You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
563 lines
18 KiB
563 lines
18 KiB
/*
|
|
*******************************************************************************
|
|
*
|
|
* © 2016 and later: Unicode, Inc. and others.
|
|
* License & terms of use: http://www.unicode.org/copyright.html
|
|
*
|
|
*******************************************************************************
|
|
*******************************************************************************
|
|
*
|
|
* Copyright (C) 2003-2006, International Business Machines
|
|
* Corporation and others. All Rights Reserved.
|
|
*
|
|
*******************************************************************************
|
|
* file name: uit_len8.c
|
|
* encoding: UTF-8
|
|
* tab size: 8 (not used)
|
|
* indentation:4
|
|
*
|
|
* created on: 2003feb10
|
|
* created by: Markus W. Scherer
|
|
*
|
|
* This file contains the implementation of the "lenient UTF-8" UCharIterator
|
|
* as used in the uciter8 sample code.
|
|
* UTF-8-style macros are defined as well as the UCharIterator.
|
|
* The macros are incomplete (do not assemble code points from pairs of
|
|
* surrogates, see comment below)
|
|
* but sufficient for the iterator.
|
|
*/
|
|
|
|
#include <string.h>
|
|
#include "unicode/utypes.h"
|
|
#include "unicode/uiter.h"
|
|
|
|
/* lenient UTF-8/CESU-8 macros ---------------------------------------------- */
|
|
|
|
/*
|
|
* This code leniently reads 8-bit Unicode strings,
|
|
* which could contain a mix of UTF-8 and CESU-8.
|
|
* More precisely:
|
|
* - supplementary code points may be encoded with dedicated 4-byte sequences
|
|
* (UTF-8 style)
|
|
* - supplementary code points may be encoded with
|
|
* pairs of 3-byte sequences, one for each surrogate of the UTF-16 form
|
|
* (CESU-8 style)
|
|
* - single surrogates are allowed, encoded with their "natural" 3-byte sequences
|
|
*
|
|
* Limitation:
|
|
* Right now, the macros do not attempt to assemble code points from pairs of
|
|
* separately encoded surrogates.
|
|
* This would not be sufficient for processing based on these macros,
|
|
* but it is sufficient for a UCharIterator that returns only UChars anyway.
|
|
*
|
|
* The code is copied and modified from utf_impl.c and utf8.h.
|
|
*
|
|
* Change 2006feb08: Much of the implementation code is replaced by calling
|
|
* the utf_impl.c functions which accept a new "strict" parameter value
|
|
* of -2 implementing exactly this leniency.
|
|
*/
|
|
|
|
#define L8_NEXT(s, i, length, c) { \
|
|
(c)=(uint8_t)(s)[(i)++]; \
|
|
if((c)>=0x80) { \
|
|
if(U8_IS_LEAD(c)) { \
|
|
(c)=utf8_nextCharSafeBody((const uint8_t *)s, &(i), (int32_t)(length), c, -2); \
|
|
} else { \
|
|
(c)=U_SENTINEL; \
|
|
} \
|
|
} \
|
|
}
|
|
|
|
#define L8_PREV(s, start, i, c) { \
|
|
(c)=(uint8_t)(s)[--(i)]; \
|
|
if((c)>=0x80) { \
|
|
if((c)<=0xbf) { \
|
|
(c)=utf8_prevCharSafeBody((const uint8_t *)s, start, &(i), c, -2); \
|
|
} else { \
|
|
(c)=U_SENTINEL; \
|
|
} \
|
|
} \
|
|
}
|
|
|
|
/* lenient-8 UCharIterator -------------------------------------------------- */
|
|
|
|
/*
|
|
* This is a copy of the UTF-8 UCharIterator in uiter.cpp,
|
|
* except that it uses the lenient-8-bit-Unicode macros above.
|
|
*/
|
|
|
|
/*
|
|
* Minimal implementation:
|
|
* Maintain a single-UChar buffer for an additional surrogate.
|
|
* The caller must not modify start and limit because they are used internally.
|
|
*
|
|
* Use UCharIterator fields as follows:
|
|
* context pointer to UTF-8 string
|
|
* length UTF-16 length of the string; -1 until lazy evaluation
|
|
* start current UTF-8 index
|
|
* index current UTF-16 index; may be -1="unknown" after setState()
|
|
* limit UTF-8 length of the string
|
|
* reservedField supplementary code point
|
|
*
|
|
* Since UCharIterator delivers 16-bit code units, the iteration can be
|
|
* currently in the middle of the byte sequence for a supplementary code point.
|
|
* In this case, reservedField will contain that code point and start will
|
|
* point to after the corresponding byte sequence. The UTF-16 index will be
|
|
* one less than what it would otherwise be corresponding to the UTF-8 index.
|
|
* Otherwise, reservedField will be 0.
|
|
*/
|
|
|
|
/*
|
|
* Possible optimization for NUL-terminated UTF-8 and UTF-16 strings:
|
|
* Add implementations that do not call strlen() for iteration but check for NUL.
|
|
*/
|
|
|
|
static int32_t U_CALLCONV
|
|
lenient8IteratorGetIndex(UCharIterator *iter, UCharIteratorOrigin origin) {
|
|
switch(origin) {
|
|
case UITER_ZERO:
|
|
case UITER_START:
|
|
return 0;
|
|
case UITER_CURRENT:
|
|
if(iter->index<0) {
|
|
/* the current UTF-16 index is unknown after setState(), count from the beginning */
|
|
const uint8_t *s;
|
|
UChar32 c;
|
|
int32_t i, limit, index;
|
|
|
|
s=(const uint8_t *)iter->context;
|
|
i=index=0;
|
|
limit=iter->start; /* count up to the UTF-8 index */
|
|
while(i<limit) {
|
|
L8_NEXT(s, i, limit, c);
|
|
if(c<=0xffff) {
|
|
++index;
|
|
} else {
|
|
index+=2;
|
|
}
|
|
}
|
|
|
|
iter->start=i; /* just in case setState() did not get us to a code point boundary */
|
|
if(i==iter->limit) {
|
|
iter->length=index; /* in case it was <0 or wrong */
|
|
}
|
|
if(iter->reservedField!=0) {
|
|
--index; /* we are in the middle of a supplementary code point */
|
|
}
|
|
iter->index=index;
|
|
}
|
|
return iter->index;
|
|
case UITER_LIMIT:
|
|
case UITER_LENGTH:
|
|
if(iter->length<0) {
|
|
const uint8_t *s;
|
|
UChar32 c;
|
|
int32_t i, limit, length;
|
|
|
|
s=(const uint8_t *)iter->context;
|
|
if(iter->index<0) {
|
|
/*
|
|
* the current UTF-16 index is unknown after setState(),
|
|
* we must first count from the beginning to here
|
|
*/
|
|
i=length=0;
|
|
limit=iter->start;
|
|
|
|
/* count from the beginning to the current index */
|
|
while(i<limit) {
|
|
L8_NEXT(s, i, limit, c);
|
|
if(c<=0xffff) {
|
|
++length;
|
|
} else {
|
|
length+=2;
|
|
}
|
|
}
|
|
|
|
/* assume i==limit==iter->start, set the UTF-16 index */
|
|
iter->start=i; /* just in case setState() did not get us to a code point boundary */
|
|
iter->index= iter->reservedField!=0 ? length-1 : length;
|
|
} else {
|
|
i=iter->start;
|
|
length=iter->index;
|
|
if(iter->reservedField!=0) {
|
|
++length;
|
|
}
|
|
}
|
|
|
|
/* count from the current index to the end */
|
|
limit=iter->limit;
|
|
while(i<limit) {
|
|
L8_NEXT(s, i, limit, c);
|
|
if(c<=0xffff) {
|
|
++length;
|
|
} else {
|
|
length+=2;
|
|
}
|
|
}
|
|
iter->length=length;
|
|
}
|
|
return iter->length;
|
|
default:
|
|
/* not a valid origin */
|
|
/* Should never get here! */
|
|
return -1;
|
|
}
|
|
}
|
|
|
|
static int32_t U_CALLCONV
|
|
lenient8IteratorMove(UCharIterator *iter, int32_t delta, UCharIteratorOrigin origin) {
|
|
const uint8_t *s;
|
|
UChar32 c;
|
|
int32_t pos; /* requested UTF-16 index */
|
|
int32_t i; /* UTF-8 index */
|
|
UBool havePos;
|
|
|
|
/* calculate the requested UTF-16 index */
|
|
switch(origin) {
|
|
case UITER_ZERO:
|
|
case UITER_START:
|
|
pos=delta;
|
|
havePos=true;
|
|
/* iter->index<0 (unknown) is possible */
|
|
break;
|
|
case UITER_CURRENT:
|
|
if(iter->index>=0) {
|
|
pos=iter->index+delta;
|
|
havePos=true;
|
|
} else {
|
|
/* the current UTF-16 index is unknown after setState(), use only delta */
|
|
pos=0;
|
|
havePos=false;
|
|
}
|
|
break;
|
|
case UITER_LIMIT:
|
|
case UITER_LENGTH:
|
|
if(iter->length>=0) {
|
|
pos=iter->length+delta;
|
|
havePos=true;
|
|
} else {
|
|
/* pin to the end, avoid counting the length */
|
|
iter->index=-1;
|
|
iter->start=iter->limit;
|
|
iter->reservedField=0;
|
|
if(delta>=0) {
|
|
return UITER_UNKNOWN_INDEX;
|
|
} else {
|
|
/* the current UTF-16 index is unknown, use only delta */
|
|
pos=0;
|
|
havePos=false;
|
|
}
|
|
}
|
|
break;
|
|
default:
|
|
return -1; /* Error */
|
|
}
|
|
|
|
if(havePos) {
|
|
/* shortcuts: pinning to the edges of the string */
|
|
if(pos<=0) {
|
|
iter->index=iter->start=iter->reservedField=0;
|
|
return 0;
|
|
} else if(iter->length>=0 && pos>=iter->length) {
|
|
iter->index=iter->length;
|
|
iter->start=iter->limit;
|
|
iter->reservedField=0;
|
|
return iter->index;
|
|
}
|
|
|
|
/* minimize the number of L8_NEXT/PREV operations */
|
|
if(iter->index<0 || pos<iter->index/2) {
|
|
/* go forward from the start instead of backward from the current index */
|
|
iter->index=iter->start=iter->reservedField=0;
|
|
} else if(iter->length>=0 && (iter->length-pos)<(pos-iter->index)) {
|
|
/*
|
|
* if we have the UTF-16 index and length and the new position is
|
|
* closer to the end than the current index,
|
|
* then go backward from the end instead of forward from the current index
|
|
*/
|
|
iter->index=iter->length;
|
|
iter->start=iter->limit;
|
|
iter->reservedField=0;
|
|
}
|
|
|
|
delta=pos-iter->index;
|
|
if(delta==0) {
|
|
return iter->index; /* nothing to do */
|
|
}
|
|
} else {
|
|
/* move relative to unknown UTF-16 index */
|
|
if(delta==0) {
|
|
return UITER_UNKNOWN_INDEX; /* nothing to do */
|
|
} else if(-delta>=iter->start) {
|
|
/* moving backwards by more UChars than there are UTF-8 bytes, pin to 0 */
|
|
iter->index=iter->start=iter->reservedField=0;
|
|
return 0;
|
|
} else if(delta>=(iter->limit-iter->start)) {
|
|
/* moving forward by more UChars than the remaining UTF-8 bytes, pin to the end */
|
|
iter->index=iter->length; /* may or may not be <0 (unknown) */
|
|
iter->start=iter->limit;
|
|
iter->reservedField=0;
|
|
return iter->index>=0 ? iter->index : UITER_UNKNOWN_INDEX;
|
|
}
|
|
}
|
|
|
|
/* delta!=0 */
|
|
|
|
/* move towards the requested position, pin to the edges of the string */
|
|
s=(const uint8_t *)iter->context;
|
|
pos=iter->index; /* could be <0 (unknown) */
|
|
i=iter->start;
|
|
if(delta>0) {
|
|
/* go forward */
|
|
int32_t limit=iter->limit;
|
|
if(iter->reservedField!=0) {
|
|
iter->reservedField=0;
|
|
++pos;
|
|
--delta;
|
|
}
|
|
while(delta>0 && i<limit) {
|
|
L8_NEXT(s, i, limit, c);
|
|
if(c<0xffff) {
|
|
++pos;
|
|
--delta;
|
|
} else if(delta>=2) {
|
|
pos+=2;
|
|
delta-=2;
|
|
} else /* delta==1 */ {
|
|
/* stop in the middle of a supplementary code point */
|
|
iter->reservedField=c;
|
|
++pos;
|
|
break; /* delta=0; */
|
|
}
|
|
}
|
|
if(i==limit) {
|
|
if(iter->length<0 && iter->index>=0) {
|
|
iter->length= iter->reservedField==0 ? pos : pos+1;
|
|
} else if(iter->index<0 && iter->length>=0) {
|
|
iter->index= iter->reservedField==0 ? iter->length : iter->length-1;
|
|
}
|
|
}
|
|
} else /* delta<0 */ {
|
|
/* go backward */
|
|
if(iter->reservedField!=0) {
|
|
iter->reservedField=0;
|
|
i-=4; /* we stayed behind the supplementary code point; go before it now */
|
|
--pos;
|
|
++delta;
|
|
}
|
|
while(delta<0 && i>0) {
|
|
L8_PREV(s, 0, i, c);
|
|
if(c<0xffff) {
|
|
--pos;
|
|
++delta;
|
|
} else if(delta<=-2) {
|
|
pos-=2;
|
|
delta+=2;
|
|
} else /* delta==-1 */ {
|
|
/* stop in the middle of a supplementary code point */
|
|
i+=4; /* back to behind this supplementary code point for consistent state */
|
|
iter->reservedField=c;
|
|
--pos;
|
|
break; /* delta=0; */
|
|
}
|
|
}
|
|
}
|
|
|
|
iter->start=i;
|
|
if(iter->index>=0) {
|
|
return iter->index=pos;
|
|
} else {
|
|
/* we started with index<0 (unknown) so pos is bogus */
|
|
if(i<=1) {
|
|
return iter->index=i; /* reached the beginning */
|
|
} else {
|
|
/* we still don't know the UTF-16 index */
|
|
return UITER_UNKNOWN_INDEX;
|
|
}
|
|
}
|
|
}
|
|
|
|
static UBool U_CALLCONV
|
|
lenient8IteratorHasNext(UCharIterator *iter) {
|
|
return iter->reservedField!=0 || iter->start<iter->limit;
|
|
}
|
|
|
|
static UBool U_CALLCONV
|
|
lenient8IteratorHasPrevious(UCharIterator *iter) {
|
|
return iter->start>0;
|
|
}
|
|
|
|
static UChar32 U_CALLCONV
|
|
lenient8IteratorCurrent(UCharIterator *iter) {
|
|
if(iter->reservedField!=0) {
|
|
return U16_TRAIL(iter->reservedField);
|
|
} else if(iter->start<iter->limit) {
|
|
const uint8_t *s=(const uint8_t *)iter->context;
|
|
UChar32 c;
|
|
int32_t i=iter->start;
|
|
|
|
L8_NEXT(s, i, iter->limit, c);
|
|
if(c<0) {
|
|
return 0xfffd;
|
|
} else if(c<=0xffff) {
|
|
return c;
|
|
} else {
|
|
return U16_LEAD(c);
|
|
}
|
|
} else {
|
|
return U_SENTINEL;
|
|
}
|
|
}
|
|
|
|
static UChar32 U_CALLCONV
|
|
lenient8IteratorNext(UCharIterator *iter) {
|
|
int32_t index;
|
|
|
|
if(iter->reservedField!=0) {
|
|
UChar trail=U16_TRAIL(iter->reservedField);
|
|
iter->reservedField=0;
|
|
if((index=iter->index)>=0) {
|
|
iter->index=index+1;
|
|
}
|
|
return trail;
|
|
} else if(iter->start<iter->limit) {
|
|
const uint8_t *s=(const uint8_t *)iter->context;
|
|
UChar32 c;
|
|
|
|
L8_NEXT(s, iter->start, iter->limit, c);
|
|
if((index=iter->index)>=0) {
|
|
iter->index=++index;
|
|
if(iter->length<0 && iter->start==iter->limit) {
|
|
iter->length= c<=0xffff ? index : index+1;
|
|
}
|
|
} else if(iter->start==iter->limit && iter->length>=0) {
|
|
iter->index= c<=0xffff ? iter->length : iter->length-1;
|
|
}
|
|
if(c<0) {
|
|
return 0xfffd;
|
|
} else if(c<=0xffff) {
|
|
return c;
|
|
} else {
|
|
iter->reservedField=c;
|
|
return U16_LEAD(c);
|
|
}
|
|
} else {
|
|
return U_SENTINEL;
|
|
}
|
|
}
|
|
|
|
static UChar32 U_CALLCONV
|
|
lenient8IteratorPrevious(UCharIterator *iter) {
|
|
int32_t index;
|
|
|
|
if(iter->reservedField!=0) {
|
|
UChar lead=U16_LEAD(iter->reservedField);
|
|
iter->reservedField=0;
|
|
iter->start-=4; /* we stayed behind the supplementary code point; go before it now */
|
|
if((index=iter->index)>0) {
|
|
iter->index=index-1;
|
|
}
|
|
return lead;
|
|
} else if(iter->start>0) {
|
|
const uint8_t *s=(const uint8_t *)iter->context;
|
|
UChar32 c;
|
|
|
|
L8_PREV(s, 0, iter->start, c);
|
|
if((index=iter->index)>0) {
|
|
iter->index=index-1;
|
|
} else if(iter->start<=1) {
|
|
iter->index= c<=0xffff ? iter->start : iter->start+1;
|
|
}
|
|
if(c<0) {
|
|
return 0xfffd;
|
|
} else if(c<=0xffff) {
|
|
return c;
|
|
} else {
|
|
iter->start+=4; /* back to behind this supplementary code point for consistent state */
|
|
iter->reservedField=c;
|
|
return U16_TRAIL(c);
|
|
}
|
|
} else {
|
|
return U_SENTINEL;
|
|
}
|
|
}
|
|
|
|
static uint32_t U_CALLCONV
|
|
lenient8IteratorGetState(const UCharIterator *iter) {
|
|
uint32_t state=(uint32_t)(iter->start<<1);
|
|
if(iter->reservedField!=0) {
|
|
state|=1;
|
|
}
|
|
return state;
|
|
}
|
|
|
|
static void U_CALLCONV
|
|
lenient8IteratorSetState(UCharIterator *iter, uint32_t state, UErrorCode *pErrorCode) {
|
|
if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
|
|
/* do nothing */
|
|
} else if(iter==NULL) {
|
|
*pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
|
|
} else if(state==lenient8IteratorGetState(iter)) {
|
|
/* setting to the current state: no-op */
|
|
} else {
|
|
int32_t index=(int32_t)(state>>1); /* UTF-8 index */
|
|
state&=1; /* 1 if in surrogate pair, must be index>=4 */
|
|
|
|
if((state==0 ? index<0 : index<4) || iter->limit<index) {
|
|
*pErrorCode=U_INDEX_OUTOFBOUNDS_ERROR;
|
|
} else {
|
|
iter->start=index; /* restore UTF-8 byte index */
|
|
if(index<=1) {
|
|
iter->index=index;
|
|
} else {
|
|
iter->index=-1; /* unknown UTF-16 index */
|
|
}
|
|
if(state==0) {
|
|
iter->reservedField=0;
|
|
} else {
|
|
/* verified index>=4 above */
|
|
UChar32 c;
|
|
L8_PREV((const uint8_t *)iter->context, 0, index, c);
|
|
if(c<=0xffff) {
|
|
*pErrorCode=U_INDEX_OUTOFBOUNDS_ERROR;
|
|
} else {
|
|
iter->reservedField=c;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const UCharIterator lenient8Iterator={
|
|
0, 0, 0, 0, 0, 0,
|
|
lenient8IteratorGetIndex,
|
|
lenient8IteratorMove,
|
|
lenient8IteratorHasNext,
|
|
lenient8IteratorHasPrevious,
|
|
lenient8IteratorCurrent,
|
|
lenient8IteratorNext,
|
|
lenient8IteratorPrevious,
|
|
NULL,
|
|
lenient8IteratorGetState,
|
|
lenient8IteratorSetState
|
|
};
|
|
|
|
U_CAPI void U_EXPORT2
|
|
uiter_setLenient8(UCharIterator *iter, const char *s, int32_t length) {
|
|
if(iter!=0) {
|
|
if(s!=0 && length>=-1) {
|
|
*iter=lenient8Iterator;
|
|
iter->context=s;
|
|
if(length>=0) {
|
|
iter->limit=length;
|
|
} else {
|
|
iter->limit=(int32_t)strlen(s);
|
|
}
|
|
iter->length= iter->limit<=1 ? iter->limit : -1;
|
|
} else {
|
|
/* set no-op iterator */
|
|
uiter_setString(iter, NULL, 0);
|
|
}
|
|
}
|
|
}
|