michael@0: /*
michael@0: ******************************************************************************
michael@0: *
michael@0: *   Copyright (C) 1999-2012, International Business Machines
michael@0: *   Corporation and others.  All Rights Reserved.
michael@0: *
michael@0: ******************************************************************************
michael@0: *   file name:  utf_impl.c
michael@0: *   encoding:   US-ASCII
michael@0: *   tab size:   8 (not used)
michael@0: *   indentation:4
michael@0: *
michael@0: *   created on: 1999sep13
michael@0: *   created by: Markus W. Scherer
michael@0: *
michael@0: *   This file provides implementation functions for macros in the utfXX.h
michael@0: *   that would otherwise be too long as macros.
michael@0: */
michael@0: 
michael@0: /* set import/export definitions */
michael@0: #ifndef U_UTF8_IMPL
michael@0: #   define U_UTF8_IMPL
michael@0: #endif
michael@0: 
michael@0: #include "unicode/utypes.h"
michael@0: #include "unicode/utf.h"
michael@0: #include "unicode/utf8.h"
michael@0: #include "unicode/utf_old.h"
michael@0: #include "uassert.h"
michael@0: 
michael@0: /*
michael@0:  * This table could be replaced on many machines by
michael@0:  * a few lines of assembler code using an
michael@0:  * "index of first 0-bit from msb" instruction and
michael@0:  * one or two more integer instructions.
michael@0:  *
michael@0:  * For example, on an i386, do something like
michael@0:  * - MOV AL, leadByte
michael@0:  * - NOT AL         (8-bit, leave b15..b8==0..0, reverse only b7..b0)
michael@0:  * - MOV AH, 0
michael@0:  * - BSR BX, AX     (16-bit)
michael@0:  * - MOV AX, 6      (result)
michael@0:  * - JZ finish      (ZF==1 if leadByte==0xff)
michael@0:  * - SUB AX, BX (result)
michael@0:  * -finish:
michael@0:  * (BSR: Bit Scan Reverse, scans for a 1-bit, starting from the MSB)
michael@0:  *
michael@0:  * In Unicode, all UTF-8 byte sequences with more than 4 bytes are illegal;
michael@0:  * lead bytes above 0xf4 are illegal.
michael@0:  * We keep them in this table for skipping long ISO 10646-UTF-8 sequences.
michael@0:  */
michael@0: U_EXPORT const uint8_t 
michael@0: utf8_countTrailBytes[256]={
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0: 
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0: 
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0:     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
michael@0: 
michael@0:     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
michael@0:     1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
michael@0: 
michael@0:     2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
michael@0:     3, 3, 3, 3, 3,
michael@0:     3, 3, 3,    /* illegal in Unicode */
michael@0:     4, 4, 4, 4, /* illegal in Unicode */
michael@0:     5, 5,       /* illegal in Unicode */
michael@0:     0, 0        /* illegal bytes 0xfe and 0xff */
michael@0: };
michael@0: 
michael@0: static const UChar32
michael@0: utf8_minLegal[4]={ 0, 0x80, 0x800, 0x10000 };
michael@0: 
michael@0: static const UChar32
michael@0: utf8_errorValue[6]={
michael@0:     UTF8_ERROR_VALUE_1, UTF8_ERROR_VALUE_2, UTF_ERROR_VALUE, 0x10ffff,
michael@0:     0x3ffffff, 0x7fffffff
michael@0: };
michael@0: 
michael@0: static UChar32
michael@0: errorValue(int32_t count, int8_t strict) {
michael@0:     if(strict>=0) {
michael@0:         return utf8_errorValue[count];
michael@0:     } else if(strict==-3) {
michael@0:         return 0xfffd;
michael@0:     } else {
michael@0:         return U_SENTINEL;
michael@0:     }
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Handle the non-inline part of the U8_NEXT() and U8_NEXT_FFFD() macros
michael@0:  * and their obsolete sibling UTF8_NEXT_CHAR_SAFE().
michael@0:  *
michael@0:  * U8_NEXT() supports NUL-terminated strings indicated via length<0.
michael@0:  *
michael@0:  * The "strict" parameter controls the error behavior:
michael@0:  * <0  "Safe" behavior of U8_NEXT():
michael@0:  *     -1: All illegal byte sequences yield U_SENTINEL=-1.
michael@0:  *     -2: Same as -1, except for lenient treatment of surrogate code points as legal.
michael@0:  *         Some implementations use this for roundtripping of
michael@0:  *         Unicode 16-bit strings that are not well-formed UTF-16, that is, they
michael@0:  *         contain unpaired surrogates.
michael@0:  *     -3: All illegal byte sequences yield U+FFFD.
michael@0:  *  0  Obsolete "safe" behavior of UTF8_NEXT_CHAR_SAFE(..., FALSE):
michael@0:  *     All illegal byte sequences yield a positive code point such that this
michael@0:  *     result code point would be encoded with the same number of bytes as
michael@0:  *     the illegal sequence.
michael@0:  * >0  Obsolete "strict" behavior of UTF8_NEXT_CHAR_SAFE(..., TRUE):
michael@0:  *     Same as the obsolete "safe" behavior, but non-characters are also treated
michael@0:  *     like illegal sequences.
michael@0:  *
michael@0:  * Note that a UBool is the same as an int8_t.
michael@0:  */
michael@0: U_CAPI UChar32 U_EXPORT2
michael@0: utf8_nextCharSafeBody(const uint8_t *s, int32_t *pi, int32_t length, UChar32 c, UBool strict) {
michael@0:     int32_t i=*pi;
michael@0:     uint8_t count=U8_COUNT_TRAIL_BYTES(c);
michael@0:     U_ASSERT(count <= 5); /* U8_COUNT_TRAIL_BYTES returns value 0...5 */
michael@0:     if(i+count<=length || length<0) {
michael@0:         uint8_t trail;
michael@0: 
michael@0:         U8_MASK_LEAD_BYTE(c, count);
michael@0:         /* support NUL-terminated strings: do not read beyond the first non-trail byte */
michael@0:         switch(count) {
michael@0:         /* each branch falls through to the next one */
michael@0:         case 0:
michael@0:             /* count==0 for illegally leading trail bytes and the illegal bytes 0xfe and 0xff */
michael@0:         case 5:
michael@0:         case 4:
michael@0:             /* count>=4 is always illegal: no more than 3 trail bytes in Unicode's UTF-8 */
michael@0:             break;
michael@0:         case 3:
michael@0:             trail=s[i++]-0x80;
michael@0:             c=(c<<6)|trail;
michael@0:             /* c>=0x110 would result in code point>0x10ffff, outside Unicode */
michael@0:             if(c>=0x110 || trail>0x3f) { break; }
michael@0:         case 2:
michael@0:             trail=s[i++]-0x80;
michael@0:             c=(c<<6)|trail;
michael@0:             /*
michael@0:              * test for a surrogate d800..dfff unless we are lenient:
michael@0:              * before the last (c<<6), a surrogate is c=360..37f
michael@0:              */
michael@0:             if(((c&0xffe0)==0x360 && strict!=-2) || trail>0x3f) { break; }
michael@0:         case 1:
michael@0:             trail=s[i++]-0x80;
michael@0:             c=(c<<6)|trail;
michael@0:             if(trail>0x3f) { break; }
michael@0:             /* correct sequence - all trail bytes have (b7..b6)==(10) */
michael@0:             if(c>=utf8_minLegal[count] &&
michael@0:                     /* strict: forbid non-characters like U+fffe */
michael@0:                     (strict<=0 || !U_IS_UNICODE_NONCHAR(c))) {
michael@0:                 *pi=i;
michael@0:                 return c;
michael@0:             }
michael@0:         /* no default branch to optimize switch()  - all values are covered */
michael@0:         }
michael@0:     } else {
michael@0:         /* too few bytes left */
michael@0:         count=length-i;
michael@0:     }
michael@0: 
michael@0:     /* error handling */
michael@0:     i=*pi;
michael@0:     while(count>0 && U8_IS_TRAIL(s[i])) {
michael@0:         ++i;
michael@0:         --count;
michael@0:     }
michael@0:     c=errorValue(i-*pi, strict);
michael@0:     *pi=i;
michael@0:     return c;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: utf8_appendCharSafeBody(uint8_t *s, int32_t i, int32_t length, UChar32 c, UBool *pIsError) {
michael@0:     if((uint32_t)(c)<=0x7ff) {
michael@0:         if((i)+1<(length)) {
michael@0:             (s)[(i)++]=(uint8_t)(((c)>>6)|0xc0);
michael@0:             (s)[(i)++]=(uint8_t)(((c)&0x3f)|0x80);
michael@0:             return i;
michael@0:         }
michael@0:     } else if((uint32_t)(c)<=0xffff) {
michael@0:         /* Starting with Unicode 3.2, surrogate code points must not be encoded in UTF-8. */
michael@0:         if((i)+2<(length) && !U_IS_SURROGATE(c)) {
michael@0:             (s)[(i)++]=(uint8_t)(((c)>>12)|0xe0);
michael@0:             (s)[(i)++]=(uint8_t)((((c)>>6)&0x3f)|0x80);
michael@0:             (s)[(i)++]=(uint8_t)(((c)&0x3f)|0x80);
michael@0:             return i;
michael@0:         }
michael@0:     } else if((uint32_t)(c)<=0x10ffff) {
michael@0:         if((i)+3<(length)) {
michael@0:             (s)[(i)++]=(uint8_t)(((c)>>18)|0xf0);
michael@0:             (s)[(i)++]=(uint8_t)((((c)>>12)&0x3f)|0x80);
michael@0:             (s)[(i)++]=(uint8_t)((((c)>>6)&0x3f)|0x80);
michael@0:             (s)[(i)++]=(uint8_t)(((c)&0x3f)|0x80);
michael@0:             return i;
michael@0:         }
michael@0:     }
michael@0:     /* c>0x10ffff or not enough space, write an error value */
michael@0:     if(pIsError!=NULL) {
michael@0:         *pIsError=TRUE;
michael@0:     } else {
michael@0:         length-=i;
michael@0:         if(length>0) {
michael@0:             int32_t offset;
michael@0:             if(length>3) {
michael@0:                 length=3;
michael@0:             }
michael@0:             s+=i;
michael@0:             offset=0;
michael@0:             c=utf8_errorValue[length-1];
michael@0:             UTF8_APPEND_CHAR_UNSAFE(s, offset, c);
michael@0:             i=i+offset;
michael@0:         }
michael@0:     }
michael@0:     return i;
michael@0: }
michael@0: 
michael@0: U_CAPI UChar32 U_EXPORT2
michael@0: utf8_prevCharSafeBody(const uint8_t *s, int32_t start, int32_t *pi, UChar32 c, UBool strict) {
michael@0:     int32_t i=*pi;
michael@0:     uint8_t b, count=1, shift=6;
michael@0: 
michael@0:     if(!U8_IS_TRAIL(c)) { return errorValue(0, strict); }
michael@0: 
michael@0:     /* extract value bits from the last trail byte */
michael@0:     c&=0x3f;
michael@0: 
michael@0:     for(;;) {
michael@0:         if(i<=start) {
michael@0:             /* no lead byte at all */
michael@0:             return errorValue(0, strict);
michael@0:         }
michael@0: 
michael@0:         /* read another previous byte */
michael@0:         b=s[--i];
michael@0:         if((uint8_t)(b-0x80)<0x7e) { /* 0x80<=b<0xfe */
michael@0:             if(b&0x40) {
michael@0:                 /* lead byte, this will always end the loop */
michael@0:                 uint8_t shouldCount=U8_COUNT_TRAIL_BYTES(b);
michael@0: 
michael@0:                 if(count==shouldCount) {
michael@0:                     /* set the new position */
michael@0:                     *pi=i;
michael@0:                     U8_MASK_LEAD_BYTE(b, count);
michael@0:                     c|=(UChar32)b<<shift;
michael@0:                     if(count>=4 || c>0x10ffff || c<utf8_minLegal[count] || (U_IS_SURROGATE(c) && strict!=-2) || (strict>0 && U_IS_UNICODE_NONCHAR(c))) {
michael@0:                         /* illegal sequence or (strict and non-character) */
michael@0:                         if(count>=4) {
michael@0:                             count=3;
michael@0:                         }
michael@0:                         c=errorValue(count, strict);
michael@0:                     } else {
michael@0:                         /* exit with correct c */
michael@0:                     }
michael@0:                 } else {
michael@0:                     /* the lead byte does not match the number of trail bytes */
michael@0:                     /* only set the position to the lead byte if it would
michael@0:                        include the trail byte that we started with */
michael@0:                     if(count<shouldCount) {
michael@0:                         *pi=i;
michael@0:                         c=errorValue(count, strict);
michael@0:                     } else {
michael@0:                         c=errorValue(0, strict);
michael@0:                     }
michael@0:                 }
michael@0:                 break;
michael@0:             } else if(count<5) {
michael@0:                 /* trail byte */
michael@0:                 c|=(UChar32)(b&0x3f)<<shift;
michael@0:                 ++count;
michael@0:                 shift+=6;
michael@0:             } else {
michael@0:                 /* more than 5 trail bytes is illegal */
michael@0:                 c=errorValue(0, strict);
michael@0:                 break;
michael@0:             }
michael@0:         } else {
michael@0:             /* single-byte character precedes trailing bytes */
michael@0:             c=errorValue(0, strict);
michael@0:             break;
michael@0:         }
michael@0:     }
michael@0:     return c;
michael@0: }
michael@0: 
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: utf8_back1SafeBody(const uint8_t *s, int32_t start, int32_t i) {
michael@0:     /* i had been decremented once before the function call */
michael@0:     int32_t I=i, Z;
michael@0:     uint8_t b;
michael@0: 
michael@0:     /* read at most the 6 bytes s[Z] to s[i], inclusively */
michael@0:     if(I-5>start) {
michael@0:         Z=I-5;
michael@0:     } else {
michael@0:         Z=start;
michael@0:     }
michael@0: 
michael@0:     /* return I if the sequence starting there is long enough to include i */
michael@0:     do {
michael@0:         b=s[I];
michael@0:         if((uint8_t)(b-0x80)>=0x7e) { /* not 0x80<=b<0xfe */
michael@0:             break;
michael@0:         } else if(b>=0xc0) {
michael@0:             if(U8_COUNT_TRAIL_BYTES(b)>=(i-I)) {
michael@0:                 return I;
michael@0:             } else {
michael@0:                 break;
michael@0:             }
michael@0:         }
michael@0:     } while(Z<=--I);
michael@0: 
michael@0:     /* return i itself to be consistent with the FWD_1 macro */
michael@0:     return i;
michael@0: }