This repository has been archived on 2023-08-20. You can view files and clone it, but cannot push or open issues or pull requests.
yap-6.3/C/text.c
2016-05-30 11:22:47 +01:00

1679 lines
44 KiB
C

/*************************************************************************
* *
* YAP Prolog *
* *
* Yap Prolog was developed at NCCUP - Universidade do Porto *
* *
* Copyright L.Damas, V. Santos Costa and Universidade do Porto 1985-- *
* *
**************************************************************************
* *
* File: strings.c *
* comments: General-conversion of character sequences. *
* *
* Last rev: $Date: 2008-07-24 16:02:00 $,$Author: vsc $ *
* *
*************************************************************************/
#include "Yap.h"
#include "YapHeap.h"
#include "YapText.h"
#include "Yatom.h"
#include "eval.h"
#include "yapio.h"
#include <string.h>
#include <wchar.h>
#ifndef HAVE_WCSNLEN
inline static size_t min_size(size_t i, size_t j) { return (i < j ? i : j); }
#define wcsnlen(S, N) min_size(N, wcslen(S))
#endif
static inline unsigned char *getChar(unsigned char *p, int *c) {
*c = *p;
return p + 1;
}
static inline wchar_t *getWchar(wchar_t *p, int *c) {
*c = *p;
return p + 1;
}
#ifndef NAN
#define NAN (0.0 / 0.0)
#endif
static Term Globalize(Term v USES_REGS) {
if (!IsVarTerm(v = Deref(v))) {
return v;
}
if (VarOfTerm(v) > HR && VarOfTerm(v) < LCL0) {
Bind_Local(VarOfTerm(v), MkVarTerm());
v = Deref(v);
}
return v;
}
static Int SkipListCodes(unsigned char **bufp, Term *l, Term **tailp,
Int *atoms, bool *wide, seq_tv_t *inp USES_REGS) {
Int length = 0;
Term *s; /* slow */
Term v; /* temporary */
*wide = false;
size_t max = 1;
unsigned char *st0 = *bufp, *st;
unsigned char *smax = NULL;
do_derefa(v, l, derefa_unk, derefa_nonvar);
*tailp = l;
s = l;
if (inp->type & YAP_STRING_TRUNC) {
max = inp->max;
} else {
max = 0; // basically, this will never be reached;
}
if (!st0) {
if (inp->type & YAP_STRING_MALLOC) {
*bufp = st0 = (unsigned char *)malloc(MAXPATHLEN + 1);
smax = st0 + (MAXPATHLEN - 8); // give 8 bytes for max UTF-8 size + '\0';
} else {
*bufp = st0 = (unsigned char *)Yap_PreAllocCodeSpace();
smax = (unsigned char *)AuxTop -
8; // give 8 bytes for max UTF-8 size + '\0';
}
} else if (inp->sz > 0) {
smax = st0 + (inp->sz - 8); // give 8 bytes for max UTF-8 size + '\0';
} else {
// AUX_ERROR( *l, 2*(length+1), st0, unsigned char);
return 0;
}
*bufp = st = st0;
if (*l == TermNil) {
return 0;
}
if (IsPairTerm(*l)) {
Term hd0 = HeadOfTerm(*l);
if (IsVarTerm(hd0)) {
return -INSTANTIATION_ERROR;
}
// are we looking for atoms/codes?
// whatever the case, we should be consistent throughout,
// so we should be consistent with the first arg.
if (*atoms == 1) {
if (!IsIntegerTerm(hd0)) {
return -INSTANTIATION_ERROR;
}
} else if (*atoms == 2) {
if (!IsAtomTerm(hd0)) {
return -TYPE_ERROR_ATOM;
}
}
do {
int ch;
length++;
if (length == max) {
*st++ = '\0';
}
{
Term hd = Deref(RepPair(*l)[0]);
if (IsVarTerm(hd)) {
return -INSTANTIATION_ERROR;
} else if (IsAtomTerm(hd)) {
(*atoms)++;
if (*atoms < length) {
*tailp = l;
return -TYPE_ERROR_NUMBER;
}
if (IsWideAtom(AtomOfTerm(hd))) {
int ch;
if ((RepAtom(AtomOfTerm(hd))->WStrOfAE)[1] != '\0') {
length = -REPRESENTATION_ERROR_CHARACTER;
}
ch = RepAtom(AtomOfTerm(hd))->WStrOfAE[0];
*wide = true;
} else {
AtomEntry *ae = RepAtom(AtomOfTerm(hd));
if ((ae->StrOfAE)[1] != '\0') {
length = -REPRESENTATION_ERROR_CHARACTER;
} else {
ch = RepAtom(AtomOfTerm(hd))->StrOfAE[0];
*wide |= ch > 0x80;
}
}
} else if (IsIntegerTerm(hd)) {
ch = IntegerOfTerm(hd);
if (*atoms)
length = -TYPE_ERROR_ATOM;
else if (ch < 0) {
*tailp = l;
length = -DOMAIN_ERROR_NOT_LESS_THAN_ZERO;
} else {
*wide |= ch > 0x80;
}
} else {
length = -TYPE_ERROR_INTEGER;
}
if (length < 0) {
*tailp = l;
return length;
}
}
// now copy char to buffer
size_t chsz = put_utf8(st, ch);
if (smax <= st + chsz) {
*st++ = '\0';
*tailp = l;
return length;
} else {
st += chsz;
}
l = RepPair(*l) + 1;
do_derefa(v, l, derefa2_unk, derefa2_nonvar);
} while (*l != *s && IsPairTerm(*l));
}
if (IsVarTerm(*l)) {
return -INSTANTIATION_ERROR;
}
if (*l != TermNil) {
return -TYPE_ERROR_LIST;
}
st[0] = '\0';
*tailp = l;
return length;
}
static void *to_buffer(void *buf, Term t, seq_tv_t *inp, bool *widep,
Int *atoms, size_t *lenp USES_REGS) {
CELL *r = NULL;
Int n;
if (!buf) {
inp->sz = *lenp;
}
unsigned char *bufc = buf;
n = SkipListCodes(&bufc, &t, &r, atoms, widep, inp PASS_REGS);
if (n < 0) {
LOCAL_Error_TYPE = -n;
LOCAL_Error_Term = *r;
return NULL;
}
*lenp = n;
return bufc;
}
static void *Yap_ListOfCodesToBuffer(void *buf, Term t, seq_tv_t *inp,
bool *widep, size_t *lenp USES_REGS) {
Int atoms = 1; // we only want lists of atoms
return to_buffer(buf, t, inp, widep, &atoms, lenp PASS_REGS);
}
static void *Yap_ListOfAtomsToBuffer(void *buf, Term t, seq_tv_t *inp,
bool *widep, size_t *lenp USES_REGS) {
Int atoms = 2; // we only want lists of integer codes
return to_buffer(buf, t, inp, widep, &atoms, lenp PASS_REGS);
}
static void *Yap_ListToBuffer(void *buf, Term t, seq_tv_t *inp, bool *widep,
size_t *lenp USES_REGS) {
Int atoms = 0; // we accept both types of lists.
return to_buffer(buf, t, inp, widep, &atoms, lenp PASS_REGS);
}
#if USE_GEN_TYPE_ERROR
static yap_error_number gen_type_error(int flags) {
if ((flags & (YAP_STRING_STRING | YAP_STRING_ATOM | YAP_STRING_INT |
YAP_STRING_FLOAT | YAP_STRING_ATOMS_CODES | YAP_STRING_BIG)) ==
(YAP_STRING_STRING | YAP_STRING_ATOM | YAP_STRING_INT | YAP_STRING_FLOAT |
YAP_STRING_ATOMS_CODES | YAP_STRING_BIG))
return TYPE_ERROR_TEXT;
if ((flags & (YAP_STRING_STRING | YAP_STRING_ATOM | YAP_STRING_INT |
YAP_STRING_FLOAT | YAP_STRING_BIG)) ==
(YAP_STRING_STRING | YAP_STRING_ATOM | YAP_STRING_INT | YAP_STRING_FLOAT |
YAP_STRING_BIG))
return TYPE_ERROR_ATOMIC;
if ((flags & (YAP_STRING_INT | YAP_STRING_FLOAT | YAP_STRING_BIG)) ==
(YAP_STRING_INT | YAP_STRING_FLOAT | YAP_STRING_BIG))
return TYPE_ERROR_NUMBER;
if (flags & YAP_STRING_ATOM)
return TYPE_ERROR_ATOM;
if (flags & YAP_STRING_STRING)
return TYPE_ERROR_STRING;
if (flags & (YAP_STRING_CODES | YAP_STRING_ATOMS))
return TYPE_ERROR_LIST;
return TYPE_ERROR_NUMBER;
}
#endif
void *Yap_readText(void *buf, seq_tv_t *inp, encoding_t *enc, int *minimal,
size_t *lengp USES_REGS) {
char *s, *s0 = buf;
wchar_t *ws;
bool wide;
/* we know what the term is */
if (!(inp->type & (YAP_STRING_CHARS | YAP_STRING_WCHARS))) {
if (!(inp->type & YAP_STRING_TERM)) {
if (IsVarTerm(inp->val.t)) {
LOCAL_Error_TYPE = INSTANTIATION_ERROR;
} else if (!IsAtomTerm(inp->val.t) && inp->type == YAP_STRING_ATOM) {
LOCAL_Error_TYPE = TYPE_ERROR_ATOM;
} else if (!IsStringTerm(inp->val.t) && inp->type == YAP_STRING_STRING) {
LOCAL_Error_TYPE = TYPE_ERROR_STRING;
} else if (!IsPairTerm(inp->val.t) && !IsStringTerm(inp->val.t) &&
inp->type == (YAP_STRING_ATOMS_CODES | YAP_STRING_STRING)) {
LOCAL_Error_TYPE = TYPE_ERROR_LIST;
} else if (!IsNumTerm(inp->val.t) &&
(inp->type & (YAP_STRING_INT | YAP_STRING_FLOAT |
YAP_STRING_BIG)) == inp->type) {
LOCAL_Error_TYPE = TYPE_ERROR_NUMBER;
}
LOCAL_Error_Term = inp->val.t;
}
}
if (LOCAL_Error_TYPE != YAP_NO_ERROR)
return NULL;
// this is a term, extract the UTF8 representation
if (IsStringTerm(inp->val.t) && inp->type & YAP_STRING_STRING) {
const char *s = StringOfTerm(inp->val.t);
*enc = ENC_ISO_UTF8;
*minimal = FALSE;
if (lengp)
*lengp = strlen(s);
return (void *)s;
}
if (IsAtomTerm(inp->val.t) && inp->type & YAP_STRING_ATOM) {
// this is a term, extract to a buffer, and representation is wide
*minimal = TRUE;
Atom at = AtomOfTerm(inp->val.t);
if (IsWideAtom(at)) {
ws = at->WStrOfAE;
*lengp = wcslen(ws);
*enc = ENC_WCHAR;
return ws;
} else {
s = (char *)at->StrOfAE;
*lengp = strlen(s);
*enc = ENC_ISO_LATIN1;
return s;
}
}
if (((inp->type & (YAP_STRING_CODES | YAP_STRING_ATOMS)) ==
(YAP_STRING_CODES | YAP_STRING_ATOMS))) {
s = Yap_ListToBuffer(s0, inp->val.t, inp, &wide, lengp PASS_REGS);
// this is a term, extract to a sfer, and representation is wide
*minimal = true;
*enc = ENC_ISO_UTF8;
return s;
}
if (inp->type == YAP_STRING_CODES) {
s = Yap_ListOfCodesToBuffer(s0, inp->val.t, inp, &wide, lengp PASS_REGS);
// this is a term, extract to a sfer, and representation is wide
*minimal = true;
*enc = ENC_ISO_UTF8;
return s;
}
if (inp->type == YAP_STRING_ATOMS) {
s = Yap_ListOfAtomsToBuffer(s0, inp->val.t, inp, &wide, lengp PASS_REGS);
// this is a term, extract to a buffer, and representation is wide
*minimal = true;
*enc = ENC_ISO_UTF8;
return s;
}
if (inp->type & YAP_STRING_INT && IsIntegerTerm(inp->val.t)) {
if (s0)
s = s0;
else
s = Yap_PreAllocCodeSpace();
AUX_ERROR(inp->val.t, LOCAL_MAX_SIZE, s, char);
if (snprintf(s, LOCAL_MAX_SIZE - 1, Int_FORMAT, IntegerOfTerm(inp->val.t)) <
0) {
AUX_ERROR(inp->val.t, 2 * LOCAL_MAX_SIZE, s, char);
}
*enc = ENC_ISO_LATIN1;
*lengp = strlen(s);
return s;
}
if (inp->type & YAP_STRING_FLOAT && IsFloatTerm(inp->val.t)) {
if (s0)
s = s0;
else
s = Yap_PreAllocCodeSpace();
AUX_ERROR(inp->val.t, LOCAL_MAX_SIZE, s, char);
if (!Yap_FormatFloat(FloatOfTerm(inp->val.t), &s, LOCAL_MAX_SIZE - 1)) {
AUX_ERROR(inp->val.t, 2 * LOCAL_MAX_SIZE, s, char);
}
*lengp = strlen(s);
*enc = ENC_ISO_LATIN1;
return s;
}
#if USE_GMP
if (inp->type & YAP_STRING_BIG && IsBigIntTerm(inp->val.t)) {
if (s0)
s = s0;
else
s = Yap_PreAllocCodeSpace();
if (!Yap_mpz_to_string(Yap_BigIntOfTerm(inp->val.t), s, LOCAL_MAX_SIZE - 1,
10)) {
AUX_ERROR(inp->val.t, LOCAL_MAX_SIZE, s, char);
}
*enc = ENC_ISO_LATIN1;
*lengp = strlen(s);
return s;
}
#endif
if (inp->type & YAP_STRING_TERM) {
encoding_t enc = ENC_ISO_UTF8;
char *o = Yap_TermToString(inp->val.t, lengp, enc, 0);
return o;
}
if (inp->type & YAP_STRING_CHARS) {
*enc = inp->enc;
if (inp->type & YAP_STRING_NCHARS)
*lengp = inp->sz;
else
*lengp = strlen(inp->val.c);
return (void *)inp->val.c;
}
if (inp->type & YAP_STRING_WCHARS) {
*enc = ENC_WCHAR;
if (inp->type & YAP_STRING_NCHARS)
*lengp = inp->sz;
else
*lengp = wcslen(inp->val.w);
return (void *)inp->val.w;
}
return NULL;
}
static Term write_strings(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
size_t min = 0, max = leng;
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_NCHARS)
min = out->sz;
if (out->type & YAP_STRING_TRUNC && out->max < max)
max = out->max;
}
switch (enc) {
case ENC_ISO_UTF8: {
unsigned char *s = s0, *lim = s + (max = strlen_utf8(s));
Term t = init_tstring(PASS_REGS1);
unsigned char *cp = s, *buf;
LOCAL_TERM_ERROR(t, 2 * (lim - s));
buf = buf_from_tstring(HR);
while (*cp && cp < lim) {
utf8proc_int32_t chr;
cp += get_utf8(cp, -1, &chr);
buf += put_utf8(buf, chr);
}
if (max >= min)
*buf++ = '\0';
else
while (max < min) {
max++;
buf += put_utf8(buf, '\0');
}
close_tstring(buf PASS_REGS);
out->val.t = t;
} break;
case ENC_ISO_LATIN1: {
unsigned char *s = s0, *lim = s + (max = strlen_latin_utf8(s0));
Term t = init_tstring(PASS_REGS1);
unsigned char *cp = s;
unsigned char *buf;
utf8proc_int32_t chr;
LOCAL_TERM_ERROR(t, 2 * (lim - s));
buf = buf_from_tstring(HR);
while (cp < lim) {
cp = getChar(cp, &chr);
buf += put_utf8(buf, chr);
}
if (max >= min)
*buf++ = '\0';
else
while (max < min) {
max++;
buf += put_utf8(buf, chr);
}
close_tstring(buf PASS_REGS);
out->val.t = t;
} break;
case ENC_WCHAR: {
wchar_t *s = s0, *lim = s + (max = strlen_ucs2_utf8(s0));
Term t = init_tstring(PASS_REGS1);
wchar_t *wp = s;
unsigned char *buf;
LOCAL_TERM_ERROR(t, 2 * (lim - s));
buf = buf_from_tstring(HR);
while (wp < lim) {
utf8proc_int32_t chr;
wp = getWchar(wp, &chr);
buf += put_utf8(buf, chr);
}
if (max >= min)
*buf++ = '\0';
else
while (max < min) {
max++;
buf += put_utf8(buf, '\0');
}
close_tstring(buf PASS_REGS);
out->val.t = t;
} break;
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
return out->val.t;
}
static Term write_atoms(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
Term t = AbsPair(HR);
size_t sz = 0;
size_t max = leng;
if (leng == 0) {
out->val.t = t;
return TermNil;
}
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_TRUNC && out->max < max)
max = out->max;
}
switch (enc) {
case ENC_ISO_UTF8: {
unsigned char *s = s0, *lim = s + strnlen((char *)s, max);
unsigned char *cp = s;
wchar_t w[2];
w[1] = '\0';
LOCAL_TERM_ERROR(t, 2 * (lim - s));
while (cp < lim && *cp) {
utf8proc_int32_t chr;
CELL *cl;
cp += get_utf8(cp, -1, &chr);
if (chr == '\0')
break;
w[0] = chr;
cl = HR;
HR += 2;
cl[0] = MkAtomTerm(Yap_LookupMaybeWideAtom(w));
cl[1] = AbsPair(HR);
sz++;
if (sz == max)
break;
}
break;
}
case ENC_ISO_LATIN1: {
unsigned char *s = s0, *lim = s + strnlen(s0, max);
unsigned char *cp = s;
char w[2];
w[1] = '\0';
LOCAL_TERM_ERROR(t, 2 * (lim - s));
while (cp < lim) {
utf8proc_int32_t chr;
cp = getChar(cp, &chr);
if (chr == '\0')
break;
w[0] = chr;
HR[0] = MkAtomTerm(Yap_LookupAtom(w));
HR[1] = AbsPair(HR + 2);
HR += 2;
sz++;
if (sz == max)
break;
}
break;
}
case ENC_WCHAR: {
wchar_t *s = s0, *lim = s + wcsnlen(s, max);
wchar_t *cp = s;
wchar_t w[2];
w[1] = '\0';
LOCAL_TERM_ERROR(t, 2 * (lim - s));
while (*cp && cp < lim) {
utf8proc_int32_t chr;
cp = getWchar(cp, &chr);
if (chr == '\0')
break;
w[0] = chr;
HR[0] = MkAtomTerm(Yap_LookupMaybeWideAtom(w));
HR[1] = AbsPair(HR + 2);
HR += 2;
sz++;
if (sz == max)
break;
}
break;
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
if (out->type & YAP_STRING_DIFF) {
if (sz == 0)
t = out->dif;
else
HR[-1] = Globalize(out->dif PASS_REGS);
} else {
if (sz == 0)
t = TermNil;
else
HR[-1] = TermNil;
}
out->val.t = t;
return (t);
}
static Term write_codes(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
Term t = AbsPair(HR);
size_t min = 0, max = leng;
size_t sz = 0;
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_NCHARS)
min = out->sz;
if (out->type & YAP_STRING_TRUNC && out->max < max)
max = out->max;
}
switch (enc) {
case ENC_ISO_UTF8: {
unsigned char *s = s0, *lim = s + strnlen(s0, max);
unsigned char *cp = s;
LOCAL_TERM_ERROR(t, 2 * (lim - s));
while (*cp && cp < lim) {
utf8proc_int32_t chr;
cp += get_utf8(cp, -1, &chr);
HR[0] = MkIntTerm(chr);
HR[1] = AbsPair(HR + 2);
HR += 2;
sz++;
if (sz == max)
break;
}
break;
}
case ENC_ISO_LATIN1: {
unsigned char *s = s0, *lim = s + strnlen(s0, max);
unsigned char *cp = s;
LOCAL_TERM_ERROR(t, 2 * (lim - s));
while (cp < lim) {
utf8proc_int32_t chr;
cp = getChar(cp, &chr);
HR[0] = MkIntTerm(chr);
HR[1] = AbsPair(HR + 2);
HR += 2;
sz++;
if (sz == max)
break;
}
break;
}
case ENC_WCHAR: {
wchar_t *s = s0, *lim = s + wcsnlen(s, max);
wchar_t *cp = s;
LOCAL_TERM_ERROR(t, 2 * (lim - s));
while (cp < lim) {
utf8proc_int32_t chr;
cp = getWchar(cp, &chr);
HR[0] = MkIntTerm(chr);
HR[1] = AbsPair(HR + 2);
HR += 2;
sz++;
if (sz == max)
break;
}
break;
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
while (sz < min) {
HR[0] = MkIntTerm(MkIntTerm(0));
HR[1] = AbsPair(HR + 2);
HR += 2;
sz++;
}
if (out->type & YAP_STRING_DIFF) {
if (sz == 0)
t = out->dif;
else
HR[-1] = Globalize(out->dif PASS_REGS);
} else {
if (sz == 0)
t = TermNil;
else
HR[-1] = TermNil;
}
out->val.t = t;
return (t);
}
static Atom write_atom(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
size_t max = leng;
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_TRUNC && out->max < max)
max = out->max;
}
switch (enc) {
case ENC_ISO_UTF8: {
unsigned char *s = s0, *lim = s + strnlen(s0, max);
wchar_t *buf = malloc(sizeof(wchar_t) * ((lim + 2) - s)), *ptr = buf;
Atom at;
while (*s && s < lim) {
utf8proc_int32_t chr;
s += get_utf8(s, -1, &chr);
*ptr++ = chr;
}
*ptr++ = '\0';
at = Yap_LookupMaybeWideAtomWithLength(buf, max);
free(buf);
out->val.a = at;
return at;
}
case ENC_ISO_LATIN1: {
char *s = s0;
Atom at;
max = strnlen(s, max);
at = Yap_LookupAtomWithLength(s, max);
out->val.a = at;
return at;
}
case ENC_WCHAR: {
wchar_t *s = s0;
Atom at;
max = wcsnlen(s, max);
out->val.a = at = Yap_LookupMaybeWideAtomWithLength(s, max);
return at;
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc));
}
return NULL;
}
static size_t write_wbuffer(void *s0, seq_tv_t *out, encoding_t enc,
int minimal, size_t leng USES_REGS) {
size_t min = 0, max = leng, sz_end, sz;
out->enc = ENC_WCHAR;
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_NCHARS)
min = out->sz;
if (out->type & YAP_STRING_TRUNC && out->max < max)
max = out->max;
}
if (out->enc != enc ||
out->type & (YAP_STRING_WITH_BUFFER | YAP_STRING_MALLOC)) {
if (enc != ENC_WCHAR) {
sz = strlen((char *)s0) + 1;
} else {
sz = wcslen((wchar_t *)s0) + 1;
}
if (sz < min)
sz = min;
sz *= sizeof(wchar_t);
if (out->type & (YAP_STRING_MALLOC)) {
out->val.w = malloc(sz);
} else if (!(out->type & (YAP_STRING_WITH_BUFFER))) {
if (ASP - (sz / sizeof(CELL) + 1) > HR + 1024) {
out->val.w =
(wchar_t *)(ASP - ((sz * sizeof(wchar_t *) / sizeof(CELL) + 1)));
} else
return -1;
}
} else {
out->val.w = s0;
sz_end = (wcslen(s0) + 1) * sizeof(wchar_t);
}
if (out->enc == ENC_WCHAR) {
switch (enc) {
case ENC_WCHAR:
if (out->type & (YAP_STRING_WITH_BUFFER | YAP_STRING_MALLOC)) {
wchar_t *s = s0;
size_t n = wcslen(s) + 1;
if (n < min)
n = min;
memcpy(out->val.c, s0, n * sizeof(wchar_t));
out->val.w[n] = '\0';
sz_end = n + 1;
}
case ENC_ISO_UTF8: {
unsigned char *s = s0, *lim = s + (max = strnlen(s0, max));
unsigned char *cp = s;
wchar_t *buf0, *buf;
buf = buf0 = out->val.w;
if (!buf)
return -1;
while (*cp && cp < lim) {
utf8proc_int32_t chr;
cp += get_utf8(cp, -1, &chr);
*buf++ = chr;
}
if (max >= min)
*buf++ = '\0';
else
while (max < min) {
utf8proc_int32_t chr;
max++;
cp += get_utf8(cp, -1, &chr);
*buf++ = chr;
}
*buf = '\0';
sz_end = (buf - buf0) + 1;
} break;
case ENC_ISO_LATIN1: {
char *s = s0;
size_t n = strlen(s), i;
if (n < min)
n = min;
for (i = 0; i < n; i++)
out->val.w[i] = s[i];
out->val.w[n] = '\0';
sz_end = n + 1;
} break;
default:
sz_end = -1;
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
}
sz_end *= sizeof(wchar_t);
if (out->type & (YAP_STRING_MALLOC)) {
out->val.c = realloc(out->val.c, sz_end);
}
out->sz = sz_end;
return sz_end;
}
size_t write_buffer(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
size_t min = 0, max = leng, sz_end;
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_NCHARS)
min = out->sz;
if (out->type & YAP_STRING_TRUNC && out->max < max)
max = out->max;
}
if (out->enc != enc) {
size_t sz;
if (enc != ENC_WCHAR)
sz = strlen((char *)s0) + 1;
else
sz = wcslen((wchar_t *)s0) + 1;
if (sz < min)
sz = min;
if (!minimal)
sz *= 4;
if (out->type & (YAP_STRING_MALLOC)) {
out->val.c = malloc(sz);
} else if (!(out->type & (YAP_STRING_WITH_BUFFER))) {
if (ASP - (sz / sizeof(CELL) + 1) > HR + 1024) {
out->val.c = Yap_PreAllocCodeSpace();
}
}
} else {
out->val.c = s0;
}
if (out->enc == ENC_ISO_UTF8) {
switch (enc) {
case ENC_ISO_UTF8:
if (out->type & (YAP_STRING_WITH_BUFFER | YAP_STRING_MALLOC)) {
char *s = s0;
size_t n = strlen(s) + 1;
out->val.c[n] = '\0';
sz_end = n + 1;
} else {
sz_end = strlen(out->val.c) + 1;
}
break;
case ENC_ISO_LATIN1: {
unsigned char *s = s0, *lim = s + (max = strnlen(s0, max));
unsigned char *cp = s, *buf0, *buf;
buf = buf0 = s0;
if (!buf)
return -1;
while (*cp && cp < lim) {
utf8proc_int32_t chr;
chr = *cp++;
buf += put_utf8(buf, chr);
}
if (max >= min)
*buf++ = '\0';
else
while (max < min) {
max++;
utf8proc_int32_t chr;
chr = *cp++;
buf += put_utf8(buf, chr);
}
buf[0] = '\0';
sz_end = (buf + 1) - buf0;
} break;
case ENC_WCHAR: {
wchar_t *s = s0;
unsigned char *buf = out->val.uc;
size_t n = wcslen(s), i;
if (n < min)
n = min;
for (i = 0; i < n; i++) {
utf8proc_int32_t chr = s[i];
buf += put_utf8(buf, chr);
}
*buf++ = '\0';
sz_end = (buf + 1) - out->val.uc;
} break;
default:
sz_end = -1;
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
} else if (out->enc == ENC_ISO_LATIN1) {
switch (enc) {
case ENC_ISO_LATIN1:
if (out->type & YAP_STRING_WITH_BUFFER) {
char *s = s0;
size_t n = strlen(s), i;
if (n < min)
n = min;
memcpy(out->val.c, s0, n);
for (i = 0; i < n; i++)
out->val.w[i] = s[i];
out->val.w[n] = '\0';
sz_end = (n + 1) * sizeof(wchar_t);
} else {
sz_end = strlen(out->val.c) + 1;
}
break;
case ENC_ISO_UTF8: {
unsigned char *s = s0, *lim = s + (max = strnlen(s0, max));
unsigned char *cp = s;
unsigned char *buf0, *buf;
buf = buf0 = out->val.uc;
if (!buf)
return -1;
while (*cp && cp < lim) {
utf8proc_int32_t chr;
cp += get_utf8(cp, -1, &chr);
*buf++ = chr;
}
if (max >= min)
*buf++ = '\0';
else
while (max < min) {
utf8proc_int32_t chr;
max++;
cp += get_utf8(cp, -1, &chr);
*buf++ = chr;
}
sz_end = buf - out->val.uc;
} break;
case ENC_WCHAR: {
wchar_t *s = s0;
size_t n = wcslen(s), i;
if (n < min)
n = min;
for (i = 0; i < n; i++)
out->val.c[i] = s[i];
out->val.c[n] = '\0';
sz_end = n + 1;
} break;
default:
sz_end = -1;
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
} else {
// no other encodings are supported.
sz_end = -1;
}
if (out->type & (YAP_STRING_MALLOC)) {
out->val.c = realloc(out->val.c, sz_end);
}
out->sz = sz_end;
return sz_end;
}
static size_t write_length(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
size_t max = -1;
if (out->type & (YAP_STRING_NCHARS | YAP_STRING_TRUNC)) {
if (out->type & YAP_STRING_NCHARS && out->sz != (size_t)-1)
return out->sz;
if (out->type & YAP_STRING_TRUNC)
max = out->max;
}
switch (enc) {
case ENC_ISO_UTF8: {
const unsigned char *s = s0;
return strlen_utf8(s);
}
case ENC_ISO_LATIN1: {
const char *s = s0;
return strnlen(s, max);
}
case ENC_WCHAR: {
const wchar_t *s = s0;
return wcsnlen(s, max);
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
return (size_t)-1;
}
static Term write_number(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
int size USES_REGS) {
return Yap_StringToNumberTerm(s0, &enc);
}
static Term string_to_term(void *s0, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
Term o = out->val.t =
Yap_StringToTerm(s0, strlen(s0) + 1, &enc, GLOBAL_MaxPriority, NULL);
return o;
}
bool write_Text(void *inp, seq_tv_t *out, encoding_t enc, int minimal,
size_t leng USES_REGS) {
/* we know what the term is */
switch (out->type & YAP_TYPE_MASK) {
case YAP_STRING_STRING:
out->val.t = write_strings(inp, out, enc, minimal, leng PASS_REGS);
return out->val.t != 0;
case YAP_STRING_ATOMS:
out->val.t = write_atoms(inp, out, enc, minimal, leng PASS_REGS);
return out->val.t != 0;
case YAP_STRING_CODES:
out->val.t = write_codes(inp, out, enc, minimal, leng PASS_REGS);
return out->val.t != 0;
case YAP_STRING_LENGTH:
out->val.l = write_length(inp, out, enc, minimal, leng PASS_REGS);
return out->val.l != (size_t)(-1);
case YAP_STRING_ATOM:
out->val.a = write_atom(inp, out, enc, minimal, leng PASS_REGS);
return out->val.a != NULL;
case YAP_STRING_INT | YAP_STRING_FLOAT | YAP_STRING_BIG:
out->val.t = write_number(inp, out, enc, minimal, leng PASS_REGS);
return out->val.t != 0;
case YAP_STRING_CHARS: {
size_t sz = write_buffer(inp, out, enc, minimal, leng PASS_REGS);
return ((Int)sz > 0);
}
case YAP_STRING_WCHARS: {
size_t sz = write_wbuffer(inp, out, enc, minimal, leng PASS_REGS);
return ((Int)sz > 0);
}
default:
if (!(out->type & YAP_STRING_TERM))
return 0;
if (out->type & (YAP_STRING_INT | YAP_STRING_FLOAT | YAP_STRING_BIG))
if ((out->val.t = write_number(inp, out, enc, minimal, leng PASS_REGS)) !=
0L)
return out->val.t != 0;
if (out->type & (YAP_STRING_ATOM))
if (write_atom(inp, out, enc, minimal, leng PASS_REGS) != NIL) {
Atom at = out->val.a;
if (at != NIL)
out->val.t = MkAtomTerm(at);
return at != NIL;
}
if ((out->val.t = string_to_term(inp, out, enc, minimal, leng PASS_REGS)) !=
0L)
return out->val.t != 0;
}
return false;
}
static size_t upcase(void *s0, seq_tv_t *out, encoding_t enc USES_REGS) {
size_t max = -1;
switch (enc) {
case ENC_ISO_UTF8: {
unsigned char *s = s0;
while (*s) {
// assumes the two code have always the same size;
utf8proc_int32_t chr;
get_utf8(s, -1, &chr);
chr = utf8proc_toupper(chr);
s += put_utf8(s, chr);
}
return true;
}
case ENC_ISO_LATIN1: {
unsigned char *s = s0;
utf8proc_int32_t chr;
while ((chr = *s)) {
// assumes the two code have always the same size;
chr = *s;
chr = utf8proc_toupper(chr);
*s++ = chr;
}
return true;
}
case ENC_WCHAR: {
wchar_t *s = s0;
utf8proc_int32_t chr;
while ((chr = *s)) {
// assumes the two code have always the same size;
chr = *s;
chr = utf8proc_toupper(chr);
*s++ = chr;
}
return true;
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
return false;
}
static size_t downcase(void *s0, seq_tv_t *out, encoding_t enc USES_REGS) {
size_t max = -1;
switch (enc) {
case ENC_ISO_UTF8: {
unsigned char *s = s0;
while (*s) {
// assumes the two code have always the same size;
utf8proc_int32_t chr;
get_utf8(s, -1, &chr);
chr = utf8proc_tolower(chr);
s += put_utf8(s, chr);
}
return true;
}
case ENC_ISO_LATIN1: {
unsigned char *s = s0;
utf8proc_int32_t chr;
while ((chr = *s)) {
// assumes the two code have always the same size;
chr = *s;
chr = utf8proc_tolower(chr);
*s++ = chr;
}
return true;
}
case ENC_WCHAR: {
wchar_t *s = s0;
utf8proc_int32_t chr;
while ((chr = *s)) {
// assumes the two code have always the same size;
chr = *s;
chr = utf8proc_tolower(chr);
*s++ = chr;
}
return true;
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
return false;
}
int Yap_CVT_Text(seq_tv_t *inp, seq_tv_t *out USES_REGS) {
encoding_t enc;
int minimal = FALSE;
char *buf;
size_t leng;
buf = Yap_readText(NULL, inp, &enc, &minimal, &leng PASS_REGS);
if (!buf)
return 0L;
if (out->type & (YAP_STRING_UPCASE|YAP_STRING_DOWNCASE)) {
if (out->type & YAP_STRING_UPCASE) {
if (!upcase(buf, out, enc))
return false;
}
if (out->type & YAP_STRING_DOWNCASE) {
if (!downcase(buf, out, enc))
return false;
}
}
return write_Text(buf, out, enc, minimal, leng PASS_REGS);
}
static void *compute_end(void *s0, encoding_t enc) {
switch (enc) {
case ENC_ISO_LATIN1:
case ENC_ISO_UTF8: {
char *s = (char *)s0;
return s + (1 + strlen(s));
}
case ENC_WCHAR: {
wchar_t *s = (wchar_t *)s0;
return s + (1 + wcslen(s));
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
return NULL;
}
static void *advance_Text(void *s, int l, encoding_t enc) {
switch (enc) {
case ENC_ISO_LATIN1:
return ((char *)s) + l;
case ENC_ISO_UTF8:
return (char *)skip_utf8(s, l);
case ENC_WCHAR:
return ((wchar_t *)s) + l;
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc), __FUNCTION__);
}
return s;
}
static int cmp_Text(void *s1, void *s2, int l, encoding_t enc1,
encoding_t enc2) {
Int i;
switch (enc1) {
case ENC_ISO_LATIN1: {
char *w1 = (char *)s1;
switch (enc2) {
case ENC_ISO_LATIN1:
return strncmp(s1, s2, l);
case ENC_ISO_UTF8: {
utf8proc_int32_t chr1, chr2;
unsigned char *w2 = s2;
for (i = 0; i < l; i++) {
chr1 = *w1++;
w2 += get_utf8(w2, -1, &chr2);
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
case ENC_WCHAR: {
utf8proc_int32_t chr1, chr2;
wchar_t *w2 = s2;
for (i = 0; i < l; i++) {
chr1 = *w1++;
chr2 = *w2++;
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc2), __FUNCTION__);
}
}
case ENC_ISO_UTF8: {
unsigned char *w1 = s1;
switch (enc2) {
case ENC_ISO_LATIN1: {
utf8proc_int32_t chr1, chr2;
unsigned char *w2 = s2;
for (i = 0; i < l; i++) {
chr2 = *w2++;
w1 += get_utf8(w1, -1, &chr1);
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
case ENC_ISO_UTF8: {
utf8proc_int32_t chr1, chr2;
unsigned char *w2 = s2;
for (i = 0; i < l; i++) {
w2 += get_utf8(w2, -1, &chr2);
w1 += get_utf8(w1, -1, &chr1);
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
case ENC_WCHAR: {
utf8proc_int32_t chr1, chr2;
wchar_t *w2 = s2;
for (i = 0; i < l; i++) {
chr2 = *w2++;
w1 += get_utf8(w1, -1, &chr1);
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc2), __FUNCTION__);
}
}
case ENC_WCHAR: {
wchar_t *w1 = (wchar_t *)s1;
switch (enc2) {
case ENC_ISO_LATIN1: {
utf8proc_int32_t chr1, chr2;
char *w2 = s2;
for (i = 0; i < l; i++) {
chr1 = *w1++;
chr2 = *w2++;
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
case ENC_ISO_UTF8: {
utf8proc_int32_t chr1, chr2;
unsigned char *w2 = s2;
for (i = 0; i < l; i++) {
chr1 = *w1++;
w2 += get_utf8(w2, -1, &chr2);
if (chr1 - chr2)
return chr1 - chr2;
}
}
return 0;
case ENC_WCHAR:
return wcsncmp(s1, s2, l);
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc2), __FUNCTION__);
}
}
default:
Yap_Error(SYSTEM_ERROR_INTERNAL, TermNil, "Unsupported Encoding ~s in %s",
enc_name(enc1), __FUNCTION__);
}
return 0;
}
static void *concat(int n, seq_tv_t *out, void *sv[], encoding_t encv[],
size_t lengv[] USES_REGS) {
if (out->type == YAP_STRING_STRING) {
/* we assume we concatenate strings only, or ASCII stuff like numbers */
Term t = init_tstring(PASS_REGS1);
unsigned char *buf = buf_from_tstring(HR);
int i;
for (i = 0; i < n; i++) {
if (encv[i] == ENC_WCHAR) {
wchar_t *ptr = sv[i];
utf8proc_int32_t chr;
while ((chr = *ptr++))
buf += put_utf8(buf, chr);
} else if (encv[i] == ENC_ISO_LATIN1) {
char *ptr = sv[i];
utf8proc_int32_t chr;
while ((chr = *ptr++))
buf += put_utf8(buf, chr);
} else {
char *ptr = sv[i];
utf8proc_int32_t chr;
while ((chr = *ptr++))
*buf++ = chr;
}
}
*buf++ = '\0';
close_tstring(buf PASS_REGS);
out->val.t = t;
return HR;
} else {
encoding_t enc = ENC_ISO_LATIN1;
size_t sz = 0;
int i;
for (i = 0; i < n; i++) {
if (encv[i] != ENC_ISO_LATIN1) {
enc = ENC_WCHAR;
}
sz += write_length(sv[i], out, encv[i], FALSE, lengv[i] PASS_REGS);
}
if (enc == ENC_WCHAR) {
/* wide atom */
wchar_t *buf = (wchar_t *)HR;
Atom at;
LOCAL_ERROR(MkAtomTerm(Yap_LookupWideAtom(buf)), sz + 3);
for (i = 0; i < n; i++) {
if (encv[i] == ENC_WCHAR) {
wchar_t *ptr = sv[i];
utf8proc_int32_t chr;
while ((chr = *ptr++) != '\0')
*buf++ = chr;
} else if (encv[i] == ENC_ISO_LATIN1) {
char *ptr = sv[i];
utf8proc_int32_t chr;
while ((chr = *ptr++) != '\0')
*buf++ = (unsigned char)chr;
} else {
unsigned char *ptr = sv[i];
utf8proc_int32_t chr;
while ((ptr += get_utf8(ptr, -1, &chr)) != NULL) {
if (chr == '\0')
break;
else
*buf++ = chr;
}
}
}
*buf++ = '\0';
at = out->val.a = Yap_LookupWideAtom((wchar_t *)HR);
return at;
} else {
/* atom */
char *buf = (char *)HR;
Atom at;
LOCAL_ERROR(MkAtomTerm(Yap_LookupAtom(buf)), sz / sizeof(CELL) + 3);
for (i = 0; i < n; i++) {
char *ptr = sv[i];
utf8proc_int32_t chr;
while ((chr = *ptr++) != '\0')
*buf++ = chr;
}
*buf++ = '\0';
at = out->val.a = Yap_LookupAtom((const char *)HR);
return at;
}
}
return NULL;
}
static void *slice(size_t min, size_t max, void *buf, seq_tv_t *out,
encoding_t enc USES_REGS) {
if (out->type == YAP_STRING_STRING) {
/* we assume we concatenate strings only, or ASCII stuff like numbers */
Term t = init_tstring(PASS_REGS1);
unsigned char *nbuf = buf_from_tstring(HR);
if (enc == ENC_WCHAR) {
wchar_t *ptr = (wchar_t *)buf + min;
utf8proc_int32_t chr;
while (min++ < max) {
chr = *ptr++;
nbuf += put_utf8(nbuf, chr);
}
} else if (enc == ENC_ISO_LATIN1) {
unsigned char *ptr = (unsigned char *)buf + min;
utf8proc_int32_t chr;
while (min++ < max) {
chr = *ptr++;
nbuf += put_utf8(nbuf, chr);
}
} else {
unsigned char *ptr = skip_utf8(buf, min);
utf8proc_int32_t chr;
if (!ptr)
return NULL;
while (min++ < max) {
ptr += get_utf8(ptr, -1, &chr);
nbuf += put_utf8(nbuf, chr);
}
}
*nbuf++ = '\0';
close_tstring(nbuf PASS_REGS);
out->val.t = t;
return (void *)StringOfTerm(t);
} else {
Atom at;
/* atom */
if (enc == ENC_WCHAR) {
/* wide atom */
wchar_t *nbuf = (wchar_t *)HR;
wchar_t *ptr = (wchar_t *)buf + min;
if (max > min) {
LOCAL_ERROR(MkAtomTerm(Yap_LookupWideAtom(buf)),
(max - min) * sizeof(wchar_t));
memcpy(nbuf, ptr, (max - min) * sizeof(wchar_t));
}
nbuf[max - min] = '\0';
at = Yap_LookupMaybeWideAtom(nbuf);
} else if (enc == ENC_ISO_LATIN1) {
/* atom */
char *nbuf = (char *)HR;
if (max > min) {
char *ptr = (char *)buf + min;
LOCAL_ERROR(MkAtomTerm(Yap_LookupAtom(buf)), max - min);
memcpy(nbuf, ptr, (max - min));
}
nbuf[max - min] = '\0';
at = Yap_LookupAtom(nbuf);
} else {
/* atom */
wchar_t *nbuf = (wchar_t *)HR;
unsigned char *ptr = skip_utf8((unsigned char *)buf, min);
utf8proc_int32_t chr;
LOCAL_ERROR(MkAtomTerm(Yap_LookupAtom(buf)), max - min);
while (min++ < max) {
ptr += get_utf8(ptr, -1, &chr);
*nbuf++ = chr;
}
nbuf[0] = '\0';
at = Yap_LookupMaybeWideAtom((wchar_t *)HR);
}
out->val.a = at;
return at->StrOfAE;
}
return NULL;
}
//
// Out must be an atom or a string
void *Yap_Concat_Text(int n, seq_tv_t inp[], seq_tv_t *out USES_REGS) {
encoding_t *encv;
void **bufv;
int minimal = FALSE;
void *buf;
size_t leng, *lengv;
int i;
Term t = ARG1;
bufv = (void **)malloc(n * sizeof(void *));
HEAP_TERM_ERROR(bufv, void *);
encv = (encoding_t *)malloc(n * sizeof(encoding_t));
HEAP_ERROR(encv, encoding_t);
buf = NULL;
for (i = 0; i < n; i++) {
void *nbuf =
Yap_readText(buf, inp + i, encv + i, &minimal, &leng PASS_REGS);
if (!nbuf)
return 0L;
bufv[i] = nbuf;
if ((char *)nbuf >= AuxBase && (char *)nbuf < AuxTop) {
buf = compute_end(nbuf, encv[i]);
}
}
lengv = (size_t *)malloc(n * sizeof(size_t));
HEAP_ERROR(lengv, size_t);
buf = concat(n, out, bufv, encv, lengv PASS_REGS);
free(bufv);
free(lengv);
free(encv);
return buf;
}
//
// out must be an atom or a string
void *Yap_Splice_Text(int n, size_t cuts[], seq_tv_t *inp, encoding_t encv[],
seq_tv_t outv[] USES_REGS) {
encoding_t enc;
int minimal = FALSE;
void *buf, *store;
size_t l, leng;
int i, min;
buf = Yap_readText(NULL, inp, &enc, &minimal, &leng PASS_REGS);
if (!buf)
return NULL;
l = write_length(buf, inp, enc, minimal, leng PASS_REGS);
/* where to allocate next is the most complicated part */
if ((char *)buf >= AuxBase && (char *)buf < AuxTop) {
store = compute_end(buf, enc);
} else {
store = NULL;
}
if (!cuts) {
if (n == 2) {
size_t l0, l1;
size_t leng0, leng1;
encoding_t enc0, enc1;
int minimal0, minimal1;
void *buf0, *buf1;
if (outv[0].val.t) {
buf0 = Yap_readText(store, outv, &enc0, &minimal0, &leng0 PASS_REGS);
if (!buf0)
return NULL;
l0 = write_length(buf0, outv, enc, minimal0, leng0 PASS_REGS);
if (cmp_Text(buf, buf0, l0, enc, enc0) != 0)
return NULL;
l1 = l - l0;
buf1 = slice(l0, l, buf, outv + 1, enc PASS_REGS);
if (encv)
encv[1] = enc;
return buf1;
} else /* if (outv[1].val.t) */ {
buf1 =
Yap_readText(store, outv + 1, &enc1, &minimal1, &leng1 PASS_REGS);
if (!buf1)
return NULL;
l1 = write_length(buf1, outv + 1, enc1, minimal1, leng1 PASS_REGS);
if (l < l1)
return NULL;
l0 = l - l1;
if (cmp_Text(advance_Text(buf, l0, enc), buf1, l1, enc, enc1) != 0)
return NULL;
buf0 = slice(0, l0, buf, outv, enc PASS_REGS);
if (encv)
encv[0] = enc;
return buf0;
}
}
}
for (i = 0; i < n; i++) {
if (i == 0)
min = 0;
else
min = cuts[i - 1];
slice(min, cuts[i], buf, outv + i, enc PASS_REGS);
if (!(outv[i].val.a))
return NULL;
if (encv)
encv[i] = enc;
}
return (void *)outv;
;
}
/**
* Function to convert a generic text term (string, atom, list of codes, list of
atoms) into a buff
er.
*
* @param t the term
* @param buf the buffer, if NULL a buffer is malloced, and the user should
reclai it
* @param len buffer size
* @param enc encoding (UTF-8 is strongly recommended)
*
* @return the buffer, or NULL in case of failure. If so, Yap_Error may be
called.
*/
const char *Yap_TextTermToText(Term t, char *buf, size_t len, encoding_t enc) {
CACHE_REGS
seq_tv_t inp, out;
inp.val.t = t;
if (IsAtomTerm(t)) {
inp.type = YAP_STRING_ATOM;
if (IsWideAtom(AtomOfTerm(t)))
inp.enc = ENC_WCHAR;
else
inp.enc = ENC_ISO_LATIN1;
} else if (IsStringTerm(t)) {
inp.type = YAP_STRING_STRING;
inp.enc = ENC_ISO_UTF8;
} else if (IsPairTerm(t)) {
inp.type = (YAP_STRING_CODES | YAP_STRING_ATOMS);
} else {
Yap_Error(TYPE_ERROR_TEXT, t, NULL);
return false;
}
out.enc = enc;
out.type = YAP_STRING_CHARS;
if (!buf) {
inp.type |= YAP_STRING_MALLOC;
out.type |= YAP_STRING_MALLOC;
}
out.val.c = buf;
if (!Yap_CVT_Text(&inp, &out PASS_REGS))
return NULL;
return out.val.c;
}
/**
* Convert from a text buffer (8-bit) to a term that has the same type as
* _Tguide_
*
* @param s the buffer
* @param tguide the guide
*
* @return the term
*/
Term Yap_MkTextTerm(const char *s, encoding_t enc, Term tguide) {
CACHE_REGS
if (IsAtomTerm(tguide))
return MkAtomTerm(Yap_LookupAtom(s));
if (IsStringTerm(tguide))
return MkStringTerm(s);
if (IsPairTerm(tguide) && IsAtomTerm(HeadOfTerm(tguide))) {
return Yap_CharsToListOfAtoms(s, enc PASS_REGS);
}
return Yap_CharsToListOfCodes(s, enc PASS_REGS);
}