1024 lines
20 KiB
C
1024 lines
20 KiB
C
/*
|
|
* $Id: misc.c,v 1.20 2006-09-10 16:04:34 bacon Exp $
|
|
*/
|
|
|
|
#include <xp/awk/awk_i.h>
|
|
|
|
#ifndef XP_AWK_STAND_ALONE
|
|
#include <xp/bas/assert.h>
|
|
#include <xp/bas/stdarg.h>
|
|
#endif
|
|
|
|
static int __vprintf (
|
|
xp_awk_t* awk, const xp_char_t* fmt, xp_va_list ap);
|
|
static int __vsprintf (
|
|
xp_awk_t* awk, xp_char_t* buf, xp_size_t size,
|
|
const xp_char_t* fmt, xp_va_list ap);
|
|
|
|
static xp_char_t* __adjust_format (xp_awk_t* awk, const xp_char_t* format);
|
|
|
|
xp_long_t xp_awk_strtolong (
|
|
xp_awk_t* awk, const xp_char_t* str,
|
|
int base, const xp_char_t** endptr)
|
|
{
|
|
xp_long_t n = 0;
|
|
const xp_char_t* p;
|
|
int digit, negative = 0;
|
|
|
|
xp_assert (base < 37);
|
|
|
|
p = str; while (XP_AWK_ISSPACE(awk,*p)) p++;
|
|
|
|
while (*p != XP_T('\0'))
|
|
{
|
|
if (*p == XP_T('-'))
|
|
{
|
|
negative = ~negative;
|
|
p++;
|
|
}
|
|
else if (*p == XP_T('+')) p++;
|
|
else break;
|
|
}
|
|
|
|
if (base == 0)
|
|
{
|
|
if (*p == XP_T('0'))
|
|
{
|
|
p++;
|
|
if (*p == XP_T('x') || *p == XP_T('X'))
|
|
{
|
|
p++; base = 16;
|
|
}
|
|
else if (*p == XP_T('b') || *p == XP_T('B'))
|
|
{
|
|
p++; base = 2;
|
|
}
|
|
else base = 8;
|
|
}
|
|
else base = 10;
|
|
}
|
|
else if (base == 16)
|
|
{
|
|
if (*p == XP_T('0') &&
|
|
(*(p+1) == XP_T('x') || *(p+1) == XP_T('X'))) p += 2;
|
|
}
|
|
else if (base == 2)
|
|
{
|
|
if (*p == XP_T('0') &&
|
|
(*(p+1) == XP_T('b') || *(p+1) == XP_T('B'))) p += 2;
|
|
}
|
|
|
|
while (*p != XP_T('\0'))
|
|
{
|
|
if (*p >= XP_T('0') && *p <= XP_T('9'))
|
|
digit = *p - XP_T('0');
|
|
else if (*p >= XP_T('A') && *p <= XP_T('Z'))
|
|
digit = *p - XP_T('A') + 10;
|
|
else if (*p >= XP_T('a') && *p <= XP_T('z'))
|
|
digit = *p - XP_T('a') + 10;
|
|
else break;
|
|
|
|
if (digit >= base) break;
|
|
n = n * base + digit;
|
|
|
|
p++;
|
|
}
|
|
|
|
if (endptr != XP_NULL) *endptr = p;
|
|
return (negative)? -n: n;
|
|
}
|
|
|
|
|
|
/*
|
|
* xp_awk_strtoreal is almost a replica of strtod.
|
|
*
|
|
* strtod.c --
|
|
*
|
|
* Source code for the "strtod" library procedure.
|
|
*
|
|
* Copyright (c) 1988-1993 The Regents of the University of California.
|
|
* Copyright (c) 1994 Sun Microsystems, Inc.
|
|
*
|
|
* Permission to use, copy, modify, and distribute this
|
|
* software and its documentation for any purpose and without
|
|
* fee is hereby granted, provided that the above copyright
|
|
* notice appear in all copies. The University of California
|
|
* makes no representations about the suitability of this
|
|
* software for any purpose. It is provided "as is" without
|
|
* express or implied warranty.
|
|
*/
|
|
|
|
#define MAX_EXPONENT 511
|
|
|
|
xp_real_t xp_awk_strtoreal (xp_awk_t* awk, const xp_char_t* str)
|
|
{
|
|
/*
|
|
* Table giving binary powers of 10. Entry is 10^2^i.
|
|
* Used to convert decimal exponents into floating-point numbers.
|
|
*/
|
|
static xp_real_t powersOf10[] = {
|
|
10., 100., 1.0e4, 1.0e8, 1.0e16,
|
|
1.0e32, 1.0e64, 1.0e128, 1.0e256
|
|
};
|
|
|
|
xp_real_t fraction, dblExp, * d;
|
|
const xp_char_t* p;
|
|
xp_cint_t c;
|
|
int exp = 0; /* Exponent read from "EX" field */
|
|
|
|
/*
|
|
* Exponent that derives from the fractional part. Under normal
|
|
* circumstatnces, it is the negative of the number of digits in F.
|
|
* However, if I is very long, the last digits of I get dropped
|
|
* (otherwise a long I with a large negative exponent could cause an
|
|
* unnecessary overflow on I alone). In this case, frac_exp is
|
|
* incremented one for each dropped digit.
|
|
*/
|
|
|
|
int frac_exp;
|
|
int mantSize; /* Number of digits in mantissa. */
|
|
int decPt; /* Number of mantissa digits BEFORE decimal point */
|
|
const xp_char_t *pExp; /* Temporarily holds location of exponent in string */
|
|
int sign = 0, expSign = 0;
|
|
|
|
p = str;
|
|
|
|
/* Strip off leading blanks and check for a sign */
|
|
while (XP_AWK_ISSPACE(awk,*p)) p++;
|
|
|
|
while (*p != XP_T('\0'))
|
|
{
|
|
if (*p == XP_T('-'))
|
|
{
|
|
sign = ~sign;
|
|
p++;
|
|
}
|
|
else if (*p == XP_T('+')) p++;
|
|
else break;
|
|
}
|
|
|
|
/* Count the number of digits in the mantissa (including the decimal
|
|
* point), and also locate the decimal point. */
|
|
decPt = -1;
|
|
for (mantSize = 0; ; mantSize++) {
|
|
c = *p;
|
|
if (!XP_AWK_ISDIGIT (awk, c)) {
|
|
if ((c != XP_T('.')) || (decPt >= 0)) break;
|
|
decPt = mantSize;
|
|
}
|
|
p++;
|
|
}
|
|
|
|
/*
|
|
* Now suck up the digits in the mantissa. Use two integers to
|
|
* collect 9 digits each (this is faster than using floating-point).
|
|
* If the mantissa has more than 18 digits, ignore the extras, since
|
|
* they can't affect the value anyway.
|
|
*/
|
|
pExp = p;
|
|
p -= mantSize;
|
|
if (decPt < 0)
|
|
{
|
|
decPt = mantSize;
|
|
}
|
|
else
|
|
{
|
|
mantSize -= 1; /* One of the digits was the point */
|
|
}
|
|
|
|
if (mantSize > 18)
|
|
{
|
|
frac_exp = decPt - 18;
|
|
mantSize = 18;
|
|
}
|
|
else
|
|
{
|
|
frac_exp = decPt - mantSize;
|
|
}
|
|
|
|
if (mantSize == 0)
|
|
{
|
|
fraction = 0.0;
|
|
/*p = str;*/
|
|
goto done;
|
|
}
|
|
else
|
|
{
|
|
int frac1, frac2;
|
|
frac1 = 0;
|
|
for ( ; mantSize > 9; mantSize -= 1)
|
|
{
|
|
c = *p;
|
|
p++;
|
|
if (c == XP_T('.'))
|
|
{
|
|
c = *p;
|
|
p++;
|
|
}
|
|
frac1 = 10 * frac1 + (c - XP_T('0'));
|
|
}
|
|
frac2 = 0;
|
|
for (; mantSize > 0; mantSize -= 1) {
|
|
c = *p;
|
|
p++;
|
|
if (c == XP_T('.'))
|
|
{
|
|
c = *p;
|
|
p++;
|
|
}
|
|
frac2 = 10*frac2 + (c - XP_T('0'));
|
|
}
|
|
fraction = (1.0e9 * frac1) + frac2;
|
|
}
|
|
|
|
/* Skim off the exponent */
|
|
p = pExp;
|
|
if ((*p == XP_T('E')) || (*p == XP_T('e')))
|
|
{
|
|
p++;
|
|
if (*p == XP_T('-'))
|
|
{
|
|
expSign = 1;
|
|
p++;
|
|
}
|
|
else
|
|
{
|
|
if (*p == XP_T('+')) p++;
|
|
expSign = 0;
|
|
}
|
|
if (!XP_AWK_ISDIGIT (awk, *p))
|
|
{
|
|
/* p = pExp; */
|
|
goto done;
|
|
}
|
|
while (XP_AWK_ISDIGIT (awk, *p))
|
|
{
|
|
exp = exp * 10 + (*p - XP_T('0'));
|
|
p++;
|
|
}
|
|
}
|
|
|
|
if (expSign) exp = frac_exp - exp;
|
|
else exp = frac_exp + exp;
|
|
|
|
/*
|
|
* Generate a floating-point number that represents the exponent.
|
|
* Do this by processing the exponent one bit at a time to combine
|
|
* many powers of 2 of 10. Then combine the exponent with the
|
|
* fraction.
|
|
*/
|
|
if (exp < 0)
|
|
{
|
|
expSign = 1;
|
|
exp = -exp;
|
|
}
|
|
else expSign = 0;
|
|
|
|
if (exp > MAX_EXPONENT) exp = MAX_EXPONENT;
|
|
|
|
dblExp = 1.0;
|
|
|
|
for (d = powersOf10; exp != 0; exp >>= 1, d++)
|
|
{
|
|
if (exp & 01) dblExp *= *d;
|
|
}
|
|
|
|
if (expSign) fraction /= dblExp;
|
|
else fraction *= dblExp;
|
|
|
|
done:
|
|
return (sign)? -fraction: fraction;
|
|
}
|
|
|
|
xp_size_t xp_awk_longtostr (
|
|
xp_long_t value, int radix, const xp_char_t* prefix,
|
|
xp_char_t* buf, xp_size_t size)
|
|
{
|
|
xp_long_t t, rem;
|
|
xp_size_t len, ret, i;
|
|
xp_size_t prefix_len;
|
|
|
|
prefix_len = (prefix != XP_NULL)? xp_awk_strlen(prefix): 0;
|
|
|
|
t = value;
|
|
if (t == 0)
|
|
{
|
|
/* zero */
|
|
if (buf == XP_NULL) return prefix_len + 1;
|
|
|
|
if (size < prefix_len+1)
|
|
{
|
|
/* buffer too small */
|
|
return (xp_size_t)-1;
|
|
}
|
|
|
|
for (i = 0; i < prefix_len; i++) buf[i] = prefix[i];
|
|
buf[prefix_len] = XP_T('0');
|
|
if (size > prefix_len+1) buf[prefix_len+1] = XP_T('\0');
|
|
return 1;
|
|
}
|
|
|
|
/* non-zero values */
|
|
len = prefix_len;
|
|
if (t < 0) { t = -t; len++; }
|
|
while (t > 0) { len++; t /= radix; }
|
|
|
|
if (buf == XP_NULL)
|
|
{
|
|
/* if buf is not given, return the number of bytes required */
|
|
return len;
|
|
}
|
|
|
|
if (size < len) return (xp_size_t)-1; /* buffer too small */
|
|
if (size > len) buf[len] = XP_T('\0');
|
|
ret = len;
|
|
|
|
t = value;
|
|
if (t < 0) t = -t;
|
|
|
|
while (t > 0)
|
|
{
|
|
rem = t % radix;
|
|
if (rem >= 10)
|
|
buf[--len] = (xp_char_t)rem + XP_T('a') - 10;
|
|
else
|
|
buf[--len] = (xp_char_t)rem + XP_T('0');
|
|
t /= radix;
|
|
}
|
|
|
|
if (value < 0)
|
|
{
|
|
for (i = 1; i <= prefix_len; i++)
|
|
{
|
|
buf[i] = prefix[i-1];
|
|
len--;
|
|
}
|
|
buf[--len] = XP_T('-');
|
|
}
|
|
else
|
|
{
|
|
for (i = 0; i < prefix_len; i++) buf[i] = prefix[i];
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
xp_char_t* xp_awk_strdup (xp_awk_t* awk, const xp_char_t* str)
|
|
{
|
|
xp_char_t* tmp;
|
|
|
|
tmp = (xp_char_t*) XP_AWK_MALLOC (
|
|
awk, (xp_awk_strlen(str) + 1) * xp_sizeof(xp_char_t));
|
|
if (tmp == XP_NULL) return XP_NULL;
|
|
|
|
xp_awk_strcpy (tmp, str);
|
|
return tmp;
|
|
}
|
|
|
|
xp_char_t* xp_awk_strxdup (xp_awk_t* awk, const xp_char_t* str, xp_size_t len)
|
|
{
|
|
xp_char_t* tmp;
|
|
|
|
tmp = (xp_char_t*) XP_AWK_MALLOC (
|
|
awk, (len + 1) * xp_sizeof(xp_char_t));
|
|
if (tmp == XP_NULL) return XP_NULL;
|
|
|
|
xp_awk_strncpy (tmp, str, len);
|
|
return tmp;
|
|
}
|
|
|
|
xp_char_t* xp_awk_strxdup2 (
|
|
xp_awk_t* awk,
|
|
const xp_char_t* str1, xp_size_t len1,
|
|
const xp_char_t* str2, xp_size_t len2)
|
|
{
|
|
xp_char_t* tmp;
|
|
|
|
tmp = (xp_char_t*) XP_AWK_MALLOC (
|
|
awk, (len1 + len2 + 1) * xp_sizeof(xp_char_t));
|
|
if (tmp == XP_NULL) return XP_NULL;
|
|
|
|
xp_awk_strncpy (tmp, str1, len1);
|
|
xp_awk_strncpy (tmp + len1, str2, len2);
|
|
return tmp;
|
|
}
|
|
|
|
xp_size_t xp_awk_strlen (const xp_char_t* str)
|
|
{
|
|
const xp_char_t* p = str;
|
|
while (*p != XP_T('\0')) p++;
|
|
return p - str;
|
|
}
|
|
|
|
xp_size_t xp_awk_strcpy (xp_char_t* buf, const xp_char_t* str)
|
|
{
|
|
xp_char_t* org = buf;
|
|
while ((*buf++ = *str++) != XP_T('\0'));
|
|
return buf - org - 1;
|
|
}
|
|
|
|
xp_size_t xp_awk_strncpy (xp_char_t* buf, const xp_char_t* str, xp_size_t len)
|
|
{
|
|
const xp_char_t* end = str + len;
|
|
while (str < end) *buf++ = *str++;
|
|
*buf = XP_T('\0');
|
|
return len;
|
|
}
|
|
|
|
int xp_awk_strcmp (const xp_char_t* s1, const xp_char_t* s2)
|
|
{
|
|
while (*s1 == *s2)
|
|
{
|
|
if (*s1 == XP_C('\0')) return 0;
|
|
s1++, s2++;
|
|
}
|
|
|
|
return (*s1 > *s2)? 1: -1;
|
|
}
|
|
|
|
int xp_awk_strxncmp (
|
|
const xp_char_t* s1, xp_size_t len1,
|
|
const xp_char_t* s2, xp_size_t len2)
|
|
{
|
|
const xp_char_t* end1 = s1 + len1;
|
|
const xp_char_t* end2 = s2 + len2;
|
|
|
|
while (s1 < end1 && s2 < end2 && *s1 == *s2) s1++, s2++;
|
|
if (s1 == end1 && s2 == end2) return 0;
|
|
if (*s1 == *s2) return (s1 < end1)? 1: -1;
|
|
return (*s1 > *s2)? 1: -1;
|
|
}
|
|
|
|
int xp_awk_strxncasecmp (
|
|
xp_awk_t* awk,
|
|
const xp_char_t* s1, xp_size_t len1,
|
|
const xp_char_t* s2, xp_size_t len2)
|
|
{
|
|
xp_char_t c1, c2;
|
|
const xp_char_t* end1 = s1 + len1;
|
|
const xp_char_t* end2 = s2 + len2;
|
|
|
|
c1 = XP_AWK_TOUPPER (awk, *s1);
|
|
c2 = XP_AWK_TOUPPER (awk, *s2);
|
|
while (s1 < end1 && s2 < end2 && c1 == c2)
|
|
{
|
|
s1++, s2++;
|
|
c1 = XP_AWK_TOUPPER (awk, *s1);
|
|
c2 = XP_AWK_TOUPPER (awk, *s2);
|
|
}
|
|
if (s1 == end1 && s2 == end2) return 0;
|
|
if (c1 == c2) return (s1 < end1)? 1: -1;
|
|
return (c1 > c2)? 1: -1;
|
|
}
|
|
|
|
xp_char_t* xp_awk_strxnstr (
|
|
const xp_char_t* str, xp_size_t strsz,
|
|
const xp_char_t* sub, xp_size_t subsz)
|
|
{
|
|
const xp_char_t* end, * subp;
|
|
|
|
if (subsz == 0) return (xp_char_t*)str;
|
|
if (strsz < subsz) return XP_NULL;
|
|
|
|
end = str + strsz - subsz;
|
|
subp = sub + subsz;
|
|
|
|
while (str <= end) {
|
|
const xp_char_t* x = str;
|
|
const xp_char_t* y = sub;
|
|
|
|
while (xp_true) {
|
|
if (y >= subp) return (xp_char_t*)str;
|
|
if (*x != *y) break;
|
|
x++; y++;
|
|
}
|
|
|
|
str++;
|
|
}
|
|
|
|
return XP_NULL;
|
|
}
|
|
|
|
xp_char_t* xp_awk_strtok (
|
|
xp_awk_run_t* run, const xp_char_t* s,
|
|
const xp_char_t* delim, xp_char_t** tok, xp_size_t* tok_len)
|
|
{
|
|
return xp_awk_strxntok (
|
|
run, s, xp_awk_strlen(s),
|
|
delim, xp_awk_strlen(delim), tok, tok_len);
|
|
}
|
|
|
|
xp_char_t* xp_awk_strxtok (
|
|
xp_awk_run_t* run, const xp_char_t* s, xp_size_t len,
|
|
const xp_char_t* delim, xp_char_t** tok, xp_size_t* tok_len)
|
|
{
|
|
return xp_awk_strxntok (
|
|
run, s, len,
|
|
delim, xp_awk_strlen(delim), tok, tok_len);
|
|
}
|
|
|
|
xp_char_t* xp_awk_strntok (
|
|
xp_awk_run_t* run, const xp_char_t* s,
|
|
const xp_char_t* delim, xp_size_t delim_len,
|
|
xp_char_t** tok, xp_size_t* tok_len)
|
|
{
|
|
return xp_awk_strxntok (
|
|
run, s, xp_awk_strlen(s),
|
|
delim, delim_len, tok, tok_len);
|
|
}
|
|
|
|
xp_char_t* xp_awk_strxntok (
|
|
xp_awk_run_t* run, const xp_char_t* s, xp_size_t len,
|
|
const xp_char_t* delim, xp_size_t delim_len,
|
|
xp_char_t** tok, xp_size_t* tok_len)
|
|
{
|
|
const xp_char_t* p = s, *d;
|
|
const xp_char_t* end = s + len;
|
|
const xp_char_t* sp = XP_NULL, * ep = XP_NULL;
|
|
const xp_char_t* delim_end = delim + delim_len;
|
|
xp_char_t c;
|
|
int delim_mode;
|
|
|
|
#define __DELIM_NULL 0
|
|
#define __DELIM_EMPTY 1
|
|
#define __DELIM_SPACES 2
|
|
#define __DELIM_NOSPACES 3
|
|
#define __DELIM_COMPOSITE 4
|
|
if (delim == XP_NULL) delim_mode = __DELIM_NULL;
|
|
else
|
|
{
|
|
delim_mode = __DELIM_EMPTY;
|
|
|
|
for (d = delim; d < delim_end; d++)
|
|
{
|
|
if (XP_AWK_ISSPACE(run->awk,*d))
|
|
{
|
|
if (delim_mode == __DELIM_EMPTY)
|
|
delim_mode = __DELIM_SPACES;
|
|
else if (delim_mode == __DELIM_NOSPACES)
|
|
{
|
|
delim_mode = __DELIM_COMPOSITE;
|
|
break;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if (delim_mode == __DELIM_EMPTY)
|
|
delim_mode = __DELIM_NOSPACES;
|
|
else if (delim_mode == __DELIM_SPACES)
|
|
{
|
|
delim_mode = __DELIM_COMPOSITE;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (delim_mode == __DELIM_NULL)
|
|
{
|
|
/* when XP_NULL is given as "delim", it trims off the
|
|
* leading and trailing spaces characters off the source
|
|
* string "s" eventually. */
|
|
|
|
while (p < end && XP_AWK_ISSPACE(run->awk,*p)) p++;
|
|
while (p < end)
|
|
{
|
|
c = *p;
|
|
|
|
if (!XP_AWK_ISSPACE(run->awk,c))
|
|
{
|
|
if (sp == XP_NULL) sp = p;
|
|
ep = p;
|
|
}
|
|
p++;
|
|
}
|
|
}
|
|
else if (delim_mode == __DELIM_EMPTY)
|
|
{
|
|
/* each character in the source string "s" becomes a token. */
|
|
if (p < end)
|
|
{
|
|
c = *p;
|
|
sp = p;
|
|
ep = p++;
|
|
}
|
|
}
|
|
else if (delim_mode == __DELIM_SPACES)
|
|
{
|
|
/* each token is delimited by space characters. all leading
|
|
* and trailing spaces are removed. */
|
|
|
|
while (p < end && XP_AWK_ISSPACE(run->awk,*p)) p++;
|
|
while (p < end)
|
|
{
|
|
c = *p;
|
|
if (XP_AWK_ISSPACE(run->awk,c)) break;
|
|
if (sp == XP_NULL) sp = p;
|
|
ep = p++;
|
|
}
|
|
while (p < end && XP_AWK_ISSPACE(run->awk,*p)) p++;
|
|
}
|
|
else if (delim_mode == __DELIM_NOSPACES)
|
|
{
|
|
/* each token is delimited by one of charaters
|
|
* in the delimeter set "delim". */
|
|
if (run->rex.ignorecase)
|
|
{
|
|
while (p < end)
|
|
{
|
|
c = XP_AWK_TOUPPER(run->awk, *p);
|
|
for (d = delim; d < delim_end; d++)
|
|
{
|
|
if (c == XP_AWK_TOUPPER(run->awk,*d)) goto exit_loop;
|
|
}
|
|
|
|
if (sp == XP_NULL) sp = p;
|
|
ep = p++;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
while (p < end)
|
|
{
|
|
c = *p;
|
|
for (d = delim; d < delim_end; d++)
|
|
{
|
|
if (c == *d) goto exit_loop;
|
|
}
|
|
|
|
if (sp == XP_NULL) sp = p;
|
|
ep = p++;
|
|
}
|
|
}
|
|
}
|
|
else /* if (delim_mode == __DELIM_COMPOSITE) */
|
|
{
|
|
/* each token is delimited by one of non-space charaters
|
|
* in the delimeter set "delim". however, all space characters
|
|
* surrounding the token are removed */
|
|
while (p < end && XP_AWK_ISSPACE(run->awk,*p)) p++;
|
|
if (run->rex.ignorecase)
|
|
{
|
|
while (p < end)
|
|
{
|
|
c = XP_AWK_TOUPPER(run->awk, *p);
|
|
if (XP_AWK_ISSPACE(run->awk,c))
|
|
{
|
|
p++;
|
|
continue;
|
|
}
|
|
for (d = delim; d < delim_end; d++)
|
|
{
|
|
if (c == XP_AWK_TOUPPER(run->awk,*d)) goto exit_loop;
|
|
}
|
|
if (sp == XP_NULL) sp = p;
|
|
ep = p++;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
while (p < end)
|
|
{
|
|
c = *p;
|
|
if (XP_AWK_ISSPACE(run->awk,c))
|
|
{
|
|
p++;
|
|
continue;
|
|
}
|
|
for (d = delim; d < delim_end; d++)
|
|
{
|
|
if (c == *d) goto exit_loop;
|
|
}
|
|
if (sp == XP_NULL) sp = p;
|
|
ep = p++;
|
|
}
|
|
}
|
|
}
|
|
|
|
exit_loop:
|
|
if (sp == XP_NULL)
|
|
{
|
|
*tok = XP_NULL;
|
|
*tok_len = (xp_size_t)0;
|
|
}
|
|
else
|
|
{
|
|
*tok = (xp_char_t*)sp;
|
|
*tok_len = ep - sp + 1;
|
|
}
|
|
|
|
/* if XP_NULL is returned, this function should not be called anymore */
|
|
if (p >= end) return XP_NULL;
|
|
if (delim_mode == __DELIM_EMPTY ||
|
|
delim_mode == __DELIM_SPACES) return (xp_char_t*)p;
|
|
return (xp_char_t*)++p;
|
|
}
|
|
|
|
xp_char_t* xp_awk_strxntokbyrex (
|
|
xp_awk_run_t* run, const xp_char_t* s, xp_size_t len,
|
|
void* rex, xp_char_t** tok, xp_size_t* tok_len, int* errnum)
|
|
{
|
|
int n;
|
|
xp_char_t* match_ptr;
|
|
xp_size_t match_len, i;
|
|
xp_size_t left = len;
|
|
const xp_char_t* ptr = s;
|
|
const xp_char_t* str_ptr = s;
|
|
xp_size_t str_len = len;
|
|
|
|
while (len > 0)
|
|
{
|
|
n = xp_awk_matchrex (
|
|
run->awk, rex,
|
|
((run->rex.ignorecase)? XP_AWK_REX_IGNORECASE: 0),
|
|
ptr, left, (const xp_char_t**)&match_ptr, &match_len,
|
|
errnum);
|
|
if (n == -1) return XP_NULL;
|
|
if (n == 0)
|
|
{
|
|
/* no match has been found.
|
|
* return the entire string as a token */
|
|
*tok = (xp_char_t*)str_ptr;
|
|
*tok_len = str_len;
|
|
*errnum = XP_AWK_ENOERR;
|
|
return XP_NULL;
|
|
}
|
|
|
|
xp_assert (n == 1);
|
|
|
|
if (match_len == 0)
|
|
{
|
|
ptr++;
|
|
left--;
|
|
}
|
|
else if (run->awk->option & XP_AWK_STRIPSPACES)
|
|
{
|
|
/* match at the beginning of the input string */
|
|
if (match_ptr == s)
|
|
{
|
|
for (i = 0; i < match_len; i++)
|
|
{
|
|
if (!XP_AWK_ISSPACE(run->awk, match_ptr[i]))
|
|
goto exit_loop;
|
|
}
|
|
|
|
/* the match that are all spaces at the
|
|
* beginning of the input string is skipped */
|
|
ptr += match_len;
|
|
left -= match_len;
|
|
str_ptr = s + match_len;
|
|
str_len -= match_len;
|
|
}
|
|
else break;
|
|
}
|
|
else break;
|
|
}
|
|
|
|
exit_loop:
|
|
if (len == 0)
|
|
{
|
|
*tok = (xp_char_t*)str_ptr;
|
|
*tok_len = str_len;
|
|
*errnum = XP_AWK_ENOERR;
|
|
return XP_NULL;
|
|
}
|
|
|
|
*tok = (xp_char_t*)str_ptr;
|
|
*tok_len = match_ptr - str_ptr;
|
|
|
|
for (i = 0; i < match_len; i++)
|
|
{
|
|
if (!XP_AWK_ISSPACE(run->awk, match_ptr[i]))
|
|
{
|
|
*errnum = XP_AWK_ENOERR;
|
|
return match_ptr+match_len;
|
|
}
|
|
}
|
|
|
|
*errnum = XP_AWK_ENOERR;
|
|
|
|
if (run->awk->option & XP_AWK_STRIPSPACES)
|
|
{
|
|
return (match_ptr+match_len >= s+len)?
|
|
XP_NULL: (match_ptr+match_len);
|
|
}
|
|
else
|
|
{
|
|
return (match_ptr+match_len > s+len)?
|
|
XP_NULL: (match_ptr+match_len);
|
|
}
|
|
}
|
|
|
|
int xp_awk_printf (xp_awk_t* awk, const xp_char_t* fmt, ...)
|
|
{
|
|
int n;
|
|
xp_va_list ap;
|
|
|
|
xp_va_start (ap, fmt);
|
|
n = __vprintf (awk, fmt, ap);
|
|
xp_va_end (ap);
|
|
return n;
|
|
}
|
|
|
|
static int __vprintf (xp_awk_t* awk, const xp_char_t* fmt, xp_va_list ap)
|
|
{
|
|
int n;
|
|
xp_char_t* nf = __adjust_format (awk, fmt);
|
|
if (nf == XP_NULL) return -1;
|
|
|
|
#ifdef XP_CHAR_IS_MCHAR
|
|
n = vprintf (nf, ap);
|
|
#else
|
|
n = vwprintf (nf, ap);
|
|
#endif
|
|
|
|
XP_AWK_FREE (awk, nf);
|
|
return n;
|
|
}
|
|
|
|
int xp_awk_sprintf (
|
|
xp_awk_t* awk, xp_char_t* buf, xp_size_t size,
|
|
const xp_char_t* fmt, ...)
|
|
{
|
|
int n;
|
|
xp_va_list ap;
|
|
|
|
xp_va_start (ap, fmt);
|
|
n = __vsprintf (awk, buf, size, fmt, ap);
|
|
xp_va_end (ap);
|
|
return n;
|
|
}
|
|
|
|
static int __vsprintf (
|
|
xp_awk_t* awk, xp_char_t* buf, xp_size_t size,
|
|
const xp_char_t* fmt, xp_va_list ap)
|
|
{
|
|
int n;
|
|
xp_char_t* nf = __adjust_format (awk, fmt);
|
|
if (nf == XP_NULL) return -1;
|
|
|
|
#if defined(dos) || defined(__dos)
|
|
n = vsprintf (buf, nf, ap); /* TODO: write your own vsnprintf */
|
|
#elif defined(XP_CHAR_IS_MCHAR)
|
|
n = vsnprintf (buf, size, nf, ap);
|
|
#elif defined(_WIN32)
|
|
n = _vsnwprintf (buf, size, nf, ap);
|
|
#else
|
|
n = vswprintf (buf, size, nf, ap);
|
|
#endif
|
|
XP_AWK_FREE (awk, nf);
|
|
return n;
|
|
}
|
|
|
|
#define MOD_SHORT 1
|
|
#define MOD_LONG 2
|
|
#define MOD_LONGLONG 3
|
|
|
|
#define ADDC(str,c) \
|
|
do { \
|
|
if (xp_awk_str_ccat(&str, c) == (xp_size_t)-1) { \
|
|
xp_awk_str_close (&str); \
|
|
return XP_NULL; \
|
|
} \
|
|
} while (0)
|
|
|
|
static xp_char_t* __adjust_format (xp_awk_t* awk, const xp_char_t* format)
|
|
{
|
|
const xp_char_t* fp = format;
|
|
xp_char_t* tmp;
|
|
xp_awk_str_t str;
|
|
xp_char_t ch;
|
|
int modifier;
|
|
|
|
if (xp_awk_str_open (&str, 256, awk) == XP_NULL) return XP_NULL;
|
|
|
|
while (*fp != XP_T('\0'))
|
|
{
|
|
while (*fp != XP_T('\0') && *fp != XP_T('%'))
|
|
{
|
|
ADDC (str, *fp++);
|
|
}
|
|
|
|
if (*fp == XP_T('\0')) break;
|
|
xp_assert (*fp == XP_T('%'));
|
|
|
|
ch = *fp++;
|
|
ADDC (str, ch); /* add % */
|
|
|
|
ch = *fp++;
|
|
|
|
/* flags */
|
|
for (;;)
|
|
{
|
|
if (ch == XP_T(' ') || ch == XP_T('+') ||
|
|
ch == XP_T('-') || ch == XP_T('#'))
|
|
{
|
|
ADDC (str, ch);
|
|
}
|
|
else if (ch == XP_T('0'))
|
|
{
|
|
ADDC (str, ch);
|
|
ch = *fp++;
|
|
break;
|
|
}
|
|
else break;
|
|
|
|
ch = *fp++;
|
|
}
|
|
|
|
/* check the width */
|
|
if (ch == XP_T('*')) ADDC (str, ch);
|
|
else
|
|
{
|
|
while (XP_AWK_ISDIGIT (awk, ch))
|
|
{
|
|
ADDC (str, ch);
|
|
ch = *fp++;
|
|
}
|
|
}
|
|
|
|
/* precision */
|
|
if (ch == XP_T('.'))
|
|
{
|
|
ADDC (str, ch);
|
|
ch = *fp++;
|
|
|
|
if (ch == XP_T('*')) ADDC (str, ch);
|
|
else
|
|
{
|
|
while (XP_AWK_ISDIGIT (awk, ch))
|
|
{
|
|
ADDC (str, ch);
|
|
ch = *fp++;
|
|
}
|
|
}
|
|
}
|
|
|
|
/* modifier */
|
|
for (modifier = 0;;)
|
|
{
|
|
if (ch == XP_T('h')) modifier = MOD_SHORT;
|
|
else if (ch == XP_T('l'))
|
|
{
|
|
modifier = (modifier == MOD_LONG)? MOD_LONGLONG: MOD_LONG;
|
|
}
|
|
else break;
|
|
ch = *fp++;
|
|
}
|
|
|
|
|
|
/* type */
|
|
if (ch == XP_T('%')) ADDC (str, ch);
|
|
else if (ch == XP_T('c') || ch == XP_T('s'))
|
|
{
|
|
#if !defined(XP_CHAR_IS_MCHAR) && !defined(_WIN32)
|
|
ADDC (str, 'l');
|
|
#endif
|
|
ADDC (str, ch);
|
|
}
|
|
else if (ch == XP_T('C') || ch == XP_T('S'))
|
|
{
|
|
#ifdef _WIN32
|
|
ADDC (str, ch);
|
|
#else
|
|
#ifdef XP_CHAR_IS_MCHAR
|
|
ADDC (str, 'l');
|
|
#endif
|
|
ADDC (str, XP_AWK_TOLOWER(awk,ch));
|
|
#endif
|
|
}
|
|
else if (ch == XP_T('d') || ch == XP_T('i') ||
|
|
ch == XP_T('o') || ch == XP_T('u') ||
|
|
ch == XP_T('x') || ch == XP_T('X'))
|
|
{
|
|
if (modifier == MOD_SHORT)
|
|
{
|
|
ADDC (str, 'h');
|
|
}
|
|
else if (modifier == MOD_LONG)
|
|
{
|
|
ADDC (str, 'l');
|
|
}
|
|
else if (modifier == MOD_LONGLONG)
|
|
{
|
|
#if defined(_WIN32) && !defined(__LCC__)
|
|
ADDC (str, 'I');
|
|
ADDC (str, '6');
|
|
ADDC (str, '4');
|
|
#else
|
|
ADDC (str, 'l');
|
|
ADDC (str, 'l');
|
|
#endif
|
|
}
|
|
ADDC (str, ch);
|
|
}
|
|
else if (ch == XP_T('\0')) break;
|
|
else ADDC (str, ch);
|
|
}
|
|
|
|
tmp = XP_AWK_STR_BUF(&str);
|
|
xp_awk_str_forfeit (&str);
|
|
return tmp;
|
|
}
|
|
|