2008-12-22 12:02:22 +00:00
|
|
|
/* $Id$
|
|
|
|
|
|
|
|
Part of SWI-Prolog
|
|
|
|
|
|
|
|
Author: Jan Wielemaker and Anjo Anjewierden
|
|
|
|
E-mail: jan@swi.psy.uva.nl
|
|
|
|
WWW: http://www.swi-prolog.org
|
|
|
|
Copyright (C): 1985-2002, University of Amsterdam
|
|
|
|
|
|
|
|
This library is free software; you can redistribute it and/or
|
|
|
|
modify it under the terms of the GNU Lesser General Public
|
|
|
|
License as published by the Free Software Foundation; either
|
|
|
|
version 2.1 of the License, or (at your option) any later version.
|
|
|
|
|
|
|
|
This library 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
|
|
|
|
Lesser General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU Lesser General Public
|
|
|
|
License along with this library; if not, write to the Free Software
|
|
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "pl-incl.h"
|
|
|
|
#include "pl-ctype.h"
|
|
|
|
#include "pl-utf8.h"
|
|
|
|
#include <errno.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#ifdef __WINDOWS__
|
|
|
|
#include "pl-mswchar.h" /* Terrible hack */
|
|
|
|
#endif
|
2009-06-17 05:26:03 +01:00
|
|
|
#if HAVE_LIMITS_H
|
|
|
|
#include <limits.h> /* solaris compatibility */
|
|
|
|
#endif
|
2008-12-22 12:02:22 +00:00
|
|
|
|
|
|
|
#undef LD
|
|
|
|
#define LD LOCAL_LD
|
|
|
|
|
2008-12-22 13:49:44 +00:00
|
|
|
#ifdef __SWI_PROLOG__
|
2008-12-22 12:02:22 +00:00
|
|
|
static inline word
|
|
|
|
valHandle__LD(term_t r ARG_LD)
|
|
|
|
{ Word p = valTermRef(r);
|
|
|
|
|
|
|
|
deRef(p);
|
|
|
|
return *p;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define valHandle(r) valHandle__LD(r PASS_LD)
|
|
|
|
#define setHandle(h, w) (*valTermRef(h) = (w))
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
|
|
/*******************************
|
|
|
|
* UNIFIED TEXT STUFF *
|
|
|
|
*******************************/
|
|
|
|
|
|
|
|
static inline size_t
|
|
|
|
bufsize_text(PL_chars_t *text, size_t len)
|
|
|
|
{ size_t unit;
|
|
|
|
|
|
|
|
switch(text->encoding)
|
|
|
|
{ case ENC_ISO_LATIN_1:
|
|
|
|
case ENC_ASCII:
|
|
|
|
case ENC_UTF8:
|
|
|
|
case ENC_ANSI:
|
|
|
|
unit = sizeof(char);
|
|
|
|
break;
|
|
|
|
case ENC_WCHAR:
|
|
|
|
unit = sizeof(pl_wchar_t);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
unit = sizeof(char); /*NOTREACHED*/
|
|
|
|
}
|
|
|
|
|
|
|
|
return len*unit;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
PL_save_text(PL_chars_t *text, int flags)
|
|
|
|
{ if ( (flags & BUF_MALLOC) && text->storage != PL_CHARS_MALLOC )
|
|
|
|
{ size_t bl = bufsize_text(text, text->length+1);
|
|
|
|
void *new = PL_malloc(bl);
|
|
|
|
|
|
|
|
memcpy(new, text->text.t, bl);
|
|
|
|
text->text.t = new;
|
|
|
|
text->storage = PL_CHARS_MALLOC;
|
|
|
|
} else if ( text->storage == PL_CHARS_LOCAL )
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
|
|
|
size_t bl = bufsize_text(text, text->length+1);
|
|
|
|
|
|
|
|
addMultipleBuffer(b, text->text.t, bl, char);
|
|
|
|
text->text.t = baseBuffer(b, char);
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2010-02-22 09:35:47 +00:00
|
|
|
/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
|
|
|
PL_from_stack_text() moves a string from the stack, so it won't get
|
|
|
|
corrupted if GC/shift comes along.
|
|
|
|
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
|
|
|
|
|
|
|
|
static void
|
|
|
|
PL_from_stack_text(PL_chars_t *text)
|
|
|
|
{ if ( text->storage == PL_CHARS_STACK )
|
|
|
|
{ size_t bl = bufsize_text(text, text->length+1);
|
|
|
|
|
|
|
|
if ( bl < sizeof(text->buf) )
|
|
|
|
{ memcpy(text->buf, text->text.t, bl);
|
|
|
|
text->text.t = text->buf;
|
|
|
|
text->storage = PL_CHARS_LOCAL;
|
|
|
|
} else
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
|
|
|
|
|
|
|
addMultipleBuffer(b, text->text.t, bl, char);
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
int
|
|
|
|
PL_get_text__LD(term_t l, PL_chars_t *text, int flags ARG_LD)
|
2010-02-22 09:35:47 +00:00
|
|
|
{ word w = valHandle(l);
|
2008-12-22 12:02:22 +00:00
|
|
|
|
|
|
|
if ( (flags & CVT_ATOM) && isAtom(w) )
|
|
|
|
{ if ( !get_atom_text(w, text) )
|
|
|
|
goto maybe_write;
|
|
|
|
} else if ( (flags & CVT_STRING) && isString(w) )
|
|
|
|
{ if ( !get_string_text(w, text PASS_LD) )
|
|
|
|
goto maybe_write;
|
2010-02-22 09:35:47 +00:00
|
|
|
PL_from_stack_text(text);
|
2008-12-22 12:02:22 +00:00
|
|
|
} else if ( (flags & CVT_INTEGER) && isInteger(w) )
|
|
|
|
{ number n;
|
|
|
|
|
|
|
|
PL_get_number(l, &n);
|
|
|
|
switch(n.type)
|
|
|
|
{ case V_INTEGER:
|
|
|
|
sprintf(text->buf, INT64_FORMAT, n.value.i);
|
|
|
|
text->text.t = text->buf;
|
|
|
|
text->length = strlen(text->text.t);
|
|
|
|
text->storage = PL_CHARS_LOCAL;
|
|
|
|
break;
|
|
|
|
#ifdef O_GMP
|
|
|
|
case V_MPZ:
|
|
|
|
{ size_t sz = mpz_sizeinbase(n.value.mpz, 10) + 2;
|
|
|
|
Buffer b = findBuffer(BUF_RING);
|
|
|
|
|
|
|
|
growBuffer(b, sz);
|
|
|
|
mpz_get_str(b->base, 10, n.value.mpz);
|
|
|
|
b->top = b->base + strlen(b->base);
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->length = entriesBuffer(b, char);
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
}
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
text->canonical = TRUE;
|
2010-02-22 09:35:47 +00:00
|
|
|
} else if ( (flags & CVT_FLOAT) && isFloat(w) )
|
|
|
|
{ format_float(valFloat(w), text->buf, LD->float_format);
|
2008-12-22 12:02:22 +00:00
|
|
|
text->text.t = text->buf;
|
|
|
|
text->length = strlen(text->text.t);
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
text->storage = PL_CHARS_LOCAL;
|
|
|
|
text->canonical = TRUE;
|
|
|
|
} else if ( (flags & CVT_LIST) &&
|
|
|
|
(isList(w) || isNil(w)) )
|
|
|
|
{ Buffer b;
|
|
|
|
|
|
|
|
if ( (b = codes_or_chars_to_buffer(l, BUF_RING, FALSE)) )
|
|
|
|
{ text->length = entriesBuffer(b, char);
|
|
|
|
addBuffer(b, EOS, char);
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
} else if ( (b = codes_or_chars_to_buffer(l, BUF_RING, TRUE)) )
|
|
|
|
{ text->length = entriesBuffer(b, pl_wchar_t);
|
|
|
|
addBuffer(b, EOS, pl_wchar_t);
|
|
|
|
text->text.w = baseBuffer(b, pl_wchar_t);
|
|
|
|
text->encoding = ENC_WCHAR;
|
|
|
|
} else
|
|
|
|
goto maybe_write;
|
|
|
|
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
text->canonical = TRUE;
|
|
|
|
} else if ( (flags & CVT_VARIABLE) && isVar(w) )
|
|
|
|
{ text->text.t = varName(l, text->buf);
|
|
|
|
text->length = strlen(text->text.t);
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
text->storage = PL_CHARS_LOCAL;
|
|
|
|
text->canonical = TRUE;
|
|
|
|
} else if ( (flags & CVT_WRITE) )
|
|
|
|
{ IOENC encodings[3];
|
|
|
|
IOENC *enc;
|
|
|
|
char *r;
|
|
|
|
|
|
|
|
case_write:
|
|
|
|
encodings[0] = ENC_ISO_LATIN_1;
|
|
|
|
encodings[1] = ENC_WCHAR;
|
|
|
|
encodings[2] = ENC_UNKNOWN;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
for(enc = encodings; *enc != ENC_UNKNOWN; enc++)
|
|
|
|
{ size_t size;
|
|
|
|
IOSTREAM *fd;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
r = text->buf;
|
|
|
|
size = sizeof(text->buf);
|
|
|
|
fd = Sopenmem(&r, &size, "w");
|
|
|
|
fd->encoding = *enc;
|
|
|
|
if ( PL_write_term(fd, l, 1200, 0) &&
|
|
|
|
Sputcode(EOS, fd) >= 0 &&
|
|
|
|
Sflush(fd) >= 0 )
|
|
|
|
{ text->encoding = *enc;
|
|
|
|
text->storage = (r == text->buf ? PL_CHARS_LOCAL : PL_CHARS_MALLOC);
|
|
|
|
text->canonical = TRUE;
|
|
|
|
|
|
|
|
if ( *enc == ENC_ISO_LATIN_1 )
|
|
|
|
{ text->length = size-1;
|
|
|
|
text->text.t = r;
|
|
|
|
} else
|
|
|
|
{ text->length = (size/sizeof(pl_wchar_t))-1;
|
|
|
|
text->text.w = (pl_wchar_t *)r;
|
|
|
|
}
|
|
|
|
|
|
|
|
Sclose(fd);
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
} else
|
|
|
|
{ Sclose(fd);
|
|
|
|
if ( r != text->buf )
|
|
|
|
Sfree(r);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
goto error;
|
|
|
|
} else
|
|
|
|
{ goto error;
|
|
|
|
}
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
|
|
|
|
maybe_write:
|
|
|
|
if ( (flags & CVT_WRITE) )
|
|
|
|
goto case_write;
|
|
|
|
|
|
|
|
error:
|
2010-02-22 09:35:47 +00:00
|
|
|
if ( canBind(w) && (flags & CVT_VARNOFAIL) )
|
|
|
|
return 2;
|
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
if ( (flags & CVT_EXCEPTION) )
|
|
|
|
{ atom_t expected;
|
|
|
|
|
|
|
|
if ( flags & CVT_LIST )
|
|
|
|
expected = ATOM_text;
|
|
|
|
else if ( flags & CVT_NUMBER )
|
|
|
|
expected = ATOM_atomic;
|
|
|
|
else
|
|
|
|
expected = ATOM_atom;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
return PL_error(NULL, 0, NULL, ERR_TYPE, expected, l);
|
|
|
|
}
|
|
|
|
|
|
|
|
fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
atom_t
|
|
|
|
textToAtom(PL_chars_t *text)
|
|
|
|
{ PL_canonise_text(text);
|
|
|
|
|
|
|
|
if ( text->encoding == ENC_ISO_LATIN_1 )
|
|
|
|
{ return lookupAtom(text->text.t, text->length);
|
|
|
|
} else
|
|
|
|
{ return lookupUCSAtom(text->text.w, text->length);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
word
|
|
|
|
textToString(PL_chars_t *text)
|
|
|
|
{ PL_canonise_text(text);
|
|
|
|
|
|
|
|
if ( text->encoding == ENC_ISO_LATIN_1 )
|
|
|
|
{ return globalString(text->length, text->text.t);
|
|
|
|
} else
|
|
|
|
{ return globalWString(text->length, text->text.w);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_unify_text(term_t term, term_t tail, PL_chars_t *text, int type)
|
|
|
|
{ switch(type)
|
|
|
|
{ case PL_ATOM:
|
|
|
|
{ atom_t a = textToAtom(text);
|
|
|
|
int rval = _PL_unify_atomic(term, a);
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
PL_unregister_atom(a);
|
|
|
|
return rval;
|
|
|
|
}
|
|
|
|
case PL_STRING:
|
|
|
|
{ word w = textToString(text);
|
|
|
|
|
2010-02-22 09:35:47 +00:00
|
|
|
if ( w )
|
|
|
|
return _PL_unify_atomic(term, w);
|
|
|
|
else
|
|
|
|
return FALSE;
|
2008-12-22 12:02:22 +00:00
|
|
|
}
|
|
|
|
case PL_CODE_LIST:
|
|
|
|
case PL_CHAR_LIST:
|
|
|
|
{ if ( text->length == 0 )
|
|
|
|
{ if ( tail )
|
|
|
|
{ GET_LD
|
|
|
|
PL_put_term(tail, term);
|
|
|
|
return TRUE;
|
|
|
|
} else
|
|
|
|
{ return PL_unify_nil(term);
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
{ GET_LD
|
2010-02-22 09:35:47 +00:00
|
|
|
term_t l = PL_new_term_ref();
|
|
|
|
Word p0, p;
|
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
switch(text->encoding)
|
|
|
|
{ case ENC_ISO_LATIN_1:
|
|
|
|
{ const unsigned char *s = (const unsigned char *)text->text.t;
|
|
|
|
const unsigned char *e = &s[text->length];
|
|
|
|
|
2010-02-22 09:35:47 +00:00
|
|
|
if ( !(p0 = p = INIT_SEQ_CODES(text->length)) )
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
if ( type == PL_CODE_LIST ) {
|
|
|
|
for( ; s < e; s++)
|
|
|
|
p = EXTEND_SEQ_CODES(p, *s);
|
|
|
|
} else {
|
|
|
|
for( ; s < e; s++)
|
|
|
|
p = EXTEND_SEQ_ATOMS(p, *s);
|
|
|
|
}
|
2008-12-22 12:02:22 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
case ENC_WCHAR:
|
|
|
|
{ const pl_wchar_t *s = (const pl_wchar_t *)text->text.t;
|
|
|
|
const pl_wchar_t *e = &s[text->length];
|
2010-02-22 09:35:47 +00:00
|
|
|
|
|
|
|
if ( !(p0 = p = INIT_SEQ_CODES(text->length)) )
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
if ( type == PL_CODE_LIST ) {
|
|
|
|
for( ; s < e; s++)
|
|
|
|
p = EXTEND_SEQ_CODES(p, *s);
|
|
|
|
} else {
|
|
|
|
for( ; s < e; s++)
|
|
|
|
p = EXTEND_SEQ_ATOMS(p, *s);
|
|
|
|
}
|
2008-12-22 12:02:22 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
case ENC_UTF8:
|
|
|
|
{ const char *s = text->text.t;
|
|
|
|
const char *e = &s[text->length];
|
|
|
|
size_t len = utf8_strlen(s, text->length);
|
|
|
|
|
2010-02-22 09:35:47 +00:00
|
|
|
if ( !(p0 = p = INIT_SEQ_CODES(len)) )
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
if ( type == PL_CODE_LIST ) {
|
|
|
|
while (s < e) {
|
|
|
|
int chr;
|
|
|
|
|
|
|
|
s = utf8_get_char(s, &chr);
|
|
|
|
p = EXTEND_SEQ_CODES(p, chr);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
while (s < e) {
|
|
|
|
int chr;
|
|
|
|
|
|
|
|
s = utf8_get_char(s, &chr);
|
|
|
|
p = EXTEND_SEQ_ATOMS(p, chr);
|
|
|
|
}
|
|
|
|
}
|
2008-12-22 12:02:22 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
case ENC_ANSI:
|
|
|
|
{ const char *s = text->text.t;
|
|
|
|
size_t rc, n = text->length;
|
|
|
|
size_t len = 0;
|
|
|
|
mbstate_t mbs;
|
|
|
|
wchar_t wc;
|
|
|
|
|
|
|
|
memset(&mbs, 0, sizeof(mbs));
|
|
|
|
while( n > 0 && (rc=mbrtowc(&wc, s, n, &mbs)) != (size_t)-1 )
|
|
|
|
{ len++;
|
|
|
|
n -= rc;
|
|
|
|
s += rc;
|
|
|
|
}
|
2010-02-22 09:35:47 +00:00
|
|
|
|
|
|
|
if ( !(p0 = p = INIT_SEQ_CODES(len)) )
|
|
|
|
return FALSE;
|
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
memset(&mbs, 0, sizeof(mbs));
|
|
|
|
n = text->length;
|
|
|
|
|
2010-02-22 09:35:47 +00:00
|
|
|
while(n > 0)
|
|
|
|
{ rc = mbrtowc(&wc, s, n, &mbs);
|
2008-12-22 12:02:22 +00:00
|
|
|
|
|
|
|
if ( type == PL_CODE_LIST )
|
2008-12-22 13:49:44 +00:00
|
|
|
p = EXTEND_SEQ_CODES(p, wc);
|
2008-12-22 12:02:22 +00:00
|
|
|
else
|
2008-12-22 13:49:44 +00:00
|
|
|
p = EXTEND_SEQ_ATOMS(p, wc);
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
s += rc;
|
|
|
|
n -= rc;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
{ assert(0);
|
|
|
|
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-02-22 09:35:47 +00:00
|
|
|
return CLOSE_SEQ_OF_CODES(p, p0, tail, term, l );
|
2008-12-22 12:02:22 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
{ assert(0);
|
|
|
|
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_unify_text_range(term_t term, PL_chars_t *text,
|
|
|
|
size_t offset, size_t len, int type)
|
|
|
|
{ if ( offset == 0 && len == text->length )
|
|
|
|
{ return PL_unify_text(term, 0, text, type);
|
|
|
|
} else
|
|
|
|
{ PL_chars_t sub;
|
|
|
|
int rc;
|
|
|
|
|
|
|
|
if ( offset > text->length || offset + len > text->length )
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
sub.length = len;
|
|
|
|
sub.storage = PL_CHARS_HEAP;
|
|
|
|
if ( text->encoding == ENC_ISO_LATIN_1 )
|
|
|
|
{ sub.text.t = text->text.t+offset;
|
|
|
|
sub.encoding = ENC_ISO_LATIN_1;
|
|
|
|
sub.canonical = TRUE;
|
|
|
|
} else
|
|
|
|
{ sub.text.w = text->text.w+offset;
|
|
|
|
sub.encoding = ENC_WCHAR;
|
|
|
|
sub.canonical = FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
rc = PL_unify_text(term, 0, &sub, type);
|
|
|
|
|
|
|
|
PL_free_text(&sub);
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
|
|
|
int PL_promote_text(PL_chars_t *text)
|
|
|
|
|
|
|
|
Promote a text to USC if it is currently 8-bit text.
|
|
|
|
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_promote_text(PL_chars_t *text)
|
|
|
|
{ if ( text->encoding != ENC_WCHAR )
|
|
|
|
{ if ( text->storage == PL_CHARS_MALLOC )
|
|
|
|
{ pl_wchar_t *new = PL_malloc(sizeof(pl_wchar_t)*(text->length+1));
|
|
|
|
pl_wchar_t *t = new;
|
|
|
|
const unsigned char *s = (const unsigned char *)text->text.t;
|
|
|
|
const unsigned char *e = &s[text->length];
|
|
|
|
|
|
|
|
while(s<e)
|
|
|
|
{ *t++ = *s++;
|
|
|
|
}
|
|
|
|
*t = EOS;
|
|
|
|
|
|
|
|
PL_free(text->text.t);
|
|
|
|
text->text.w = new;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
text->encoding = ENC_WCHAR;
|
|
|
|
} else if ( text->storage == PL_CHARS_LOCAL &&
|
|
|
|
(text->length+1)*sizeof(pl_wchar_t) < sizeof(text->buf) )
|
|
|
|
{ unsigned char buf[sizeof(text->buf)];
|
|
|
|
unsigned char *f = buf;
|
|
|
|
unsigned char *e = &buf[text->length];
|
|
|
|
pl_wchar_t *t = (pl_wchar_t*)text->buf;
|
|
|
|
|
|
|
|
memcpy(buf, text->buf, text->length*sizeof(char));
|
|
|
|
while(f<e)
|
|
|
|
{ *t++ = *f++;
|
|
|
|
}
|
|
|
|
*t = EOS;
|
|
|
|
text->encoding = ENC_WCHAR;
|
|
|
|
} else
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
|
|
|
const unsigned char *s = (const unsigned char *)text->text.t;
|
|
|
|
const unsigned char *e = &s[text->length];
|
|
|
|
|
|
|
|
for( ; s<e; s++)
|
|
|
|
addBuffer(b, *s, pl_wchar_t);
|
|
|
|
addBuffer(b, EOS, pl_wchar_t);
|
|
|
|
|
|
|
|
text->text.w = baseBuffer(b, pl_wchar_t);
|
|
|
|
text->encoding = ENC_WCHAR;
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_demote_text(PL_chars_t *text)
|
|
|
|
{ if ( text->encoding != ENC_ISO_LATIN_1 )
|
|
|
|
{ if ( text->storage == PL_CHARS_MALLOC )
|
|
|
|
{ char *new = PL_malloc(sizeof(char)*(text->length+1));
|
|
|
|
char *t = new;
|
|
|
|
const pl_wchar_t *s = (const pl_wchar_t *)text->text.t;
|
|
|
|
const pl_wchar_t *e = &s[text->length];
|
|
|
|
|
|
|
|
while(s<e)
|
|
|
|
{ if ( *s > 0xff )
|
|
|
|
{ PL_free(new);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
*t++ = *s++ & 0xff;
|
|
|
|
}
|
|
|
|
*t = EOS;
|
|
|
|
|
|
|
|
PL_free(text->text.t);
|
|
|
|
text->text.t = new;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
} else if ( text->storage == PL_CHARS_LOCAL )
|
|
|
|
{ pl_wchar_t buf[sizeof(text->buf)/sizeof(pl_wchar_t)];
|
|
|
|
pl_wchar_t *f = buf;
|
|
|
|
pl_wchar_t *e = &buf[text->length];
|
|
|
|
char *t = text->buf;
|
|
|
|
|
|
|
|
memcpy(buf, text->buf, text->length*sizeof(pl_wchar_t));
|
|
|
|
while(f<e)
|
|
|
|
{ if ( *f > 0xff )
|
|
|
|
return FALSE;
|
|
|
|
*t++ = *f++ & 0xff;
|
|
|
|
}
|
|
|
|
*t = EOS;
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
} else
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
|
|
|
const pl_wchar_t *s = (const pl_wchar_t*)text->text.w;
|
|
|
|
const pl_wchar_t *e = &s[text->length];
|
|
|
|
|
|
|
|
for( ; s<e; s++)
|
|
|
|
{ if ( *s > 0xff )
|
|
|
|
{ unfindBuffer(BUF_RING);
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
addBuffer(b, *s&0xff, char);
|
|
|
|
}
|
|
|
|
addBuffer(b, EOS, char);
|
|
|
|
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
can_demote(PL_chars_t *text)
|
|
|
|
{ if ( text->encoding != ENC_ISO_LATIN_1 )
|
|
|
|
{ const pl_wchar_t *w = (const pl_wchar_t*)text->text.w;
|
|
|
|
const pl_wchar_t *e = &w[text->length];
|
|
|
|
|
|
|
|
for(; w<e; w++)
|
|
|
|
{ if ( *w > 0xff )
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
|
|
|
Convert text to 8-bit according to flags. May hold REP_UTF8 to convert
|
|
|
|
to UTF-8, REP_MB to convert to locale 8-bit representation or nothing to
|
|
|
|
convert to ISO Latin-1. This predicate can fail of the text cannot be
|
|
|
|
represented.
|
|
|
|
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
|
|
|
|
|
|
|
|
static int
|
|
|
|
wctobuffer(wchar_t c, mbstate_t *mbs, Buffer buf)
|
|
|
|
{ char b[MB_LEN_MAX];
|
|
|
|
size_t n;
|
|
|
|
|
|
|
|
if ( (n=wcrtomb(b, c, mbs)) != (size_t)-1 )
|
|
|
|
{ size_t i;
|
|
|
|
|
|
|
|
for(i=0; i<n; i++)
|
|
|
|
addBuffer(buf, b[i], char);
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return FALSE; /* cannot represent */
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
utf8tobuffer(wchar_t c, Buffer buf)
|
|
|
|
{ if ( c <= 0x7f )
|
|
|
|
{ addBuffer(buf, (char)c, char);
|
|
|
|
} else
|
|
|
|
{ char b[6];
|
|
|
|
char *e = b;
|
|
|
|
const char *s;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
e = utf8_put_char(e, c);
|
|
|
|
for(s=b; s<e; s++)
|
|
|
|
addBuffer(buf, *s, char);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_mb_text(PL_chars_t *text, int flags)
|
|
|
|
{ int norep = -1;
|
|
|
|
IOENC target = ((flags&REP_UTF8) ? ENC_UTF8 :
|
|
|
|
(flags&REP_MB) ? ENC_ANSI : ENC_ISO_LATIN_1);
|
|
|
|
|
|
|
|
if ( text->encoding != target )
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
switch(text->encoding)
|
|
|
|
{ case ENC_ISO_LATIN_1:
|
|
|
|
{ const unsigned char *s = (const unsigned char*)text->text.t;
|
|
|
|
const unsigned char *e = &s[text->length];
|
|
|
|
|
|
|
|
if ( target == ENC_UTF8 )
|
|
|
|
{ for( ; s<e; s++)
|
|
|
|
{ utf8tobuffer(*s, b);
|
|
|
|
}
|
|
|
|
addBuffer(b, 0, char);
|
|
|
|
} else /* if ( target == ENC_MB ) */
|
|
|
|
{ mbstate_t mbs;
|
|
|
|
|
|
|
|
memset(&mbs, 0, sizeof(mbs));
|
|
|
|
for( ; s<e; s++)
|
|
|
|
{ if ( !wctobuffer(*s, &mbs, b) )
|
|
|
|
{ unfindBuffer(BUF_RING);
|
|
|
|
norep = *s;
|
|
|
|
goto rep_error;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
wctobuffer(0, &mbs, b);
|
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case ENC_WCHAR:
|
|
|
|
{ if ( target == ENC_ISO_LATIN_1 )
|
|
|
|
{ return PL_demote_text(text);
|
|
|
|
} else
|
|
|
|
{ const pl_wchar_t *w = (const pl_wchar_t*)text->text.w;
|
|
|
|
const pl_wchar_t *e = &w[text->length];
|
|
|
|
|
|
|
|
if ( target == ENC_UTF8 )
|
|
|
|
{ for( ; w<e; w++)
|
|
|
|
{ utf8tobuffer(*w, b);
|
|
|
|
}
|
|
|
|
addBuffer(b, 0, char);
|
|
|
|
} else /* if ( target == ENC_MB ) */
|
|
|
|
{ mbstate_t mbs;
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
memset(&mbs, 0, sizeof(mbs));
|
|
|
|
for( ; w<e; w++)
|
|
|
|
{ if ( !wctobuffer(*w, &mbs, b) )
|
|
|
|
{ unfindBuffer(BUF_RING);
|
|
|
|
norep = *w;
|
|
|
|
goto rep_error;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
wctobuffer(0, &mbs, b);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
{ assert(0);
|
|
|
|
fail;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
text->length = sizeOfBuffer(b)-1;
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->encoding = target;
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
}
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
|
|
|
|
rep_error:
|
|
|
|
if ( (flags & CVT_EXCEPTION) )
|
|
|
|
{ char msg[128];
|
|
|
|
|
|
|
|
sprintf(msg,
|
|
|
|
"Cannot represent char U%04x using %s encoding",
|
2010-02-22 09:35:47 +00:00
|
|
|
norep,
|
2008-12-22 12:02:22 +00:00
|
|
|
target == ENC_ISO_LATIN_1 ? "ISO Latin-1" : "current locale");
|
|
|
|
|
|
|
|
return PL_error(NULL, 0, msg, ERR_REPRESENTATION, ATOM_encoding);
|
|
|
|
}
|
|
|
|
|
|
|
|
fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_canonise_text(PL_chars_t *text)
|
|
|
|
{ if ( !text->canonical )
|
|
|
|
{ switch(text->encoding )
|
|
|
|
{ case ENC_ISO_LATIN_1:
|
|
|
|
break; /* nothing to do */
|
|
|
|
case ENC_WCHAR:
|
|
|
|
{ const pl_wchar_t *w = (const pl_wchar_t*)text->text.w;
|
|
|
|
const pl_wchar_t *e = &w[text->length];
|
2010-02-22 09:35:47 +00:00
|
|
|
|
2008-12-22 12:02:22 +00:00
|
|
|
for(; w<e; w++)
|
|
|
|
{ if ( *w > 0xff )
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return PL_demote_text(text);
|
|
|
|
}
|
|
|
|
case ENC_UTF8:
|
|
|
|
{ const char *s = text->text.t;
|
|
|
|
const char *e = &s[text->length];
|
|
|
|
|
|
|
|
while(s<e && !(*s & 0x80))
|
|
|
|
s++;
|
|
|
|
if ( s == e )
|
|
|
|
{ text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
text->canonical = TRUE;
|
|
|
|
} else
|
|
|
|
{ int chr;
|
|
|
|
int wide = FALSE;
|
|
|
|
size_t len = s - text->text.t;
|
|
|
|
|
|
|
|
while(s<e)
|
|
|
|
{ s = utf8_get_char(s, &chr);
|
|
|
|
if ( chr > 0xff ) /* requires wide characters */
|
|
|
|
wide = TRUE;
|
|
|
|
len++;
|
|
|
|
}
|
|
|
|
|
|
|
|
s = (const char *)text->text.t;
|
|
|
|
text->length = len;
|
|
|
|
|
|
|
|
if ( wide )
|
|
|
|
{ pl_wchar_t *to = PL_malloc(sizeof(pl_wchar_t)*(len+1));
|
|
|
|
|
|
|
|
text->text.w = to;
|
|
|
|
while(s<e)
|
|
|
|
{ s = utf8_get_char(s, &chr);
|
|
|
|
*to++ = chr;
|
|
|
|
}
|
|
|
|
*to = EOS;
|
|
|
|
|
|
|
|
text->encoding = ENC_WCHAR;
|
|
|
|
text->storage = PL_CHARS_MALLOC;
|
|
|
|
} else
|
|
|
|
{ char *to = PL_malloc(len+1);
|
|
|
|
|
|
|
|
text->text.t = to;
|
|
|
|
while(s<e)
|
|
|
|
{ s = utf8_get_char(s, &chr);
|
|
|
|
*to++ = chr;
|
|
|
|
}
|
|
|
|
*to = EOS;
|
|
|
|
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
text->storage = PL_CHARS_MALLOC;
|
|
|
|
}
|
|
|
|
|
|
|
|
text->canonical = TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
}
|
|
|
|
case ENC_ANSI:
|
|
|
|
{ mbstate_t mbs;
|
|
|
|
size_t len = 0;
|
|
|
|
int iso = TRUE;
|
|
|
|
char *s = text->text.t;
|
|
|
|
size_t rc, n = text->length;
|
|
|
|
wchar_t wc;
|
|
|
|
|
|
|
|
memset(&mbs, 0, sizeof(mbs));
|
|
|
|
while( n > 0 && (rc=mbrtowc(&wc, s, n, &mbs)) != (size_t)-1 )
|
|
|
|
{ if ( wc > 0xff )
|
|
|
|
iso = FALSE;
|
|
|
|
len++;
|
|
|
|
n -= rc;
|
|
|
|
s += rc;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ( n == 0 )
|
|
|
|
{ const char *from = text->text.t;
|
|
|
|
void *do_free;
|
|
|
|
|
|
|
|
n = text->length;
|
|
|
|
memset(&mbs, 0, sizeof(mbs));
|
|
|
|
|
|
|
|
if ( text->storage == PL_CHARS_MALLOC )
|
|
|
|
do_free = text->text.t;
|
|
|
|
else
|
|
|
|
do_free = NULL;
|
|
|
|
|
|
|
|
if ( iso )
|
|
|
|
{ char *to;
|
|
|
|
|
|
|
|
text->encoding = ENC_ISO_LATIN_1;
|
|
|
|
if ( len+1 < sizeof(text->buf) )
|
|
|
|
{ text->text.t = text->buf;
|
|
|
|
text->storage = PL_CHARS_LOCAL;
|
|
|
|
} else
|
|
|
|
{ text->text.t = PL_malloc(len+1);
|
|
|
|
text->storage = PL_CHARS_MALLOC;
|
|
|
|
}
|
|
|
|
|
|
|
|
to = text->text.t;
|
|
|
|
while( n > 0 && (rc=mbrtowc(&wc, from, n, &mbs)) != (size_t)-1 )
|
|
|
|
{ *to++ = (char)wc;
|
|
|
|
n -= rc;
|
|
|
|
from += rc;
|
|
|
|
}
|
|
|
|
*to = EOS;
|
|
|
|
} else
|
|
|
|
{ wchar_t *to;
|
|
|
|
char b2[sizeof(text->buf)];
|
|
|
|
|
|
|
|
text->encoding = ENC_WCHAR;
|
|
|
|
if ( len+1 < sizeof(text->buf)/sizeof(wchar_t) )
|
|
|
|
{ if ( text->text.t == text->buf )
|
|
|
|
{ memcpy(b2, text->buf, sizeof(text->buf));
|
|
|
|
from = b2;
|
|
|
|
}
|
|
|
|
text->text.w = (wchar_t*)text->buf;
|
|
|
|
} else
|
|
|
|
{ text->text.w = PL_malloc((len+1)*sizeof(wchar_t));
|
|
|
|
text->storage = PL_CHARS_MALLOC;
|
|
|
|
}
|
|
|
|
|
|
|
|
to = text->text.w;
|
|
|
|
while( n > 0 && (rc=mbrtowc(&wc, from, n, &mbs)) != (size_t)-1 )
|
|
|
|
{ *to++ = wc;
|
|
|
|
n -= rc;
|
|
|
|
from += rc;
|
|
|
|
}
|
|
|
|
*to = EOS;
|
|
|
|
}
|
|
|
|
|
|
|
|
text->length = len;
|
|
|
|
text->canonical = TRUE;
|
|
|
|
if ( do_free )
|
|
|
|
PL_free(do_free);
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
}
|
|
|
|
|
|
|
|
fail;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
succeed;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
PL_free_text(PL_chars_t *text)
|
|
|
|
{ if ( text->storage == PL_CHARS_MALLOC )
|
|
|
|
PL_free(text->text.t);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
PL_text_recode(PL_chars_t *text, IOENC encoding)
|
|
|
|
{ if ( text->encoding != encoding )
|
|
|
|
{ switch(encoding)
|
|
|
|
{ case ENC_UTF8:
|
|
|
|
{ switch(text->encoding)
|
|
|
|
{ case ENC_ASCII:
|
|
|
|
text->encoding = ENC_UTF8;
|
|
|
|
break;
|
|
|
|
case ENC_ISO_LATIN_1:
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
|
|
|
const unsigned char *s = (const unsigned char *)text->text.t;
|
|
|
|
const unsigned char *e = &s[text->length];
|
|
|
|
char tmp[8];
|
|
|
|
|
|
|
|
for( ; s<e; s++)
|
|
|
|
{ if ( *s&0x80 )
|
|
|
|
{ const char *end = utf8_put_char(tmp, *s);
|
|
|
|
const char *q = tmp;
|
|
|
|
|
|
|
|
for(q=tmp; q<end; q++)
|
|
|
|
addBuffer(b, *q, char);
|
|
|
|
} else
|
|
|
|
{ addBuffer(b, *s, char);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
PL_free_text(text);
|
|
|
|
text->length = entriesBuffer(b, char);
|
|
|
|
addBuffer(b, EOS, char);
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->encoding = ENC_UTF8;
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case ENC_WCHAR:
|
|
|
|
{ Buffer b = findBuffer(BUF_RING);
|
|
|
|
const pl_wchar_t *s = text->text.w;
|
|
|
|
const pl_wchar_t *e = &s[text->length];
|
|
|
|
char tmp[8];
|
|
|
|
|
|
|
|
for( ; s<e; s++)
|
|
|
|
{ if ( *s > 0x7f )
|
|
|
|
{ const char *end = utf8_put_char(tmp, (int)*s);
|
|
|
|
const char *q = tmp;
|
|
|
|
|
|
|
|
for(q=tmp; q<end; q++)
|
|
|
|
addBuffer(b, *q&0xff, char);
|
|
|
|
} else
|
|
|
|
{ addBuffer(b, *s&0xff, char);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
PL_free_text(text);
|
|
|
|
text->length = entriesBuffer(b, char);
|
|
|
|
addBuffer(b, EOS, char);
|
|
|
|
text->text.t = baseBuffer(b, char);
|
|
|
|
text->encoding = ENC_UTF8;
|
|
|
|
text->storage = PL_CHARS_RING;
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
assert(0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
|
|
|
|
PL_cmp_text(PL_chars_t *t1, size_t o1,
|
|
|
|
PL_chars_t *t2, size_t o2,
|
|
|
|
size_t len)
|
|
|
|
|
|
|
|
Compares two substrings of two text representations.
|
|
|
|
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_cmp_text(PL_chars_t *t1, size_t o1, PL_chars_t *t2, size_t o2,
|
|
|
|
size_t len)
|
|
|
|
{ ssize_t l = len;
|
|
|
|
int ifeq = 0;
|
|
|
|
|
|
|
|
if ( l > (ssize_t)(t1->length - o1) )
|
|
|
|
{ l = t1->length - o1;
|
|
|
|
ifeq = -1; /* first is short */
|
|
|
|
}
|
|
|
|
if ( l > (ssize_t)(t2->length - o2) )
|
|
|
|
{ l = t2->length - o2;
|
|
|
|
if ( ifeq == 0 )
|
|
|
|
ifeq = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if ( l == 0 ) /* too long offsets */
|
|
|
|
return ifeq;
|
|
|
|
|
|
|
|
if ( t1->encoding == ENC_ISO_LATIN_1 && t2->encoding == ENC_ISO_LATIN_1 )
|
|
|
|
{ const unsigned char *s = (const unsigned char *)t1->text.t+o1;
|
|
|
|
const unsigned char *q = (const unsigned char *)t2->text.t+o2;
|
|
|
|
|
|
|
|
for(; l-- > 0 && *s == *q; s++, q++ )
|
|
|
|
;
|
|
|
|
if ( l < 0 )
|
|
|
|
return ifeq;
|
|
|
|
else
|
|
|
|
return *s > *q ? 1 : -1;
|
|
|
|
} else if ( t1->encoding == ENC_WCHAR && t2->encoding == ENC_WCHAR )
|
|
|
|
{ const pl_wchar_t *s = t1->text.w+o1;
|
|
|
|
const pl_wchar_t *q = t2->text.w+o2;
|
|
|
|
|
|
|
|
for(; l-- > 0 && *s == *q; s++, q++ )
|
|
|
|
;
|
|
|
|
if ( l < 0 )
|
|
|
|
return ifeq;
|
|
|
|
else
|
|
|
|
return *s > *q ? 1 : -1;
|
|
|
|
} else if ( t1->encoding == ENC_ISO_LATIN_1 && t2->encoding == ENC_WCHAR )
|
|
|
|
{ const unsigned char *s = (const unsigned char *)t1->text.t+o1;
|
|
|
|
const pl_wchar_t *q = t2->text.w+o2;
|
|
|
|
|
|
|
|
for(; l-- > 0 && *s == *q; s++, q++ )
|
|
|
|
;
|
|
|
|
if ( l < 0 )
|
|
|
|
return ifeq;
|
|
|
|
else
|
|
|
|
return *s > *q ? 1 : -1;
|
|
|
|
} else
|
|
|
|
{ const pl_wchar_t *s = t1->text.w+o1;
|
|
|
|
const unsigned char *q = (const unsigned char *)t2->text.t+o2;
|
|
|
|
|
|
|
|
for(; l-- > 0 && *s == *q; s++, q++ )
|
|
|
|
;
|
|
|
|
if ( l < 0 )
|
|
|
|
return ifeq;
|
|
|
|
else
|
|
|
|
return *s > *q ? 1 : -1;
|
2010-02-22 09:35:47 +00:00
|
|
|
}
|
2008-12-22 12:02:22 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
PL_concat_text(int n, PL_chars_t **text, PL_chars_t *result)
|
|
|
|
{ size_t total_length = 0;
|
|
|
|
int latin = TRUE;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for(i=0; i<n; i++)
|
|
|
|
{ if ( latin && !can_demote(text[i]) )
|
|
|
|
latin = FALSE;
|
|
|
|
total_length += text[i]->length;
|
|
|
|
}
|
|
|
|
|
|
|
|
result->canonical = TRUE;
|
|
|
|
result->length = total_length;
|
|
|
|
|
|
|
|
if ( latin )
|
|
|
|
{ char *to;
|
|
|
|
|
|
|
|
result->encoding = ENC_ISO_LATIN_1;
|
|
|
|
if ( total_length+1 < sizeof(result->buf) )
|
|
|
|
{ result->text.t = result->buf;
|
|
|
|
result->storage = PL_CHARS_LOCAL;
|
|
|
|
} else
|
|
|
|
{ result->text.t = PL_malloc(total_length+1);
|
|
|
|
result->storage = PL_CHARS_MALLOC;
|
|
|
|
}
|
|
|
|
|
|
|
|
for(to=result->text.t, i=0; i<n; i++)
|
|
|
|
{ memcpy(to, text[i]->text.t, text[i]->length);
|
|
|
|
to += text[i]->length;
|
|
|
|
}
|
|
|
|
*to = EOS;
|
|
|
|
} else
|
|
|
|
{ pl_wchar_t *to;
|
|
|
|
|
|
|
|
result->encoding = ENC_WCHAR;
|
|
|
|
if ( total_length+1 < sizeof(result->buf)/sizeof(pl_wchar_t) )
|
|
|
|
{ result->text.w = (pl_wchar_t*)result->buf;
|
|
|
|
result->storage = PL_CHARS_LOCAL;
|
|
|
|
} else
|
|
|
|
{ result->text.w = PL_malloc((total_length+1)*sizeof(pl_wchar_t));
|
|
|
|
result->storage = PL_CHARS_MALLOC;
|
|
|
|
}
|
|
|
|
|
|
|
|
for(to=result->text.w, i=0; i<n; i++)
|
|
|
|
{ if ( text[i]->encoding == ENC_WCHAR )
|
|
|
|
{ memcpy(to, text[i]->text.w, text[i]->length*sizeof(pl_wchar_t));
|
|
|
|
to += text[i]->length;
|
|
|
|
} else
|
|
|
|
{ const unsigned char *f = (const unsigned char *)text[i]->text.t;
|
|
|
|
const unsigned char *e = &f[text[i]->length];
|
|
|
|
|
|
|
|
while(f<e)
|
|
|
|
*to++ = *f++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
assert((size_t)(to-result->text.w) == total_length);
|
|
|
|
*to = EOS;
|
|
|
|
}
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
IOSTREAM *
|
|
|
|
Sopen_text(PL_chars_t *txt, const char *mode)
|
|
|
|
{ IOSTREAM *stream;
|
|
|
|
|
|
|
|
if ( !streq(mode, "r") )
|
|
|
|
{ errno = EINVAL;
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
stream = Sopen_string(NULL,
|
|
|
|
txt->text.t,
|
|
|
|
bufsize_text(txt, txt->length),
|
|
|
|
mode);
|
|
|
|
stream->encoding = txt->encoding;
|
|
|
|
|
|
|
|
return stream;
|
|
|
|
}
|