0627
};
static const UChar araLink[178]=
{
1 + 32 + 256 * 0x11,/*0x0622*/
1 + 32 + 256 * 0x13,/*0x0623*/
1 + 256 * 0x15,/*0x0624*/
1 + 32 + 256 * 0x17,/*0x0625*/
1 + 2 + 256 * 0x19,/*0x0626*/
1 + 32 + 256 * 0x1D,/*0x0627*/
1 + 2 + 256 * 0x1F,/*0x0628*/
1 + 256 * 0x23,/*0x0629*/
1 + 2 + 256 * 0x25,/*0x062A*/
1 + 2 + 256 * 0x29,/*0x062B*/
1 + 2 + 256 * 0x2D,/*0x062C*/
1 + 2 + 256 * 0x31,/*0x062D*/
1 + 2 + 256 * 0x35,/*0x062E*/
1 + 256 * 0x39,/*0x062F*/
1 + 256 * 0x3B,/*0x0630*/
1 + 256 * 0x3D,/*0x0631*/
1 + 256 * 0x3F,/*0x0632*/
1 + 2 + 256 * 0x41,/*0x0633*/
1 + 2 + 256 * 0x45,/*0x0634*/
1 + 2 + 256 * 0x49,/*0x0635*/
1 + 2 + 256 * 0x4D,/*0x0636*/
1 + 2 + 256 * 0x51,/*0x0637*/
1 + 2 + 256 * 0x55,/*0x0638*/
1 + 2 + 256 * 0x59,/*0x0639*/
1 + 2 + 256 * 0x5D,/*0x063A*/
0, 0, 0, 0, 0, /*0x063B-0x063F*/
1 + 2, /*0x0640*/
1 + 2 + 256 * 0x61,/*0x0641*/
1 + 2 + 256 * 0x65,/*0x0642*/
1 + 2 + 256 * 0x69,/*0x0643*/
1 + 2 + 16 + 256 * 0x6D,/*0x0644*/
1 + 2 + 256 * 0x71,/*0x0645*/
1 + 2 + 256 * 0x75,/*0x0646*/
1 + 2 + 256 * 0x79,/*0x0647*/
1 + 256 * 0x7D,/*0x0648*/
1 + 256 * 0x7F,/*0x0649*/
1 + 2 + 256 * 0x81,/*0x064A*/
4 + 256 * 1, /*0x064B*/
4 + 128 + 256 * 1, /*0x064C*/
4 + 128 + 256 * 1, /*0x064D*/
4 + 128 + 256 * 1, /*0x064E*/
4 + 128 + 256 * 1, /*0x064F*/
4 + 128 + 256 * 1, /*0x0650*/
4 + 64 + 256 * 3, /*0x0651*/
4 + 256 * 1, /*0x0652*/
4 + 256 * 7, /*0x0653*/
4 + 256 * 8, /*0x0654*/
4 + 256 * 8, /*0x0655*/
4 + 256 * 1, /*0x0656*/
0, 0, 0, 0, 0, /*0x0657-0x065B*/
1 + 256 * 0x85,/*0x065C*/
1 + 256 * 0x87,/*0x065D*/
1 + 256 * 0x89,/*0x065E*/
1 + 256 * 0x8B,/*0x065F*/
0, 0, 0, 0, 0, /*0x0660-0x0664*/
0, 0, 0, 0, 0, /*0x0665-0x0669*/
0, 0, 0, 0, 0, 0, /*0x066A-0x066F*/
4 + 256 * 6, /*0x0670*/
1 + 8 + 256 * 0x00,/*0x0671*/
1 + 32, /*0x0672*/
1 + 32, /*0x0673*/
0, /*0x0674*/
1 + 32, /*0x0675*/
1, 1, /*0x0676-0x0677*/
1+2, 1+2, 1+2, 1+2, 1+2, 1+2, /*0x0678-0x067D*/
1+2+8+256 * 0x06, 1+2, 1+2, 1+2, 1+2, 1+2, /*0x067E-0x0683*/
1+2, 1+2, 1+2+8+256 * 0x2A, 1+2, /*0x0684-0x0687*/
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /*0x0688-0x0691*/
1, 1, 1, 1, 1, 1, 1+8+256 * 0x3A, 1, /*0x0692-0x0699*/
1+2, 1+2, 1+2, 1+2, 1+2, 1+2, /*0x069A-0x06A3*/
1+2, 1+2, 1+2, 1+2, /*0x069A-0x06A3*/
1+2, 1+2, 1+2, 1+2, 1+2, 1+2+8+256 * 0x3E, /*0x06A4-0x06AD*/
1+2, 1+2, 1+2, 1+2, /*0x06A4-0x06AD*/
1+2, 1+2+8+256 * 0x42, 1+2, 1+2, 1+2, 1+2, /*0x06AE-0x06B7*/
1+2, 1+2, 1+2, 1+2, /*0x06AE-0x06B7*/
1+2, 1+2, 1+2, 1+2, 1+2, 1+2, /*0x06B8-0x06BF*/
1+2, 1+2, /*0x06B8-0x06BF*/
1, /*0x06C0*/
1+2, /*0x06C1*/
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /*0x06C2-0x06CB*/
1+2+8+256 * 0xAC, /*0x06CC*/
1, /*0x06CD*/
1+2, 1+2, 1+2, 1+2, /*0x06CE-0x06D1*/
1, 1 /*0x06D2-0x06D3*/
};
static const uint8_t presALink[] = {
/***********0*****1*****2*****3*****4*****5*****6*****7*****8*****9*****A*****B*****C*****D*****E*****F*/
/*FB5*/ 0, 1, 0, 0, 0, 0, 0, 1, 2,1 + 2, 0, 0, 0, 0, 0, 0,
/*FB6*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FB7*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2,1 + 2, 0, 0,
/*FB8*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1,
/*FB9*/ 2,1 + 2, 0, 1, 2,1 + 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBA*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBB*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBC*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBD*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBE*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBF*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2,1 + 2,
/*FC0*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FC1*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FC2*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FC3*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FC4*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FC5*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 4, 4,
/*FC6*/ 4, 4, 4
};
static const uint8_t presBLink[]=
{
/***********0*****1*****2*****3*****4*****5*****6*****7*****8*****9*****A*****B*****C*****D*****E*****F*/
/*FE7*/1 + 2,1 + 2,1 + 2, 0,1 + 2, 0,1 + 2,1 + 2,1 + 2,1 + 2,1 + 2,1 + 2,1 + 2,1 + 2,1 + 2,1 + 2,
/*FE8*/ 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 2,1 + 2, 0, 1, 0,
/*FE9*/ 1, 2,1 + 2, 0, 1, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,
/*FEA*/1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 0, 1, 0, 1, 0,
/*FEB*/ 1, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,
/*FEC*/1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,
/*FED*/1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,
/*FEE*/1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 2,1 + 2, 0, 1, 0,
/*FEF*/ 1, 0, 1, 2,1 + 2, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0
};
static const UChar convertFBto06[] =
{
/***********0******1******2******3******4******5******6******7******8******9******A******B******C******D******E******F***/
/*FB5*/ 0x671, 0x671, 0, 0, 0, 0, 0x07E, 0x07E, 0x07E, 0x07E, 0, 0, 0, 0, 0, 0,
/*FB6*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FB7*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x686, 0x686, 0x686, 0x686, 0, 0,
/*FB8*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x698, 0x698, 0, 0, 0x6A9, 0x6A9,
/*FB9*/ 0x6A9, 0x6A9, 0x6AF, 0x6AF, 0x6AF, 0x6AF, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBA*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBB*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBC*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBD*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBE*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
/*FBF*/ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x6CC, 0x6CC, 0x6CC, 0x6CC
};
static const UChar convertFEto06[] =
{
/***********0******1******2******3******4******5******6******7******8******9******A******B******C******D******E******F***/
/*FE7*/ 0x64B, 0x64B, 0x64C, 0x64C, 0x64D, 0x64D, 0x64E, 0x64E, 0x64F, 0x64F, 0x650, 0x650, 0x651, 0x651, 0x652, 0x652,
/*FE8*/ 0x621, 0x622, 0x622, 0x623, 0x623, 0x624, 0x624, 0x625, 0x625, 0x626, 0x626, 0x626, 0x626, 0x627, 0x627, 0x628,
/*FE9*/ 0x628, 0x628, 0x628, 0x629, 0x629, 0x62A, 0x62A, 0x62A, 0x62A, 0x62B, 0x62B, 0x62B, 0x62B, 0x62C, 0x62C, 0x62C,
/*FEA*/ 0x62C, 0x62D, 0x62D, 0x62D, 0x62D, 0x62E, 0x62E, 0x62E, 0x62E, 0x62F, 0x62F, 0x630, 0x630, 0x631, 0x631, 0x632,
/*FEB*/ 0x632, 0x633, 0x633, 0x633, 0x633, 0x634, 0x634, 0x634, 0x634, 0x635, 0x635, 0x635, 0x635, 0x636, 0x636, 0x636,
/*FEC*/ 0x636, 0x637, 0x637, 0x637, 0x637, 0x638, 0x638, 0x638, 0x638, 0x639, 0x639, 0x639, 0x639, 0x63A, 0x63A, 0x63A,
/*FED*/ 0x63A, 0x641, 0x641, 0x641, 0x641, 0x642, 0x642, 0x642, 0x642, 0x643, 0x643, 0x643, 0x643, 0x644, 0x644, 0x644,
/*FEE*/ 0x644, 0x645, 0x645, 0x645, 0x645, 0x646, 0x646, 0x646, 0x646, 0x647, 0x647, 0x647, 0x647, 0x648, 0x648, 0x649,
/*FEF*/ 0x649, 0x64A, 0x64A, 0x64A, 0x64A, 0x65C, 0x65C, 0x65D, 0x65D, 0x65E, 0x65E, 0x65F, 0x65F
};
static const uint8_t shapeTable[4][4][4]=
{
{ {0,0,0,0}, {0,0,0,0}, {0,1,0,3}, {0,1,0,1} },
{ {0,0,2,2}, {0,0,1,2}, {0,1,1,2}, {0,1,1,3} },
{ {0,0,0,0}, {0,0,0,0}, {0,1,0,3}, {0,1,0,3} },
{ {0,0,1,2}, {0,0,1,2}, {0,1,1,2}, {0,1,1,3} }
};
/*
* This function shapes European digits to Arabic-Indic digits
* in-place, writing over the input characters.
* Since we know that we are only looking for BMP code points,
* we can safely just work with code units (again, at least UTF-16).
*/
static void
_shapeToArabicDigitsWithContext(UChar *s, int32_t length,
UChar digitBase,
UBool isLogical, UBool lastStrongWasAL) {
const UBiDiProps *bdp;
UErrorCode errorCode;
int32_t i;
UChar c;
errorCode=U_ZERO_ERROR;
bdp=ubidi_getSingleton(&errorCode);
if(U_FAILURE(errorCode)) {
return;
}
digitBase-=0x30;
/* the iteration direction depends on the type of input */
if(isLogical) {
for(i=0; i<length; ++i) {
c=s[i];
switch(ubidi_getClass(bdp, c)) {
case U_LEFT_TO_RIGHT: /* L */
case U_RIGHT_TO_LEFT: /* R */
lastStrongWasAL=FALSE;
break;
case U_RIGHT_TO_LEFT_ARABIC: /* AL */
lastStrongWasAL=TRUE;
break;
case U_EUROPEAN_NUMBER: /* EN */
if(lastStrongWasAL && (uint32_t)(c-0x30)<10) {
s[i]=(UChar)(digitBase+c); /* digitBase+(c-0x30) - digitBase was modified above */
}
break;
default :
break;
}
}
} else {
for(i=length; i>0; /* pre-decrement in the body */) {
c=s[--i];
switch(ubidi_getClass(bdp, c)) {
case U_LEFT_TO_RIGHT: /* L */
case U_RIGHT_TO_LEFT: /* R */
lastStrongWasAL=FALSE;
break;
case U_RIGHT_TO_LEFT_ARABIC: /* AL */
lastStrongWasAL=TRUE;
break;
case U_EUROPEAN_NUMBER: /* EN */
if(lastStrongWasAL && (uint32_t)(c-0x30)<10) {
s[i]=(UChar)(digitBase+c); /* digitBase+(c-0x30) - digitBase was modified above */
}
break;
default :
break;
}
}
}
}
/*
*Name : invertBuffer
*Function : This function inverts the buffer, it's used
* in case the user specifies the buffer to be
* U_SHAPE_TEXT_DIRECTION_LOGICAL
*/
static void
invertBuffer(UChar *buffer,int32_t size,uint32_t options,int32_t lowlimit,int32_t highlimit) {
UChar temp;
int32_t i=0,j=0;
for(i=lowlimit,j=size-highlimit-1;i<j;i++,j--) {
temp = buffer[i];
buffer[i] = buffer[j];
buffer[j] = temp;
}
}
/*
*Name : changeLamAlef
*Function : Converts the Alef characters into an equivalent
* LamAlef location in the 0x06xx Range, this is an
* intermediate stage in the operation of the program
* later it'll be converted into the 0xFExx LamAlefs
* in the shaping function.
*/
static U_INLINE UChar
changeLamAlef(UChar ch) {
switch(ch) {
case 0x0622 :
return 0x065C;
case 0x0623 :
return 0x065D;
case 0x0625 :
return 0x065E;
case 0x0627 :
return 0x065F;
}
return 0;
}
/*
*Name : getLink
*Function : Resolves the link between the characters as
* Arabic characters have four forms :
* Isolated, Initial, Middle and Final Form
*/
static UChar
getLink(UChar ch) {
if(ch >= 0x0622 && ch <= 0x06D3) {
return(araLink[ch-0x0622]);
} else if(ch == 0x200D) {
return(3);
} else if(ch >= 0x206D && ch <= 0x206F) {
return(4);
}else if(ch >= 0xFB50 && ch <= 0xFC62) {
return(presALink[ch-0xFB50]);
} else if(ch >= 0xFE70 && ch <= 0xFEFC) {
return(presBLink[ch-0xFE70]);
}else {
return(0);
}
}
/*
*Name : countSpaces
*Function : Counts the number of spaces
* at each end of the logical buffer
*/
static void
countSpaces(UChar *dest,int32_t size,uint32_t options,int32_t *spacesCountl,int32_t *spacesCountr) {
int32_t i = 0;
int32_t countl = 0,countr = 0;
while(dest[i] == SPACE_CHAR) {
countl++;
i++;
}
while(dest[size-1] == SPACE_CHAR) {
countr++;
size--;
}
*spacesCountl = countl;
*spacesCountr = countr;
}
/*
*Name : isTashkeelChar
*Function : Returns 1 for Tashkeel characters in 06 range else return 0
*/
static U_INLINE int32_t
isTashkeelChar(UChar ch) {
return (int32_t)( ch>=0x064B && ch<= 0x0652 );
}
/*
*Name : isTashkeelCharFE
*Function : Returns 1 for Tashkeel characters in FE range else return 0
*/
static U_INLINE int32_t
isTashkeelCharFE(UChar ch) {
return (int32_t)( ch>=0xFE70 && ch<= 0xFE7F );
}
/*
*Name : isAlefChar
*Function : Returns 1 for Alef characters else return 0
*/
static U_INLINE int32_t
isAlefChar(UChar ch) {
return (int32_t)( (ch==0x0622)||(ch==0x0623)||(ch==0x0625)||(ch==0x0627) );
}
/*
*Name : isLamAlefChar
*Function : Returns 1 for LamAlef characters else return 0
*/
static U_INLINE int32_t
isLamAlefChar(UChar ch) {
return (int32_t)((ch>=0xFEF5)&&(ch<=0xFEFC) );
}
/*BIDI
*Name : isTailChar
*Function : returns 1 if the character matches one of the tail characters (0xfe73 or 0x200b) otherwise returns 0
*/
static U_INLINE int32_t
isTailChar(UChar ch) {
if(ch == OLD_TAIL_CHAR || ch == NEW_TAIL_CHAR){
return 1;
}else{
return 0;
}
}
/*BIDI
*Name : isSeenTailFamilyChar
*Function : returns 1 if the character is a seen family isolated character
* in the FE range otherwise returns 0
*/
static U_INLINE int32_t
isSeenTailFamilyChar(UChar ch) {
if(ch >= 0xfeb1 && ch < 0xfebf){
return tailFamilyIsolatedFinal [ch - 0xFEB1];
}else{
return 0;
}
}
/* Name : isSeenFamilyChar
* Function : returns 1 if the character is a seen family character in the Unicode
* 06 range otherwise returns 0
*/
static U_INLINE int32_t
isSeenFamilyChar(UChar ch){
if(ch >= 0x633 && ch <= 0x636){
return 1;
}else {
return 0;
}
}
/*Start of BIDI*/
/*
*Name : isAlefMaksouraChar
*Function : returns 1 if the character is a Alef Maksoura Final or isolated
* otherwise returns 0
*/
static U_INLINE int32_t
isAlefMaksouraChar(UChar ch) {
return (int32_t)( (ch == 0xFEEF) || ( ch == 0xFEF0) || (ch == 0x0649));
}
/*
* Name : isYehHamzaChar
* Function : returns 1 if the character is a yehHamza isolated or yehhamza
* final is found otherwise returns 0
*/
static U_INLINE int32_t
isYehHamzaChar(UChar ch) {
if((ch==0xFE89)||(ch==0xFE8A)){
return 1;
}else{
return 0;
}
}
/*
* Name: isTashkeelOnTatweelChar
* Function: Checks if the Tashkeel Character is on Tatweel or not,if the
* Tashkeel on tatweel (FE range), it returns 1 else if the
* Tashkeel with shadda on tatweel (FC range)return 2 otherwise
* returns 0
*/
static U_INLINE int32_t
isTashkeelOnTatweelChar(UChar ch){
if(ch >= 0xfe70 && ch <= 0xfe7f && ch != NEW_TAIL_CHAR && ch != 0xFE75 && ch != SHADDA_TATWEEL_CHAR)
{
return tashkeelMedial [ch - 0xFE70];
}else if( (ch >= 0xfcf2 && ch <= 0xfcf4) || (ch == SHADDA_TATWEEL_CHAR)) {
return 2;
}else{
return 0;
}
}
/*
* Name: isIsolatedTashkeelChar
* Function: Checks if the Tashkeel Character is in the isolated form
* (i.e. Unicode FE range) returns 1 else if the Tashkeel
* with shadda is in the isolated form (i.e. Unicode FC range)
* returns 2 otherwise returns 0
*/
static U_INLINE int32_t
isIsolatedTashkeelChar(UChar ch){
if(ch >= 0xfe70 && ch <= 0xfe7f && ch != NEW_TAIL_CHAR && ch != 0xFE75){
return (1 - tashkeelMedial [ch - 0xFE70]);
}else if(ch >= 0xfc5e && ch <= 0xfc63){
return 1;
}else{
return 0;
}
}
/*
*Name : calculateSize
*Function : This function calculates the destSize to be used in preflighting
* when the destSize is equal to 0
* It is used also to calculate the new destsize in case the
* destination buffer will be resized.
*/
static int32_t
calculateSize(const UChar *source, int32_t sourceLength,
int32_t destSize,uint32_t options) {
int32_t i = 0;
int lamAlefOption = 0;
int tashkeelOption = 0;
destSize = sourceLength;
if (((options&U_SHAPE_LETTERS_MASK) == U_SHAPE_LETTERS_SHAPE ||
((options&U_SHAPE_LETTERS_MASK) == U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED )) &&
((options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_RESIZE )){
lamAlefOption = 1;
}
if((options&U_SHAPE_LETTERS_MASK) == U_SHAPE_LETTERS_SHAPE &&
((options&U_SHAPE_TASHKEEL_MASK) == U_SHAPE_TASHKEEL_RESIZE ) ){
tashkeelOption = 1;
}
if(lamAlefOption || tashkeelOption){
if((options&U_SHAPE_TEXT_DIRECTION_MASK)==U_SHAPE_TEXT_DIRECTION_VISUAL_LTR) {
for(i=0;i<sourceLength;i++) {
if( ((isAlefChar(source[i]))&& (i<(sourceLength-1)) &&(source[i+1] == LAM_CHAR)) || (isTashkeelCharFE(source[i])) ) {
destSize--;
}
}
}else if((options&U_SHAPE_TEXT_DIRECTION_MASK)==U_SHAPE_TEXT_DIRECTION_LOGICAL) {
for(i=0;i<sourceLength;i++) {
if( ( (source[i] == LAM_CHAR) && (i<(sourceLength-1)) && (isAlefChar(source[i+1]))) || (isTashkeelCharFE(source[i])) ) {
destSize--;
}
}
}
}
if ((options&U_SHAPE_LETTERS_MASK) == U_SHAPE_LETTERS_UNSHAPE){
if ( (options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_RESIZE){
for(i=0;i<sourceLength;i++) {
if(isLamAlefChar(source[i]))
destSize++;
}
}
}
return destSize;
}
/*
*Name : handleTashkeelWithTatweel
*Function : Replaces Tashkeel as following:
* Case 1 :if the Tashkeel on tatweel, replace it with Tatweel.
* Case 2 :if the Tashkeel aggregated with Shadda on Tatweel, replace
* it with Shadda on Tatweel.
* Case 3: if the Tashkeel is isolated replace it with Space.
*
*/
static int32_t
handleTashkeelWithTatweel(UChar *dest, int32_t sourceLength,
int32_t destSize,uint32_t options,
UErrorCode *pErrorCode) {
int i;
for(i = 0; i < sourceLength; i++){
if((isTashkeelOnTatweelChar(dest[i]) == 1)){
dest[i] = TATWEEL_CHAR;
}else if((isTashkeelOnTatweelChar(dest[i]) == 2)){
dest[i] = SHADDA_TATWEEL_CHAR;
}else if(isIsolatedTashkeelChar(dest[i]) && dest[i] != SHADDA_CHAR){
dest[i] = SPACE_CHAR;
}
}
return sourceLength;
}
/*
*Name : handleGeneratedSpaces
*Function : The shapeUnicode function converts Lam + Alef into LamAlef + space,
* and Tashkeel to space.
* handleGeneratedSpaces function puts these generated spaces
* according to the options the user specifies. LamAlef and Tashkeel
* spaces can be replaced at begin, at end, at near or decrease the
* buffer size.
*
* There is also Auto option for LamAlef and tashkeel, which will put
* the spaces at end of the buffer (or end of text if the user used
* the option U_SHAPE_SPACES_RELATIVE_TO_TEXT_BEGIN_END).
*
* If the text type was visual_LTR and the option
* U_SHAPE_SPACES_RELATIVE_TO_TEXT_BEGIN_END was selected the END
* option will place the space at the beginning of the buffer and
* BEGIN will place the space at the end of the buffer.
*/
static int32_t
handleGeneratedSpaces(UChar *dest, int32_t sourceLength,
int32_t destSize,
uint32_t options,
UErrorCode *pErrorCode ) {
int32_t i = 0, j = 0;
int32_t count = 0;
UChar *tempbuffer=NULL;
int lamAlefOption = 0;
int tashkeelOption = 0;
int shapingMode = SHAPE_MODE;
if (shapingMode == 0){
if ( (options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_RESIZE ){
lamAlefOption = 1;
}
if ( (options&U_SHAPE_TASHKEEL_MASK) == U_SHAPE_TASHKEEL_RESIZE ){
tashkeelOption = 1;
}
}
if (lamAlefOption || tashkeelOption){
tempbuffer = (UChar *)uprv_malloc((sourceLength+1)*U_SIZEOF_UCHAR);
/* Test for NULL */
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
uprv_memset(tempbuffer, 0, (sourceLength+1)*U_SIZEOF_UCHAR);
i = j = 0; count = 0;
while(i < sourceLength) {
if ( (lamAlefOption && dest[i] == LAMALEF_SPACE_SUB) ||
(tashkeelOption && dest[i] == TASHKEEL_SPACE_SUB) ){
j--;
count++;
} else {
tempbuffer[j] = dest[i];
}
i++;
j++;
}
while(count >= 0) {
tempbuffer[i] = 0x0000;
i--;
count--;
}
uprv_memcpy(dest, tempbuffer, sourceLength*U_SIZEOF_UCHAR);
destSize = u_strlen(dest);
}
lamAlefOption = 0;
if (shapingMode == 0){
if ( (options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_NEAR ){
lamAlefOption = 1;
}
}
if (lamAlefOption){
/* Lam+Alef is already shaped into LamAlef + FFFF */
i = 0;
while(i < sourceLength) {
if(lamAlefOption&&dest[i] == LAMALEF_SPACE_SUB){
dest[i] = SPACE_CHAR;
}
i++;
}
destSize = sourceLength;
}
lamAlefOption = 0;
tashkeelOption = 0;
if (shapingMode == 0) {
if ( ((options&U_SHAPE_LAMALEF_MASK) == uShapeLamalefBegin) ||
(((options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_AUTO )
&& (spacesRelativeToTextBeginEnd==1)) ) {
lamAlefOption = 1;
}
if ( (options&U_SHAPE_TASHKEEL_MASK) == uShapeTashkeelBegin ) {
tashkeelOption = 1;
}
}
if(lamAlefOption || tashkeelOption){
tempbuffer = (UChar *)uprv_malloc((sourceLength+1)*U_SIZEOF_UCHAR);
/* Test for NULL */
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
uprv_memset(tempbuffer, 0, (sourceLength+1)*U_SIZEOF_UCHAR);
i = j = sourceLength; count = 0;
while(i >= 0) {
if ( (lamAlefOption && dest[i] == LAMALEF_SPACE_SUB) ||
(tashkeelOption && dest[i] == TASHKEEL_SPACE_SUB) ){
j++;
count++;
}else {
tempbuffer[j] = dest[i];
}
i--;
j--;
}
for(i=0 ;i < count; i++){
tempbuffer[i] = SPACE_CHAR;
}
uprv_memcpy(dest, tempbuffer, sourceLength*U_SIZEOF_UCHAR);
destSize = sourceLength;
}
lamAlefOption = 0;
tashkeelOption = 0;
if (shapingMode == 0) {
if ( ((options&U_SHAPE_LAMALEF_MASK) == uShapeLamalefEnd) ||
(((options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_AUTO )
&& (spacesRelativeToTextBeginEnd==0)) ) {
lamAlefOption = 1;
}
if ( (options&U_SHAPE_TASHKEEL_MASK) == uShapeTashkeelEnd ){
tashkeelOption = 1;
}
}
if(lamAlefOption || tashkeelOption){
tempbuffer = (UChar *)uprv_malloc((sourceLength+1)*U_SIZEOF_UCHAR);
/* Test for NULL */
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
uprv_memset(tempbuffer, 0, (sourceLength+1)*U_SIZEOF_UCHAR);
i = j = 0; count = 0;
while(i < sourceLength) {
if ( (lamAlefOption && dest[i] == LAMALEF_SPACE_SUB) ||
(tashkeelOption && dest[i] == TASHKEEL_SPACE_SUB) ){
j--;
count++;
}else {
tempbuffer[j] = dest[i];
}
i++;
j++;
}
while(count >= 0) {
tempbuffer[i] = SPACE_CHAR;
i--;
count--;
}
uprv_memcpy(dest,tempbuffer, sourceLength*U_SIZEOF_UCHAR);
destSize = sourceLength;
}
if(tempbuffer){
uprv_free(tempbuffer);
}
return destSize;
}
/*
*Name :expandCompositCharAtBegin
*Function :Expands the LamAlef character to Lam and Alef consuming the required
* space from beginning of the buffer. If the text type was visual_LTR
* and the option U_SHAPE_SPACES_RELATIVE_TO_TEXT_BEGIN_END was selected
* the spaces will be located at end of buffer.
* If there are no spaces to expand the LamAlef, an error
* will be set to U_NO_SPACE_AVAILABLE as defined in utypes.h
*/
static int32_t
expandCompositCharAtBegin(UChar *dest, int32_t sourceLength, int32_t destSize,UErrorCode *pErrorCode) {
int32_t i = 0,j = 0;
int32_t countl = 0;
UChar *tempbuffer=NULL;
tempbuffer = (UChar *)uprv_malloc((sourceLength+1)*U_SIZEOF_UCHAR);
/* Test for NULL */
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
uprv_memset(tempbuffer, 0, (sourceLength+1)*U_SIZEOF_UCHAR);
i = 0;
while(dest[i] == SPACE_CHAR) {
countl++;
i++;
}
i = j = sourceLength-1;
while(i >= 0 && j >= 0) {
if( countl>0 && isLamAlefChar(dest[i])) {
tempbuffer[j] = LAM_CHAR;
tempbuffer[j-1] = convertLamAlef[ dest[i] - 0xFEF5 ];
j--;
countl--;
}else {
if( countl == 0 && isLamAlefChar(dest[i]) ) {
*pErrorCode=U_NO_SPACE_AVAILABLE;
}
tempbuffer[j] = dest[i];
}
i--;
j--;
}
uprv_memcpy(dest, tempbuffer, sourceLength*U_SIZEOF_UCHAR);
uprv_free(tempbuffer);
destSize = sourceLength;
return destSize;
}
/*
*Name : expandCompositCharAtEnd
*Function : Expands the LamAlef character to Lam and Alef consuming the
* required space from end of the buffer. If the text type was
* Visual LTR and the option U_SHAPE_SPACES_RELATIVE_TO_TEXT_BEGIN_END
* was used, the spaces will be consumed from begin of buffer. If
* there are no spaces to expand the LamAlef, an error
* will be set to U_NO_SPACE_AVAILABLE as defined in utypes.h
*/
static int32_t
expandCompositCharAtEnd(UChar *dest, int32_t sourceLength, int32_t destSize,UErrorCode *pErrorCode) {
int32_t i = 0,j = 0;
int32_t countr = 0;
int32_t inpsize = sourceLength;
UChar *tempbuffer=NULL;
tempbuffer = (UChar *)uprv_malloc((sourceLength+1)*U_SIZEOF_UCHAR);
/* Test for NULL */
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
uprv_memset(tempbuffer, 0, (sourceLength+1)*U_SIZEOF_UCHAR);
while(dest[inpsize-1] == SPACE_CHAR) {
countr++;
inpsize--;
}
i = sourceLength - countr - 1;
j = sourceLength - 1;
while(i >= 0 && j >= 0) {
if( countr>0 && isLamAlefChar(dest[i]) ) {
tempbuffer[j] = LAM_CHAR;
tempbuffer[j-1] = convertLamAlef[ dest[i] - 0xFEF5 ];
j--;
countr--;
}else {
if ((countr == 0) && isLamAlefChar(dest[i]) ) {
*pErrorCode=U_NO_SPACE_AVAILABLE;
}
tempbuffer[j] = dest[i];
}
i--;
j--;
}
if(countr > 0) {
uprv_memmove(tempbuffer, tempbuffer+countr, sourceLength*U_SIZEOF_UCHAR);
if(u_strlen(tempbuffer) < sourceLength) {
for(i=sourceLength-1;i>=sourceLength-countr;i--) {
tempbuffer[i] = SPACE_CHAR;
}
}
}
uprv_memcpy(dest, tempbuffer, sourceLength*U_SIZEOF_UCHAR);
uprv_free(tempbuffer);
destSize = sourceLength;
return destSize;
}
/*
*Name : expandCompositCharAtNear
*Function : Expands the LamAlef character into Lam + Alef, YehHamza character
* into Yeh + Hamza, SeenFamily character into SeenFamily character
* + Tail, while consuming the space next to the character.
* If there are no spaces next to the character, an error
* will be set to U_NO_SPACE_AVAILABLE as defined in utypes.h
*/
static int32_t
expandCompositCharAtNear(UChar *dest, int32_t sourceLength, int32_t destSize,UErrorCode *pErrorCode,
int yehHamzaOption, int seenTailOption, int lamAlefOption) {
int32_t i = 0;
UChar lamalefChar, yehhamzaChar;
for(i = 0 ;i<=sourceLength-1;i++) {
if (seenTailOption && isSeenTailFamilyChar(dest[i])) {
if ((i>0) && (dest[i-1] == SPACE_CHAR) ) {
dest[i-1] = tailChar;
}else {
*pErrorCode=U_NO_SPACE_AVAILABLE;
}
}else if(yehHamzaOption && (isYehHamzaChar(dest[i])) ) {
if ((i>0) && (dest[i-1] == SPACE_CHAR) ) {
yehhamzaChar = dest[i];
dest[i] = yehHamzaToYeh[yehhamzaChar - YEH_HAMZAFE_CHAR];
dest[i-1] = HAMZAFE_CHAR;
}else {
*pErrorCode=U_NO_SPACE_AVAILABLE;
}
}else if(lamAlefOption && isLamAlefChar(dest[i+1])) {
if(dest[i] == SPACE_CHAR){
lamalefChar = dest[i+1];
dest[i+1] = LAM_CHAR;
dest[i] = convertLamAlef[ lamalefChar - 0xFEF5 ];
}else {
*pErrorCode=U_NO_SPACE_AVAILABLE;
}
}
}
destSize = sourceLength;
return destSize;
}
/*
* Name : expandCompositChar
* Function : LamAlef, need special handling, since it expands from one
* character into two characters while shaping or deshaping.
* In order to expand it, near or far spaces according to the
* options user specifies. Also buffer size can be increased.
*
* For SeenFamily characters and YehHamza only the near option is
* supported, while for LamAlef we can take spaces from begin, end,
* near or even increase the buffer size.
* There is also the Auto option for LamAlef only, which will first
* search for a space at end, begin then near, respectively.
* If there are no spaces to expand these characters, an error will be set to
* U_NO_SPACE_AVAILABLE as defined in utypes.h
*/
static int32_t
expandCompositChar(UChar *dest, int32_t sourceLength,
int32_t destSize,uint32_t options,
UErrorCode *pErrorCode, int shapingMode) {
int32_t i = 0,j = 0;
UChar *tempbuffer=NULL;
int yehHamzaOption = 0;
int seenTailOption = 0;
int lamAlefOption = 0;
if (shapingMode == 1){
if ( (options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_AUTO){
if(spacesRelativeToTextBeginEnd == 0) {
destSize = expandCompositCharAtEnd(dest, sourceLength, destSize, pErrorCode);
if(*pErrorCode == U_NO_SPACE_AVAILABLE) {
*pErrorCode = U_ZERO_ERROR;
destSize = expandCompositCharAtBegin(dest, sourceLength, destSize, pErrorCode);
}
}else {
destSize = expandCompositCharAtBegin(dest, sourceLength, destSize, pErrorCode);
if(*pErrorCode == U_NO_SPACE_AVAILABLE) {
*pErrorCode = U_ZERO_ERROR;
destSize = expandCompositCharAtEnd(dest, sourceLength, destSize, pErrorCode);
}
}
if(*pErrorCode == U_NO_SPACE_AVAILABLE) {
*pErrorCode = U_ZERO_ERROR;
destSize = expandCompositCharAtNear(dest, sourceLength, destSize, pErrorCode, yehHamzaOption,
seenTailOption, 1);
}
}
}
if (shapingMode == 1){
if ( (options&U_SHAPE_LAMALEF_MASK) == uShapeLamalefEnd){
destSize = expandCompositCharAtEnd(dest, sourceLength, destSize, pErrorCode);
}
}
if (shapingMode == 1){
if ( (options&U_SHAPE_LAMALEF_MASK) == uShapeLamalefBegin){
destSize = expandCompositCharAtBegin(dest, sourceLength, destSize, pErrorCode);
}
}
if (shapingMode == 0){
if ((options&U_SHAPE_YEHHAMZA_MASK) == U_SHAPE_YEHHAMZA_TWOCELL_NEAR){
yehHamzaOption = 1;
}
if ((options&U_SHAPE_SEEN_MASK) == U_SHAPE_SEEN_TWOCELL_NEAR){
seenTailOption = 1;
}
}
if (shapingMode == 1) {
if ( (options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_NEAR) {
lamAlefOption = 1;
}
}
if (yehHamzaOption || seenTailOption || lamAlefOption){
destSize = expandCompositCharAtNear(dest, sourceLength, destSize, pErrorCode, yehHamzaOption,
seenTailOption,lamAlefOption);
}
if (shapingMode == 1){
if ( (options&U_SHAPE_LAMALEF_MASK) == U_SHAPE_LAMALEF_RESIZE){
destSize = calculateSize(dest,sourceLength,destSize,options);
tempbuffer = (UChar *)uprv_malloc((destSize+1)*U_SIZEOF_UCHAR);
/* Test for NULL */
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
uprv_memset(tempbuffer, 0, (destSize+1)*U_SIZEOF_UCHAR);
i = j = 0;
while(i < destSize && j < destSize) {
if(isLamAlefChar(dest[i]) ) {
tempbuffer[j] = convertLamAlef[ dest[i] - 0xFEF5 ];
tempbuffer[j+1] = LAM_CHAR;
j++;
}else {
tempbuffer[j] = dest[i];
}
i++;
j++;
}
uprv_memcpy(dest, tempbuffer, destSize*U_SIZEOF_UCHAR);
destSize = sourceLength;
}
}
if(tempbuffer) {
uprv_free(tempbuffer);
}
return destSize;
}
/*
*Name : shapeUnicode
*Function : Converts an Arabic Unicode buffer in 06xx Range into a shaped
* arabic Unicode buffer in FExx Range
*/
static int32_t
shapeUnicode(UChar *dest, int32_t sourceLength,
int32_t destSize,uint32_t options,
UErrorCode *pErrorCode,
int tashkeelFlag) {
int32_t i, iend;
int32_t step;
int32_t lastPos,Nx, Nw;
unsigned int Shape;
int32_t lamalef_found = 0;
int32_t seenfamFound = 0, yehhamzaFound =0, tashkeelFound = 0;
UChar prevLink = 0, lastLink = 0, currLink, nextLink = 0;
UChar wLamalef;
/*
* Converts the input buffer from FExx Range into 06xx Range
* to make sure that all characters are in the 06xx range
* even the lamalef is converted to the special region in
* the 06xx range
*/
if ((options & U_SHAPE_PRESERVE_PRESENTATION_MASK) == U_SHAPE_PRESERVE_PRESENTATION_NOOP) {
for (i = 0; i < sourceLength; i++) {
UChar inputChar = dest[i];
if ( (inputChar >= 0xFB50) && (inputChar <= 0xFBFF)) {
UChar c = convertFBto06 [ (inputChar - 0xFB50) ];
if (c != 0)
dest[i] = c;
} else if ( (inputChar >= 0xFE70) && (inputChar <= 0xFEFC)) {
dest[i] = convertFEto06 [ (inputChar - 0xFE70) ] ;
} else {
dest[i] = inputChar ;
}
}
}
/* sets the index to the end of the buffer, together with the step point to -1 */
i = sourceLength - 1;
iend = -1;
step = -1;
/*
* This function resolves the link between the characters .
* Arabic characters have four forms :
* Isolated Form, Initial Form, Middle Form and Final Form
*/
currLink = getLink(dest[i]);
lastPos = i;
Nx = -2, Nw = 0;
while (i != iend) {
/* If high byte of currLink > 0 then more than one shape */
if ((currLink & 0xFF00) > 0 || (getLink(dest[i]) & IRRELEVANT) != 0) {
Nw = i + step;
while (Nx < 0) { /* we need to know about next char */
if(Nw == iend) {
nextLink = 0;
Nx = 3000;
} else {
nextLink = getLink(dest[Nw]);
if((nextLink & IRRELEVANT) == 0) {
Nx = Nw;
} else {
Nw = Nw + step;
}
}
}
if ( ((currLink & ALEFTYPE) > 0) && ((lastLink & LAMTYPE) > 0) ) {
lamalef_found = 1;
wLamalef = changeLamAlef(dest[i]); /*get from 0x065C-0x065f */
if ( wLamalef != 0) {
dest[i] = LAMALEF_SPACE_SUB; /* The default case is to drop the Alef and replace */
dest[lastPos] =wLamalef; /* it by LAMALEF_SPACE_SUB which is the last character in the */
i=lastPos; /* unicode private use area, this is done to make */
} /* sure that removeLamAlefSpaces() handles only the */
lastLink = prevLink; /* spaces generated during lamalef generation. */
currLink = getLink(wLamalef); /* LAMALEF_SPACE_SUB is added here and is replaced by spaces */
} /* in removeLamAlefSpaces() */
if ((i > 0) && (dest[i-1] == SPACE_CHAR)){
if ( isSeenFamilyChar(dest[i])){
seenfamFound = 1;
} else if (dest[i] == YEH_HAMZA_CHAR) {
yehhamzaFound = 1;
}
}
else if(i==0){
if ( isSeenFamilyChar(dest[i])){
seenfamFound = 1;
} else if (dest[i] == YEH_HAMZA_CHAR) {
yehhamzaFound = 1;
}
}
/*
* get the proper shape according to link ability of neighbors
* and of character; depends on the order of the shapes
* (isolated, initial, middle, final) in the compatibility area
*/
Shape = shapeTable[nextLink & (LINKR + LINKL)]
[lastLink & (LINKR + LINKL)]
[currLink & (LINKR + LINKL)];
if ((currLink & (LINKR+LINKL)) == 1) {
Shape &= 1;
} else if(isTashkeelChar(dest[i])) {
if( (lastLink & LINKL) && (nextLink & LINKR) && (tashkeelFlag == 1) &&
dest[i] != 0x064C && dest[i] != 0x064D )
{
Shape = 1;
if( (nextLink&ALEFTYPE) == ALEFTYPE && (lastLink&LAMTYPE) == LAMTYPE ) {
Shape = 0;
}
}
else {
Shape = 0;
}
}
if ((dest[i] ^ 0x0600) < 0x100) {
if ( isTashkeelChar(dest[i]) ){
if (tashkeelFlag == 2){
dest[i] = TASHKEEL_SPACE_SUB;
tashkeelFound = 1;
}else {
dest[i] = 0xFE70 + IrrelevantPos[(dest[i] - 0x064B)] + Shape;
}
}else if ((currLink & APRESENT) > 0) {
dest[i] = (UChar)(0xFB50 + (currLink >> 8) + Shape);
}else if ((currLink >> 8) > 0 && (currLink & IRRELEVANT) == 0) {
dest[i] = (UChar)(0xFE70 + (currLink >> 8) + Shape);
}
}
}
/* move one notch forward */
if ((currLink & IRRELEVANT) == 0) {
prevLink = lastLink;
lastLink = currLink;
lastPos = i;
}
i = i + step;
if (i == Nx) {
currLink = nextLink;
Nx = -2;
} else if(i != iend) {
currLink = getLink(dest[i]);
}
}
destSize = sourceLength;
if ( (lamalef_found != 0 ) || (tashkeelFound != 0) ){
destSize = handleGeneratedSpaces(dest,sourceLength,destSize,options,pErrorCode);
}
if ( (seenfamFound != 0) || (yehhamzaFound != 0) ) {
destSize = expandCompositChar(dest, sourceLength,destSize,options,pErrorCode, SHAPE_MODE);
}
return destSize;
}
/*
*Name : deShapeUnicode
*Function : Converts an Arabic Unicode buffer in FExx Range into unshaped
* arabic Unicode buffer in 06xx Range
*/
static int32_t
deShapeUnicode(UChar *dest, int32_t sourceLength,
int32_t destSize,uint32_t options,
UErrorCode *pErrorCode) {
int32_t i = 0;
int32_t lamalef_found = 0;
int32_t yehHamzaComposeEnabled = 0;
int32_t seenComposeEnabled = 0;
yehHamzaComposeEnabled = ((options&U_SHAPE_YEHHAMZA_MASK) == U_SHAPE_YEHHAMZA_TWOCELL_NEAR) ? 1 : 0;
seenComposeEnabled = ((options&U_SHAPE_SEEN_MASK) == U_SHAPE_SEEN_TWOCELL_NEAR)? 1 : 0;
/*
*This for loop changes the buffer from the Unicode FE range to
*the Unicode 06 range
*/
for(i = 0; i < sourceLength; i++) {
UChar inputChar = dest[i];
if ( (inputChar >= 0xFB50) && (inputChar <= 0xFBFF)) { /* FBxx Arabic range */
UChar c = convertFBto06 [ (inputChar - 0xFB50) ];
if (c != 0)
dest[i] = c;
} else if( (yehHamzaComposeEnabled == 1) && ((inputChar == HAMZA06_CHAR) || (inputChar == HAMZAFE_CHAR))
&& (i < (sourceLength - 1)) && isAlefMaksouraChar(dest[i+1] )) {
dest[i] = SPACE_CHAR;
dest[i+1] = YEH_HAMZA_CHAR;
} else if ( (seenComposeEnabled == 1) && (isTailChar(inputChar)) && (i< (sourceLength - 1))
&& (isSeenTailFamilyChar(dest[i+1])) ) {
dest[i] = SPACE_CHAR;
} else if (( inputChar >= 0xFE70) && (inputChar <= 0xFEF4 )) { /* FExx Arabic range */
dest[i] = convertFEto06 [ (inputChar - 0xFE70) ];
} else {
dest[i] = inputChar ;
}
if( isLamAlefChar(dest[i]) )
lamalef_found = 1;
}
destSize = sourceLength;
if (lamalef_found != 0){
destSize = expandCompositChar(dest,sourceLength,destSize,options,pErrorCode,DESHAPE_MODE);
}
return destSize;
}
/*
****************************************
* u_shapeArabic
****************************************
*/
U_CAPI int32_t U_EXPORT2
u_shapeArabic(const UChar *source, int32_t sourceLength,
UChar *dest, int32_t destCapacity,
uint32_t options,
UErrorCode *pErrorCode) {
int32_t destLength;
spacesRelativeToTextBeginEnd = 0;
uShapeLamalefBegin = U_SHAPE_LAMALEF_BEGIN;
uShapeLamalefEnd = U_SHAPE_LAMALEF_END;
uShapeTashkeelBegin = U_SHAPE_TASHKEEL_BEGIN;
uShapeTashkeelEnd = U_SHAPE_TASHKEEL_END;
/* usual error checking */
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
return 0;
}
/* make sure that no reserved options values are used; allow dest==NULL only for preflighting */
if( source==NULL || sourceLength<-1 || (dest==NULL && destCapacity!=0) || destCapacity<0 ||
(((options&U_SHAPE_TASHKEEL_MASK) > 0) &&
((options&U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED) == U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED) ) ||
(((options&U_SHAPE_TASHKEEL_MASK) > 0) &&
((options&U_SHAPE_LETTERS_MASK) == U_SHAPE_LETTERS_UNSHAPE)) ||
(options&U_SHAPE_DIGIT_TYPE_RESERVED)==U_SHAPE_DIGIT_TYPE_RESERVED ||
(options&U_SHAPE_DIGITS_MASK)==U_SHAPE_DIGITS_RESERVED ||
((options&U_SHAPE_LAMALEF_MASK) != U_SHAPE_LAMALEF_RESIZE &&
(options&U_SHAPE_AGGREGATE_TASHKEEL_MASK) != 0) ||
((options&U_SHAPE_AGGREGATE_TASHKEEL_MASK) == U_SHAPE_AGGREGATE_TASHKEEL &&
(options&U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED) != U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED)
)
{
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
return 0;
}
/* Validate lamalef options */
if(((options&U_SHAPE_LAMALEF_MASK) > 0)&&
!(((options & U_SHAPE_LAMALEF_MASK)==U_SHAPE_LAMALEF_BEGIN) ||
((options & U_SHAPE_LAMALEF_MASK)==U_SHAPE_LAMALEF_END ) ||
((options & U_SHAPE_LAMALEF_MASK)==U_SHAPE_LAMALEF_RESIZE )||
((options & U_SHAPE_LAMALEF_MASK)==U_SHAPE_LAMALEF_AUTO) ||
((options & U_SHAPE_LAMALEF_MASK)==U_SHAPE_LAMALEF_NEAR)))
{
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
return 0;
}
/* Validate Tashkeel options */
if(((options&U_SHAPE_TASHKEEL_MASK) > 0)&&
!(((options & U_SHAPE_TASHKEEL_MASK)==U_SHAPE_TASHKEEL_BEGIN) ||
((options & U_SHAPE_TASHKEEL_MASK)==U_SHAPE_TASHKEEL_END )
||((options & U_SHAPE_TASHKEEL_MASK)==U_SHAPE_TASHKEEL_RESIZE )||
((options & U_SHAPE_TASHKEEL_MASK)==U_SHAPE_TASHKEEL_REPLACE_BY_TATWEEL)))
{
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
return 0;
}
/* determine the source length */
if(sourceLength==-1) {
sourceLength=u_strlen(source);
}
if(sourceLength<=0) {
return u_terminateUChars(dest, destCapacity, 0, pErrorCode);
}
/* check that source and destination do not overlap */
if( dest!=NULL &&
((source<=dest && dest<source+sourceLength) ||
(dest<=source && source<dest+destCapacity))) {
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
return 0;
}
/* Does Options contain the new Seen Tail Unicode code point option */
if ( (options&SHAPE_TAIL_TYPE_MASK) == SHAPE_TAIL_NEW_UNICODE){
tailChar = NEW_TAIL_CHAR;
}else {
tailChar = OLD_TAIL_CHAR;
}
if((options&U_SHAPE_LETTERS_MASK)!=U_SHAPE_LETTERS_NOOP) {
UChar buffer[300];
UChar *tempbuffer, *tempsource = NULL;
int32_t outputSize, spacesCountl=0, spacesCountr=0;
if((options&U_SHAPE_AGGREGATE_TASHKEEL_MASK)>0) {
int32_t logical_order = (options&U_SHAPE_TEXT_DIRECTION_MASK) == U_SHAPE_TEXT_DIRECTION_LOGICAL;
int32_t aggregate_tashkeel =
(options&(U_SHAPE_AGGREGATE_TASHKEEL_MASK+U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED)) ==
(U_SHAPE_AGGREGATE_TASHKEEL+U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED);
int step=logical_order?1:-1;
int j=logical_order?-1:2*sourceLength;
int i=logical_order?-1:sourceLength;
int end=logical_order?sourceLength:-1;
int aggregation_possible = 1;
UChar prev = 0;
UChar prevLink, currLink = 0;
int newSourceLength = 0;
tempsource = (UChar *)uprv_malloc(2*sourceLength*U_SIZEOF_UCHAR);
if(tempsource == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
return 0;
}
while ((i+=step) != end) {
prevLink = currLink;
currLink = getLink(source[i]);
if (aggregate_tashkeel && ((prevLink|currLink)&COMBINE) == COMBINE && aggregation_possible) {
aggregation_possible = 0;
tempsource[j] = (prev<source[i]?prev:source[i])-0x064C+0xFC5E;
currLink = getLink(tempsource[j]);
} else {
aggregation_possible = 1;
tempsource[j+=step] = source[i];
prev = source[i];
newSourceLength++;
}
}
source = tempsource+(logical_order?0:j);
sourceLength = newSourceLength;
}
/* calculate destination size */
/* TODO: do we ever need to do this pure preflighting? */
if(((options&U_SHAPE_LAMALEF_MASK)==U_SHAPE_LAMALEF_RESIZE) ||
((options&U_SHAPE_TASHKEEL_MASK)==U_SHAPE_TASHKEEL_RESIZE)) {
outputSize=calculateSize(source,sourceLength,destCapacity,options);
} else {
outputSize=sourceLength;
}
if(outputSize>destCapacity) {
*pErrorCode=U_BUFFER_OVERFLOW_ERROR;
if (tempsource != NULL) uprv_free(tempsource);
return outputSize;
}
/*
* need a temporary buffer of size max(outputSize, sourceLength)
* because at first we copy source->temp
*/
if(sourceLength>outputSize) {
outputSize=sourceLength;
}
/* Start of Arabic letter shaping part */
if(outputSize<=sizeof(buffer)/U_SIZEOF_UCHAR) {
outputSize=sizeof(buffer)/U_SIZEOF_UCHAR;
tempbuffer=buffer;
} else {
tempbuffer = (UChar *)uprv_malloc(outputSize*U_SIZEOF_UCHAR);
/*Test for NULL*/
if(tempbuffer == NULL) {
*pErrorCode = U_MEMORY_ALLOCATION_ERROR;
if (tempsource != NULL) uprv_free(tempsource);
return 0;
}
}
uprv_memcpy(tempbuffer, source, sourceLength*U_SIZEOF_UCHAR);
if (tempsource != NULL){
uprv_free(tempsource);
}
if(sourceLength<outputSize) {
uprv_memset(tempbuffer+sourceLength, 0, (outputSize-sourceLength)*U_SIZEOF_UCHAR);
}
if((options&U_SHAPE_TEXT_DIRECTION_MASK) == U_SHAPE_TEXT_DIRECTION_LOGICAL) {
countSpaces(tempbuffer,sourceLength,options,&spacesCountl,&spacesCountr);
invertBuffer(tempbuffer,sourceLength,options,spacesCountl,spacesCountr);
}
if((options&U_SHAPE_TEXT_DIRECTION_MASK) == U_SHAPE_TEXT_DIRECTION_VISUAL_LTR) {
if((options&U_SHAPE_SPACES_RELATIVE_TO_TEXT_MASK) == U_SHAPE_SPACES_RELATIVE_TO_TEXT_BEGIN_END) {
spacesRelativeToTextBeginEnd = 1;
uShapeLamalefBegin = U_SHAPE_LAMALEF_END;
uShapeLamalefEnd = U_SHAPE_LAMALEF_BEGIN;
uShapeTashkeelBegin = U_SHAPE_TASHKEEL_END;
uShapeTashkeelEnd = U_SHAPE_TASHKEEL_BEGIN;
}
}
switch(options&U_SHAPE_LETTERS_MASK) {
case U_SHAPE_LETTERS_SHAPE :
if( (options&U_SHAPE_TASHKEEL_MASK)> 0
&& ((options&U_SHAPE_TASHKEEL_MASK) !=U_SHAPE_TASHKEEL_REPLACE_BY_TATWEEL)) {
/* Call the shaping function with tashkeel flag == 2 for removal of tashkeel */
destLength = shapeUnicode(tempbuffer,sourceLength,destCapacity,options,pErrorCode,2);
}else {
/* default Call the shaping function with tashkeel flag == 1 */
destLength = shapeUnicode(tempbuffer,sourceLength,destCapacity,options,pErrorCode,1);
/*After shaping text check if user wants to remove tashkeel and replace it with tatweel*/
if( (options&U_SHAPE_TASHKEEL_MASK) == U_SHAPE_TASHKEEL_REPLACE_BY_TATWEEL){
destLength = handleTashkeelWithTatweel(tempbuffer,destLength,destCapacity,options,pErrorCode);
}
}
break;
case U_SHAPE_LETTERS_SHAPE_TASHKEEL_ISOLATED :
/* Call the shaping function with tashkeel flag == 0 */
destLength = shapeUnicode(tempbuffer,sourceLength,destCapacity,options,pErrorCode,0);
break;
case U_SHAPE_LETTERS_UNSHAPE :
/* Call the deshaping function */
destLength = deShapeUnicode(tempbuffer,sourceLength,destCapacity,options,pErrorCode);
break;
default :
/* will never occur because of validity checks above */
destLength = 0;
break;
}
/*
* TODO: (markus 2002aug01)
* For as long as we always preflight the outputSize above
* we should U_ASSERT(outputSize==destLength)
* except for the adjustment above before the tempbuffer allocation
*/
if((options&U_SHAPE_TEXT_DIRECTION_MASK) == U_SHAPE_TEXT_DIRECTION_LOGICAL) {
countSpaces(tempbuffer,destLength,options,&spacesCountl,&spacesCountr);
invertBuffer(tempbuffer,destLength,options,spacesCountl,spacesCountr);
}
uprv_memcpy(dest, tempbuffer, uprv_min(destLength, destCapacity)*U_SIZEOF_UCHAR);
if(tempbuffer!=buffer) {
uprv_free(tempbuffer);
}
if(destLength>destCapacity) {
*pErrorCode=U_BUFFER_OVERFLOW_ERROR;
return destLength;
}
/* End of Arabic letter shaping part */
} else {
/*
* No letter shaping:
* just make sure the destination is large enough and copy the string.
*/
if(destCapacity<sourceLength) {
/* this catches preflighting, too */
*pErrorCode=U_BUFFER_OVERFLOW_ERROR;
return sourceLength;
}
uprv_memcpy(dest, source, sourceLength*U_SIZEOF_UCHAR);
destLength=sourceLength;
}
/*
* Perform number shaping.
* With UTF-16 or UTF-32, the length of the string is constant.
* The easiest way to do this is to operate on the destination and
* "shape" the digits in-place.
*/
if((options&U_SHAPE_DIGITS_MASK)!=U_SHAPE_DIGITS_NOOP) {
UChar digitBase;
int32_t i;
/* select the requested digit group */
switch(options&U_SHAPE_DIGIT_TYPE_MASK) {
case U_SHAPE_DIGIT_TYPE_AN:
digitBase=0x660; /* Unicode: "Arabic-Indic digits" */
break;
case U_SHAPE_DIGIT_TYPE_AN_EXTENDED:
digitBase=0x6f0; /* Unicode: "Eastern Arabic-Indic digits (Persian and Urdu)" */
break;
default:
/* will never occur because of validity checks above */
digitBase=0;
break;
}
/* perform the requested operation */
switch(options&U_SHAPE_DIGITS_MASK) {
case U_SHAPE_DIGITS_EN2AN:
/* add (digitBase-'0') to each European (ASCII) digit code point */
digitBase-=0x30;
for(i=0; i<destLength; ++i) {
if(((uint32_t)dest[i]-0x30)<10) {
dest[i]+=digitBase;
}
}
break;
case U_SHAPE_DIGITS_AN2EN:
/* subtract (digitBase-'0') from each Arabic digit code point */
for(i=0; i<destLength; ++i) {
if(((uint32_t)dest[i]-(uint32_t)digitBase)<10) {
dest[i]-=digitBase-0x30;
}
}
break;
case U_SHAPE_DIGITS_ALEN2AN_INIT_LR:
_shapeToArabicDigitsWithContext(dest, destLength,
digitBase,
(UBool)((options&U_SHAPE_TEXT_DIRECTION_MASK)==U_SHAPE_TEXT_DIRECTION_LOGICAL),
FALSE);
break;
case U_SHAPE_DIGITS_ALEN2AN_INIT_AL:
_shapeToArabicDigitsWithContext(dest, destLength,
digitBase,
(UBool)((options&U_SHAPE_TEXT_DIRECTION_MASK)==U_SHAPE_TEXT_DIRECTION_LOGICAL),
TRUE);
break;
default:
/* will never occur because of validity checks above */
break;
}
}
return u_terminateUChars(dest, destCapacity, destLength, pErrorCode);
}