diff options
Diffstat (limited to '')
-rw-r--r-- | external/unbound/compat/snprintf.c | 1036 |
1 files changed, 1036 insertions, 0 deletions
diff --git a/external/unbound/compat/snprintf.c b/external/unbound/compat/snprintf.c new file mode 100644 index 000000000..066355703 --- /dev/null +++ b/external/unbound/compat/snprintf.c @@ -0,0 +1,1036 @@ +/* snprintf - compatibility implementation of snprintf, vsnprintf + * + * Copyright (c) 2013, NLnet Labs. All rights reserved. + * + * This software is open source. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * Redistributions of source code must retain the above copyright notice, + * this list of conditions and the following disclaimer. + * + * Redistributions in binary form must reproduce the above copyright notice, + * this list of conditions and the following disclaimer in the documentation + * and/or other materials provided with the distribution. + * + * Neither the name of the NLNET LABS nor the names of its contributors may + * be used to endorse or promote products derived from this software without + * specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED + * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "config.h" +#include <stdio.h> +#include <ctype.h> +#include <string.h> +#include <stdarg.h> +#include <stdlib.h> +#include <errno.h> +#ifdef HAVE_STDINT_H +#include <stdint.h> +#endif + +/* for test */ +/* #define SNPRINTF_TEST 1 */ +#ifdef SNPRINTF_TEST +#define snprintf my_snprintf +#define vsnprintf my_vsnprintf +#endif /* SNPRINTF_TEST */ + +int snprintf(char* str, size_t size, const char* format, ...); +int vsnprintf(char* str, size_t size, const char* format, va_list arg); + +/** + * Very portable snprintf implementation, limited in functionality, + * esp. for %[capital] %[nonportable] and so on. Reduced float functionality, + * mostly in formatting and range (e+-16), for %f and %g. + * + * %s, %d, %u, %i, %x, %c, %n and %% are fully supported. + * This includes width, precision, flags 0- +, and *(arg for wid,prec). + * %f, %g, %m, %p have reduced support, support for wid,prec,flags,*, but + * less floating point range, no %e formatting for %g. + */ +int snprintf(char* str, size_t size, const char* format, ...) +{ + int r; + va_list args; + va_start(args, format); + r = vsnprintf(str, size, format, args); + va_end(args); + return r; +} + +/** add padding to string */ +static void +print_pad(char** at, size_t* left, int* ret, char p, int num) +{ + while(num--) { + if(*left > 1) { + *(*at)++ = p; + (*left)--; + } + (*ret)++; + } +} + +/** get negative symbol, 0 if none */ +static char +get_negsign(int negative, int plus, int space) +{ + if(negative) + return '-'; + if(plus) + return '+'; + if(space) + return ' '; + return 0; +} + +#define PRINT_DEC_BUFSZ 32 /* 20 is enough for 64 bit decimals */ +/** print decimal into buffer, returns length */ +static int +print_dec(char* buf, int max, unsigned int value) +{ + int i = 0; + if(value == 0) { + if(max > 0) { + buf[0] = '0'; + i = 1; + } + } else while(value && i < max) { + buf[i++] = '0' + value % 10; + value /= 10; + } + return i; +} + +/** print long decimal into buffer, returns length */ +static int +print_dec_l(char* buf, int max, unsigned long value) +{ + int i = 0; + if(value == 0) { + if(max > 0) { + buf[0] = '0'; + i = 1; + } + } else while(value && i < max) { + buf[i++] = '0' + value % 10; + value /= 10; + } + return i; +} + +/** print long decimal into buffer, returns length */ +static int +print_dec_ll(char* buf, int max, unsigned long long value) +{ + int i = 0; + if(value == 0) { + if(max > 0) { + buf[0] = '0'; + i = 1; + } + } else while(value && i < max) { + buf[i++] = '0' + value % 10; + value /= 10; + } + return i; +} + +/** print hex into buffer, returns length */ +static int +print_hex(char* buf, int max, unsigned int value) +{ + const char* h = "0123456789abcdef"; + int i = 0; + if(value == 0) { + if(max > 0) { + buf[0] = '0'; + i = 1; + } + } else while(value && i < max) { + buf[i++] = h[value & 0x0f]; + value >>= 4; + } + return i; +} + +/** print long hex into buffer, returns length */ +static int +print_hex_l(char* buf, int max, unsigned long value) +{ + const char* h = "0123456789abcdef"; + int i = 0; + if(value == 0) { + if(max > 0) { + buf[0] = '0'; + i = 1; + } + } else while(value && i < max) { + buf[i++] = h[value & 0x0f]; + value >>= 4; + } + return i; +} + +/** print long long hex into buffer, returns length */ +static int +print_hex_ll(char* buf, int max, unsigned long long value) +{ + const char* h = "0123456789abcdef"; + int i = 0; + if(value == 0) { + if(max > 0) { + buf[0] = '0'; + i = 1; + } + } else while(value && i < max) { + buf[i++] = h[value & 0x0f]; + value >>= 4; + } + return i; +} + +/** copy string into result, reversed */ +static void +spool_str_rev(char** at, size_t* left, int* ret, const char* buf, int len) +{ + int i = len; + while(i) { + if(*left > 1) { + *(*at)++ = buf[--i]; + (*left)--; + } else --i; + (*ret)++; + } +} + +/** copy string into result */ +static void +spool_str(char** at, size_t* left, int* ret, const char* buf, int len) +{ + int i; + for(i=0; i<len; i++) { + if(*left > 1) { + *(*at)++ = buf[i]; + (*left)--; + } + (*ret)++; + } +} + +/** print number formatted */ +static void +print_num(char** at, size_t* left, int* ret, int minw, int precision, + int prgiven, int zeropad, int minus, int plus, int space, + int zero, int negative, char* buf, int len) +{ + int w = len; /* excludes minus sign */ + char s = get_negsign(negative, plus, space); + if(minus) { + /* left adjust the number into the field, space padding */ + /* calc numw = [sign][zeroes][number] */ + int numw = w; + if(precision == 0 && zero) numw = 0; + if(numw < precision) numw = precision; + if(s) numw++; + + /* sign */ + if(s) print_pad(at, left, ret, s, 1); + + /* number */ + if(precision == 0 && zero) { + /* "" for the number */ + } else { + if(w < precision) + print_pad(at, left, ret, '0', precision - w); + spool_str_rev(at, left, ret, buf, len); + } + /* spaces */ + if(numw < minw) + print_pad(at, left, ret, ' ', minw - numw); + } else { + /* pad on the left of the number */ + /* calculate numw has width of [sign][zeroes][number] */ + int numw = w; + if(precision == 0 && zero) numw = 0; + if(numw < precision) numw = precision; + if(!prgiven && zeropad && numw < minw) numw = minw; + else if(s) numw++; + + /* pad with spaces */ + if(numw < minw) + print_pad(at, left, ret, ' ', minw - numw); + /* print sign (and one less zeropad if so) */ + if(s) { + print_pad(at, left, ret, s, 1); + numw--; + } + /* pad with zeroes */ + if(w < numw) + print_pad(at, left, ret, '0', numw - w); + if(precision == 0 && zero) + return; + /* print the characters for the value */ + spool_str_rev(at, left, ret, buf, len); + } +} + +/** print %d and %i */ +static void +print_num_d(char** at, size_t* left, int* ret, int value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = (value < 0); + int zero = (value == 0); + int len = print_dec(buf, (int)sizeof(buf), + (unsigned int)(negative?-value:value)); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %ld and %li */ +static void +print_num_ld(char** at, size_t* left, int* ret, long value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = (value < 0); + int zero = (value == 0); + int len = print_dec_l(buf, (int)sizeof(buf), + (unsigned long)(negative?-value:value)); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %lld and %lli */ +static void +print_num_lld(char** at, size_t* left, int* ret, long long value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = (value < 0); + int zero = (value == 0); + int len = print_dec_ll(buf, (int)sizeof(buf), + (unsigned long long)(negative?-value:value)); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %u */ +static void +print_num_u(char** at, size_t* left, int* ret, unsigned int value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); + int len = print_dec(buf, (int)sizeof(buf), value); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %lu */ +static void +print_num_lu(char** at, size_t* left, int* ret, unsigned long value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); + int len = print_dec_l(buf, (int)sizeof(buf), value); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %llu */ +static void +print_num_llu(char** at, size_t* left, int* ret, unsigned long long value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); + int len = print_dec_ll(buf, (int)sizeof(buf), value); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %x */ +static void +print_num_x(char** at, size_t* left, int* ret, unsigned int value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); + int len = print_hex(buf, (int)sizeof(buf), value); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %lx */ +static void +print_num_lx(char** at, size_t* left, int* ret, unsigned long value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); + int len = print_hex_l(buf, (int)sizeof(buf), value); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %llx */ +static void +print_num_llx(char** at, size_t* left, int* ret, unsigned long long value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); + int len = print_hex_ll(buf, (int)sizeof(buf), value); + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/** print %llp */ +static void +print_num_llp(char** at, size_t* left, int* ret, void* value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_DEC_BUFSZ]; + int negative = 0; + int zero = (value == 0); +#if defined(UINTPTR_MAX) && defined(UINT32_MAX) && (UINTPTR_MAX == UINT32_MAX) + /* avoid warning about upcast on 32bit systems */ + unsigned long long llvalue = (unsigned long)value; +#else + unsigned long long llvalue = (unsigned long long)value; +#endif + int len = print_hex_ll(buf, (int)sizeof(buf), llvalue); + if(zero) { + buf[0]=')'; + buf[1]='l'; + buf[2]='i'; + buf[3]='n'; + buf[4]='('; + len = 5; + } else { + /* put '0x' in front of the (reversed) buffer result */ + if(len < PRINT_DEC_BUFSZ) + buf[len++] = 'x'; + if(len < PRINT_DEC_BUFSZ) + buf[len++] = '0'; + } + print_num(at, left, ret, minw, precision, prgiven, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +#define PRINT_FLOAT_BUFSZ 64 /* xx.yy with 20.20 about the max */ +/** spool remainder after the decimal point to buffer, in reverse */ +static int +print_remainder(char* buf, int max, double r, int prec) +{ + unsigned long long cap = 1; + unsigned long long value; + int len, i; + if(prec > 19) prec = 19; /* max we can do */ + if(max < prec) return 0; + for(i=0; i<prec; i++) { + cap *= 10; + } + r *= (double)cap; + value = (unsigned long long)r; + /* see if we need to round up */ + if(((unsigned long long)((r - (double)value)*10.0)) >= 5) { + value++; + /* that might carry to numbers before the comma, if so, + * just ignore that rounding. failure because 64bitprintout */ + if(value >= cap) + value = cap-1; + } + len = print_dec_ll(buf, max, value); + while(len < prec) { /* pad with zeroes, e.g. if 0.0012 */ + buf[len++] = '0'; + } + if(len < max) + buf[len++] = '.'; + return len; +} + +/** spool floating point to buffer */ +static int +print_float(char* buf, int max, double value, int prec) +{ + /* as xxx.xxx if prec==0, no '.', with prec decimals after . */ + /* no conversion for NAN and INF, because we do not want to require + linking with -lm. */ + /* Thus, the conversions use 64bit integers to convert the numbers, + * which makes 19 digits before and after the decimal point the max */ + unsigned long long whole = (unsigned long long)value; + double remain = value - (double)whole; + int len = 0; + if(prec != 0) + len = print_remainder(buf, max, remain, prec); + len += print_dec_ll(buf+len, max-len, whole); + return len; +} + +/** print %f */ +static void +print_num_f(char** at, size_t* left, int* ret, double value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_FLOAT_BUFSZ]; + int negative = (value < 0); + int zero = 0; + int len; + if(!prgiven) precision = 6; + len = print_float(buf, (int)sizeof(buf), negative?-value:value, + precision); + print_num(at, left, ret, minw, 1, 0, zeropad, minus, + plus, space, zero, negative, buf, len); +} + +/* rudimentary %g support */ +static int +print_float_g(char* buf, int max, double value, int prec) +{ + unsigned long long whole = (unsigned long long)value; + double remain = value - (double)whole; + int before = 0; + int len = 0; + + /* number of digits before the decimal point */ + while(whole > 0) { + before++; + whole /= 10; + } + whole = (unsigned long long)value; + + if(prec > before && remain != 0.0) { + /* see if the last decimals are zero, if so, skip them */ + len = print_remainder(buf, max, remain, prec-before); + while(len > 0 && buf[0]=='0') { + memmove(buf, buf+1, --len); + } + } + len += print_dec_ll(buf+len, max-len, whole); + return len; +} + + +/** print %g */ +static void +print_num_g(char** at, size_t* left, int* ret, double value, + int minw, int precision, int prgiven, int zeropad, int minus, + int plus, int space) +{ + char buf[PRINT_FLOAT_BUFSZ]; + int negative = (value < 0); + int zero = 0; + int len; + if(!prgiven) precision = 6; + if(precision == 0) precision = 1; + len = print_float_g(buf, (int)sizeof(buf), negative?-value:value, + precision); + print_num(at, left, ret, minw, 1, 0, zeropad, minus, + plus, space, zero, negative, buf, len); +} + + +/** strnlen (compat implementation) */ +static int +my_strnlen(const char* s, int max) +{ + int i; + for(i=0; i<max; i++) + if(s[i]==0) + return i; + return max; +} + +/** print %s */ +static void +print_str(char** at, size_t* left, int* ret, char* s, + int minw, int precision, int prgiven, int minus) +{ + int w; + /* with prec: no more than x characters from this string, stop at 0 */ + if(prgiven) + w = my_strnlen(s, precision); + else w = (int)strlen(s); /* up to the nul */ + if(w < minw && !minus) + print_pad(at, left, ret, ' ', minw - w); + spool_str(at, left, ret, s, w); + if(w < minw && minus) + print_pad(at, left, ret, ' ', minw - w); +} + +/** print %c */ +static void +print_char(char** at, size_t* left, int* ret, int c, + int minw, int minus) +{ + if(1 < minw && !minus) + print_pad(at, left, ret, ' ', minw - 1); + print_pad(at, left, ret, c, 1); + if(1 < minw && minus) + print_pad(at, left, ret, ' ', minw - 1); +} + + +/** + * Print to string. + * str: string buffer for result. result will be null terminated. + * size: size of the buffer. null is put inside buffer. + * format: printf format string. + * arg: '...' arguments to print. + * returns number of characters. a null is printed after this. + * return number of bytes that would have been written + * if the buffer had been large enough. + * + * supported format specifiers: + * %s, %u, %d, %x, %i, %f, %g, %c, %p, %n. + * length: l, ll (for d, u, x). + * precision: 6.6d (for d, u, x) + * %f, %g precisions, 0.3f + * %20s, '.*s' + * and %%. + */ +int vsnprintf(char* str, size_t size, const char* format, va_list arg) +{ + char* at = str; + size_t left = size; + int ret = 0; + const char* fmt = format; + int conv, minw, precision, prgiven, zeropad, minus, plus, space, length; + while(*fmt) { + /* copy string before % */ + while(*fmt && *fmt!='%') { + if(left > 1) { + *at++ = *fmt++; + left--; + } else fmt++; + ret++; + } + + /* see if we are at end */ + if(!*fmt) break; + + /* fetch next argument % designation from format string */ + fmt++; /* skip the '%' */ + + /********************************/ + /* get the argument designation */ + /********************************/ + /* we must do this vararg stuff inside this function for + * portability. Hence, get_designation, and print_designation + * are not their own functions. */ + + /* printout designation: + * conversion specifier: x, d, u, s, c, n, m, p + * flags: # not supported + * 0 zeropad (on the left) + * - left adjust (right by default) + * ' ' printspace for positive number (in - position). + * + alwayssign + * fieldwidth: [1-9][0-9]* minimum field width. + * if this is * then type int next argument specifies the minwidth. + * if this is negative, the - flag is set (with positive width). + * precision: period[digits]*, %.2x. + * if this is * then type int next argument specifies the precision. + * just '.' or negative value means precision=0. + * this is mindigits to print for d, i, u, x + * this is aftercomma digits for f + * this is max number significant digits for g + * maxnumber characters to be printed for s + * length: 0-none (int), 1-l (long), 2-ll (long long) + * notsupported: hh (char), h (short), L (long double), q, j, z, t + * Does not support %m$ and *m$ argument designation as array indices. + * Does not support %#x + * + */ + minw = 0; + precision = 1; + prgiven = 0; + zeropad = 0; + minus = 0; + plus = 0; + space = 0; + length = 0; + + /* get flags in any order */ + for(;;) { + if(*fmt == '0') + zeropad = 1; + else if(*fmt == '-') + minus = 1; + else if(*fmt == '+') + plus = 1; + else if(*fmt == ' ') + space = 1; + else break; + fmt++; + } + + /* field width */ + if(*fmt == '*') { + fmt++; /* skip char */ + minw = va_arg(arg, int); + if(minw < 0) { + minus = 1; + minw = -minw; + } + } else while(*fmt >= '0' && *fmt <= '9') { + minw = minw*10 + (*fmt++)-'0'; + } + + /* precision */ + if(*fmt == '.') { + fmt++; /* skip period */ + prgiven = 1; + precision = 0; + if(*fmt == '*') { + fmt++; /* skip char */ + precision = va_arg(arg, int); + if(precision < 0) + precision = 0; + } else while(*fmt >= '0' && *fmt <= '9') { + precision = precision*10 + (*fmt++)-'0'; + } + } + + /* length */ + if(*fmt == 'l') { + fmt++; /* skip char */ + length = 1; + if(*fmt == 'l') { + fmt++; /* skip char */ + length = 2; + } + } + + /* get the conversion */ + if(!*fmt) conv = 0; + else conv = *fmt++; + + /***********************************/ + /* print that argument designation */ + /***********************************/ + switch(conv) { + case 'i': + case 'd': + if(length == 0) + print_num_d(&at, &left, &ret, va_arg(arg, int), + minw, precision, prgiven, zeropad, minus, plus, space); + else if(length == 1) + print_num_ld(&at, &left, &ret, va_arg(arg, long), + minw, precision, prgiven, zeropad, minus, plus, space); + else if(length == 2) + print_num_lld(&at, &left, &ret, + va_arg(arg, long long), + minw, precision, prgiven, zeropad, minus, plus, space); + break; + case 'u': + if(length == 0) + print_num_u(&at, &left, &ret, + va_arg(arg, unsigned int), + minw, precision, prgiven, zeropad, minus, plus, space); + else if(length == 1) + print_num_lu(&at, &left, &ret, + va_arg(arg, unsigned long), + minw, precision, prgiven, zeropad, minus, plus, space); + else if(length == 2) + print_num_llu(&at, &left, &ret, + va_arg(arg, unsigned long long), + minw, precision, prgiven, zeropad, minus, plus, space); + break; + case 'x': + if(length == 0) + print_num_x(&at, &left, &ret, + va_arg(arg, unsigned int), + minw, precision, prgiven, zeropad, minus, plus, space); + else if(length == 1) + print_num_lx(&at, &left, &ret, + va_arg(arg, unsigned long), + minw, precision, prgiven, zeropad, minus, plus, space); + else if(length == 2) + print_num_llx(&at, &left, &ret, + va_arg(arg, unsigned long long), + minw, precision, prgiven, zeropad, minus, plus, space); + break; + case 's': + print_str(&at, &left, &ret, va_arg(arg, char*), + minw, precision, prgiven, minus); + break; + case 'c': + print_char(&at, &left, &ret, va_arg(arg, int), + minw, minus); + break; + case 'n': + *va_arg(arg, int*) = ret; + break; + case 'm': + print_str(&at, &left, &ret, strerror(errno), + minw, precision, prgiven, minus); + break; + case 'p': + print_num_llp(&at, &left, &ret, va_arg(arg, void*), + minw, precision, prgiven, zeropad, minus, plus, space); + break; + case '%': + print_pad(&at, &left, &ret, '%', 1); + break; + case 'f': + print_num_f(&at, &left, &ret, va_arg(arg, double), + minw, precision, prgiven, zeropad, minus, plus, space); + break; + case 'g': + print_num_g(&at, &left, &ret, va_arg(arg, double), + minw, precision, prgiven, zeropad, minus, plus, space); + break; + /* unknown */ + default: + case 0: break; + } + } + + /* zero terminate */ + if(left > 0) + *at = 0; + return ret; +} + +#ifdef SNPRINTF_TEST + +/** do tests */ +#undef snprintf +#define DOTEST(bufsz, result, retval, ...) do { \ + char buf[bufsz]; \ + printf("now test %s\n", #__VA_ARGS__); \ + int r=my_snprintf(buf, sizeof(buf), __VA_ARGS__); \ + if(r != retval || strcmp(buf, result) != 0) { \ + printf("error test(%s) was \"%s\":%d\n", \ + ""#bufsz", "#result", "#retval", "#__VA_ARGS__, \ + buf, r); \ + exit(1); \ + } \ + r=snprintf(buf, sizeof(buf), __VA_ARGS__); \ + if(r != retval || strcmp(buf, result) != 0) { \ + printf("error test(%s) differs with system, \"%s\":%d\n", \ + ""#bufsz", "#result", "#retval", "#__VA_ARGS__, \ + buf, r); \ + exit(1); \ + } \ + printf("test(\"%s\":%d) passed\n", buf, r); \ + } while(0); + +/** test program */ +int main(void) +{ + int x = 0; + + /* bufsize, expectedstring, expectedretval, snprintf arguments */ + DOTEST(1024, "hello", 5, "hello"); + DOTEST(1024, "h", 1, "h"); + /* warning from gcc for format string, but it does work + * DOTEST(1024, "", 0, ""); */ + + DOTEST(3, "he", 5, "hello"); + DOTEST(1, "", 7, "%d", 7823089); + + /* test positive numbers */ + DOTEST(1024, "0", 1, "%d", 0); + DOTEST(1024, "1", 1, "%d", 1); + DOTEST(1024, "9", 1, "%d", 9); + DOTEST(1024, "15", 2, "%d", 15); + DOTEST(1024, "ab15cd", 6, "ab%dcd", 15); + DOTEST(1024, "167", 3, "%d", 167); + DOTEST(1024, "7823089", 7, "%d", 7823089); + DOTEST(1024, " 12", 3, "%3d", 12); + DOTEST(1024, "012", 3, "%.3d", 12); + DOTEST(1024, "012", 3, "%3.3d", 12); + DOTEST(1024, "012", 3, "%03d", 12); + DOTEST(1024, " 012", 4, "%4.3d", 12); + DOTEST(1024, "", 0, "%.0d", 0); + + /* test negative numbers */ + DOTEST(1024, "-1", 2, "%d", -1); + DOTEST(1024, "-12", 3, "%3d", -12); + DOTEST(1024, " -2", 3, "%3d", -2); + DOTEST(1024, "-012", 4, "%.3d", -12); + DOTEST(1024, "-012", 4, "%3.3d", -12); + DOTEST(1024, "-012", 4, "%4.3d", -12); + DOTEST(1024, " -012", 5, "%5.3d", -12); + DOTEST(1024, "-12", 3, "%03d", -12); + DOTEST(1024, "-02", 3, "%03d", -2); + DOTEST(1024, "-15", 3, "%d", -15); + DOTEST(1024, "-7307", 5, "%d", -7307); + DOTEST(1024, "-12 ", 5, "%-5d", -12); + DOTEST(1024, "-00012", 6, "%-.5d", -12); + + /* test + and space flags */ + DOTEST(1024, "+12", 3, "%+d", 12); + DOTEST(1024, " 12", 3, "% d", 12); + + /* test %u */ + DOTEST(1024, "12", 2, "%u", 12); + DOTEST(1024, "0", 1, "%u", 0); + DOTEST(1024, "4294967295", 10, "%u", 0xffffffff); + + /* test %x */ + DOTEST(1024, "0", 1, "%x", 0); + DOTEST(1024, "c", 1, "%x", 12); + DOTEST(1024, "12ab34cd", 8, "%x", 0x12ab34cd); + + /* test %llu, %lld */ + DOTEST(1024, "18446744073709551615", 20, "%llu", + (long long)0xffffffffffffffff); + DOTEST(1024, "-9223372036854775808", 20, "%lld", + (long long)0x8000000000000000); + DOTEST(1024, "9223372036854775808", 19, "%llu", + (long long)0x8000000000000000); + + /* test %s */ + DOTEST(1024, "hello", 5, "%s", "hello"); + DOTEST(1024, " hello", 10, "%10s", "hello"); + DOTEST(1024, "hello ", 10, "%-10s", "hello"); + DOTEST(1024, "he", 2, "%.2s", "hello"); + DOTEST(1024, " he", 4, "%4.2s", "hello"); + DOTEST(1024, " h", 4, "%4.2s", "h"); + + /* test %c */ + DOTEST(1024, "a", 1, "%c", 'a'); + /* warning from gcc for format string, but it does work + DOTEST(1024, " a", 5, "%5c", 'a'); + DOTEST(1024, "a", 1, "%.0c", 'a'); */ + + /* test %n */ + DOTEST(1024, "hello", 5, "hello%n", &x); + if(x != 5) { printf("the %%n failed\n"); exit(1); } + + /* test %m */ + errno = 0; + DOTEST(1024, "Success", 7, "%m"); + + /* test %p */ + DOTEST(1024, "0x10", 4, "%p", (void*)0x10); + DOTEST(1024, "(nil)", 5, "%p", (void*)0x0); + + /* test %% */ + DOTEST(1024, "%", 1, "%%"); + + /* test %f */ + DOTEST(1024, "0.000000", 8, "%f", 0.0); + DOTEST(1024, "0.00", 4, "%.2f", 0.0); + /* differs, "-0.00" DOTEST(1024, "0.00", 4, "%.2f", -0.0); */ + DOTEST(1024, "234.00", 6, "%.2f", 234.005); + DOTEST(1024, "8973497.1246", 12, "%.4f", 8973497.12456); + DOTEST(1024, "-12.000000", 10, "%f", -12.0); + DOTEST(1024, "6", 1, "%.0f", 6.0); + + DOTEST(1024, "6", 1, "%g", 6.0); + DOTEST(1024, "6.1", 3, "%g", 6.1); + DOTEST(1024, "6.15", 4, "%g", 6.15); + + /* These format strings are from the code of NSD, Unbound, ldns */ + + DOTEST(1024, "abcdef", 6, "%s", "abcdef"); + DOTEST(1024, "005", 3, "%03u", 5); + DOTEST(1024, "12345", 5, "%03u", 12345); + DOTEST(1024, "5", 1, "%d", 5); + DOTEST(1024, "(nil)", 5, "%p", NULL); + DOTEST(1024, "12345", 5, "%ld", (long)12345); + DOTEST(1024, "12345", 5, "%lu", (long)12345); + DOTEST(1024, " 12345", 12, "%12u", (unsigned)12345); + DOTEST(1024, "12345", 5, "%u", (unsigned)12345); + DOTEST(1024, "12345", 5, "%llu", (unsigned long long)12345); + DOTEST(1024, "12345", 5, "%x", 0x12345); + DOTEST(1024, "12345", 5, "%llx", (long long)0x12345); + DOTEST(1024, "012345", 6, "%6.6d", 12345); + DOTEST(1024, "012345", 6, "%6.6u", 12345); + DOTEST(1024, "1234.54", 7, "%g", 1234.54); + DOTEST(1024, "123456789.54", 12, "%.12g", 123456789.54); + DOTEST(1024, "3456789123456.54", 16, "%.16g", 3456789123456.54); + /* %24g does not work with 24 digits, not enough accuracy, + * the first 16 digits are correct */ + DOTEST(1024, "12345", 5, "%3.3d", 12345); + DOTEST(1024, "000", 3, "%3.3d", 0); + DOTEST(1024, "001", 3, "%3.3d", 1); + DOTEST(1024, "012", 3, "%3.3d", 12); + DOTEST(1024, "-012", 4, "%3.3d", -12); + DOTEST(1024, "he", 2, "%.2s", "hello"); + DOTEST(1024, "helloworld", 10, "%s%s", "hello", "world"); + DOTEST(1024, "he", 2, "%.*s", 2, "hello"); + DOTEST(1024, " hello", 7, "%*s", 7, "hello"); + DOTEST(1024, "hello ", 7, "%*s", -7, "hello"); + DOTEST(1024, "0", 1, "%c", '0'); + DOTEST(1024, "A", 1, "%c", 'A'); + DOTEST(1024, "", 1, "%c", 0); + DOTEST(1024, "\010", 1, "%c", 8); + DOTEST(1024, "%", 1, "%%"); + DOTEST(1024, "0a", 2, "%02x", 0x0a); + DOTEST(1024, "bd", 2, "%02x", 0xbd); + DOTEST(1024, "12", 2, "%02ld", (long)12); + DOTEST(1024, "02", 2, "%02ld", (long)2); + DOTEST(1024, "02", 2, "%02u", (unsigned)2); + DOTEST(1024, "765432", 6, "%05u", (unsigned)765432); + DOTEST(1024, "10.234", 6, "%0.3f", 10.23421); + DOTEST(1024, "123456.234", 10, "%0.3f", 123456.23421); + DOTEST(1024, "123456789.234", 13, "%0.3f", 123456789.23421); + DOTEST(1024, "123456.23", 9, "%.2f", 123456.23421); + DOTEST(1024, "123456", 6, "%.0f", 123456.23421); + DOTEST(1024, "0123", 4, "%.4x", 0x0123); + DOTEST(1024, "00000123", 8, "%.8x", 0x0123); + DOTEST(1024, "ffeb0cde", 8, "%.8x", 0xffeb0cde); + DOTEST(1024, " 987654321", 10, "%10lu", (unsigned long)987654321); + DOTEST(1024, " 987654321", 12, "%12lu", (unsigned long)987654321); + DOTEST(1024, "987654321", 9, "%i", 987654321); + DOTEST(1024, "-87654321", 9, "%i", -87654321); + DOTEST(1024, "hello ", 16, "%-16s", "hello"); + DOTEST(1024, " ", 16, "%-16s", ""); + DOTEST(1024, "a ", 16, "%-16s", "a"); + DOTEST(1024, "foobarfoobar ", 16, "%-16s", "foobarfoobar"); + DOTEST(1024, "foobarfoobarfoobar", 18, "%-16s", "foobarfoobarfoobar"); + + /* combined expressions */ + DOTEST(1024, "foo 1.0 size 512 edns", 21, + "foo %s size %d %s%s", "1.0", 512, "", "edns"); + DOTEST(15, "foo 1.0 size 5", 21, + "foo %s size %d %s%s", "1.0", 512, "", "edns"); + DOTEST(1024, "packet 1203ceff id", 18, + "packet %2.2x%2.2x%2.2x%2.2x id", 0x12, 0x03, 0xce, 0xff); + DOTEST(1024, "/tmp/testbound_123abcd.tmp", 26, "/tmp/testbound_%u%s%s.tmp", 123, "ab", "cd"); + + return 0; +} +#endif /* SNPRINTF_TEST */ |