#include #include #include #include #include #include #include "zstring.h" #ifndef _MSC_VER #include #else typedef unsigned __int64 uint64_t; typedef signed __int64 int64_t; #endif // Even though the standard C library has a function to do printf-style formatting in a // generic way, there is no standard interface to this function. So if you want to do // some printf formatting that doesn't fit in the context of the provided functions, // you need to roll your own. Why is that? // // Maybe Microsoft wants you to write a better one yourself? When used as part of a // sprintf replacement, this function is significantly faster than Microsoft's // offering. When used as part of a fprintf replacement, this function turns out to // be slower, but that's probably because the CRT's fprintf can interact with the // FILE object on a low level for better perfomance. If you sprintf into a buffer // and then fwrite that buffer, this routine wins again, though the difference isn't // great. namespace StringFormat { int Worker (OutputFunc output, void *outputData, const char *fmt, ...) { va_list arglist; int len; va_start (arglist, fmt); len = VWorker (output, outputData, fmt, arglist); va_end (arglist); return len; } static inline int writepad (OutputFunc output, void *outputData, const char *pad, int padsize, int spaceToFill) { int outlen = 0; while (spaceToFill > 0) { int count = spaceToFill > padsize ? padsize : spaceToFill; outlen += output (outputData, pad, count); spaceToFill -= count; } return outlen; } // Gasp! This is supposed to be a replacement for sprintf formatting, but // I used sprintf for doubles anyway! Oh no! static int fmt_fp (OutputFunc output, void *outputData, int flags, int precision, int width, double number, char type) { char *buff; char format[16]; int i; format[0] = '%'; i = 1; if (flags & F_MINUS) format[i++] = '-'; if (flags & F_PLUS) format[i++] = '+'; if (flags & F_ZERO) format[i++] = '0'; if (flags & F_BLANK) format[i++] = ' '; if (flags & F_HASH) format[i++] = '#'; format[i++] = '*'; format[i++] = '.'; format[i++] = '*'; format[i++] = type; format[i++] = '\0'; buff = (char *)alloca (1000 + precision); i = sprintf (buff, format, width, precision, number); return output (outputData, buff, i); } int VWorker (OutputFunc output, void *outputData, const char *fmt, va_list arglist) { const char *c; const char *base; int len = 0; int width; int precision; int flags; base = c = fmt; for (;;) { while (*c && *c != '%') { ++c; } if (*c == '\0') { return len + output (outputData, base, int(c - base)); } if (c - base > 0) { len += output (outputData, base, int(c - base)); } c++; // Gather the flags, if any for (flags = 0;; ++c) { if (*c == '-') { flags |= F_MINUS; // bit 0 } else if (*c == '+') { flags |= F_PLUS; // bit 1 } else if (*c == '0') { flags |= F_ZERO; // bit 2 } else if (*c == ' ') { flags |= F_BLANK; // bit 3 } else if (*c == '#') { flags |= F_HASH; // bit 4 } else { break; } } width = precision = -1; // Read the width, if any if (*c == '*') { ++c; width = va_arg (arglist, int); if (width < 0) { // Negative width means minus flag and positive width flags |= F_MINUS; width = -width; } } else if (*c >= '0' && *c <= '9') { width = *c++ - '0'; while (*c >= '0' && *c <= '9') { width = width * 10 + *c++ - '0'; } } // If 0 and - both appear, 0 is ignored. // If the blank and + both appear, the blank is ignored. flags &= ~((flags & 3) << 2); // Read the precision, if any if (*c == '.') { precision = 0; if (*++c == '*') { ++c; precision = va_arg (arglist, int); } else if (*c >= '0' && *c <= '9') { precision = *c++ - '0'; while (*c >= '0' && *c <= '9') { precision = precision * 10 + *c++ - '0'; } } } // Read the size prefix, if any if (*c == 'h') { if (*++c == 'h') { flags |= F_HALFHALF; ++c; } else { flags |= F_HALF; } } else if (*c == 'l') { if (*++c == 'l') { flags |= F_LONGLONG; ++c; } else { flags |= F_LONG; } } else if (*c == 'I') { if (*++c == '6') { if (*++c == '4') { flags |= F_LONGLONG; ++c; } } else { flags |= F_BIGI; } } else if (*c == 't') { flags |= F_PTRDIFF; ++c; } else if (*c == 'z') { flags |= F_SIZE; ++c; } base = c+1; // Now that that's all out of the way, we should be pointing at the type specifier { static const char hexits[18] = {'0','1','2','3','4','5','6','7','8','9','a','b','c','d','e','f','0','x'}; static const char HEXits[18] = {'0','1','2','3','4','5','6','7','8','9','A','B','C','D','E','F','0','X'}; static const char spaces[16] = {' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' ',' '}; static const char zeroes[17] = {'0','0','0','0','0','0','0','0','0','0','0','0','0','0','0','0','.'}; static const char plusprefix = '+'; static const char minusprefix = '-'; static const char dotchar = '.'; const char *prefix = NULL; int prefixlen = 0; int postprefixzeros = 0; int size = flags & 0xF000; char buffer[32], *ibuff; const char *obuff = 0; char type = *c++; int bufflen = 0; int outlen = 0; unsigned int intarg = 0; uint64_t int64arg = 0; const void *voidparg; const char *charparg; const char *xits = hexits; int inlen = len; // Using a bunch of if/else if statements is faster than a switch, because a switch generates // a jump table. A jump table means a possible data cache miss and a hefty penalty while the // cache line is loaded. if (type == 'x' || type == 'X' || type == 'p' || type == 'd' || type == 'u' || type == 'i' || type == 'o') { if (type == 'X' || type == 'p') { xits = HEXits; } if (type == 'p') { type = 'X'; voidparg = va_arg (arglist, void *); if (sizeof(void*) == sizeof(int)) { intarg = (unsigned int)(size_t)voidparg; precision = 8; size = 0; } else { int64arg = (uint64_t)(size_t)voidparg; precision = 16; size = F_LONGLONG; } } else { if (size == 0) { intarg = va_arg (arglist, int); } else if (size == F_HALFHALF) { intarg = va_arg (arglist, int); intarg = (signed char)intarg; } else if (size == F_HALF) { intarg = va_arg (arglist, int); intarg = (short)intarg; } else if (size == F_LONG) { if (sizeof(long) == sizeof(int)) intarg = va_arg (arglist, int); else { int64arg = va_arg (arglist, int64_t); size = F_LONGLONG; } } else if (size == F_BIGI) { if (sizeof(void*) == sizeof(int)) intarg = va_arg (arglist, int); else { int64arg = va_arg (arglist, int64_t); size = F_LONGLONG; } } else if (size == F_LONGLONG) { int64arg = va_arg (arglist, int64_t); } else if (size == F_PTRDIFF) { if (sizeof(ptrdiff_t) == sizeof(int)) intarg = va_arg (arglist, int); else { int64arg = va_arg (arglist, int64_t); size = F_LONGLONG; } } else if (size == F_SIZE) { if (sizeof(size_t) == sizeof(int)) intarg = va_arg (arglist, int); else { int64arg = va_arg (arglist, int64_t); size = F_LONGLONG; } } else { intarg = va_arg (arglist, int); } } if (precision < 0) precision = 1; ibuff = &buffer[sizeof(buffer)]; if (size == F_LONGLONG) { if (int64arg == 0) { flags |= F_ZEROVALUE; } else { if (type == 'o') { // Octal: Dump digits until it fits in an unsigned int while (int64arg > UINT_MAX) { *--ibuff = char(int64arg & 7) + '0'; int64arg >>= 3; } intarg = int(int64arg); } else if (type == 'x' || type == 'X') { // Hexadecimal: Dump digits until it fits in an unsigned int while (int64arg > UINT_MAX) { *--ibuff = xits[int64arg & 15]; int64arg >>= 4; } intarg = int(int64arg); } else { if (type != 'u') { // If a signed number is negative, set the negative flag and make it positive. int64_t sint64arg = (int64_t)int64arg; if (sint64arg < 0) { flags |= F_NEGATIVE; sint64arg = -sint64arg; int64arg = sint64arg; } flags |= F_SIGNED; type = 'u'; } // If an unsigned int64 is too big to fit in an unsigned int, dump out // digits until it is sufficiently small. while (int64arg > INT_MAX) { *--ibuff = char(int64arg % 10) + '0'; int64arg /= 10; } intarg = (unsigned int)(int64arg); } } } else { if (intarg == 0) { flags |= F_ZEROVALUE; } else if (type == 'i' || type == 'd') { // If a signed int is negative, set the negative flag and make it positive. signed int sintarg = (signed int)intarg; if (sintarg < 0) { flags |= F_NEGATIVE; sintarg = -sintarg; intarg = sintarg; } flags |= F_SIGNED; type = 'u'; } } if (flags & F_ZEROVALUE) { if (precision != 0) { *--ibuff = '0'; } } else if (type == 'u') { // Decimal while (intarg != 0) { *--ibuff = char(intarg % 10) + '0'; intarg /= 10; } } else if (type == 'o') { // Octal while (intarg != 0) { *--ibuff = char(intarg & 7) + '0'; intarg >>= 3; } } else { // Hexadecimal while (intarg != 0) { *--ibuff = xits[intarg & 15]; intarg >>= 4; } } // Check for prefix (only for non-decimal, which are always unsigned) if ((flags & (F_HASH|F_ZEROVALUE)) == F_HASH) { if (type == 'o') { if (bufflen >= precision) { prefix = zeroes; prefixlen = 1; } } else if (type == 'x' || type == 'X') { prefix = xits + 16; prefixlen = 2; } } bufflen = (int)(ptrdiff_t)(&buffer[sizeof(buffer)] - ibuff); obuff = ibuff; if (precision >= 0) { postprefixzeros = precision - bufflen; if (postprefixzeros < 0) postprefixzeros = 0; // flags &= ~F_ZERO; } } else if (type == 'c') { intarg = va_arg (arglist, int); buffer[0] = char(intarg); bufflen = 1; obuff = buffer; } else if (type == 's') { charparg = va_arg (arglist, const char *); if (charparg == NULL) { obuff = "(null)"; bufflen = 6; } else { obuff = charparg; if (precision < 0) { bufflen = (int)strlen (charparg); } else { for (bufflen = 0; bufflen < precision && charparg[bufflen] != '\0'; ++bufflen) { /* empty */ } } } } else if (type == '%') { // Just print a '%': Output it with the next stage. base--; continue; } else if (type == 'n') { if (size == F_HALFHALF) { *va_arg (arglist, char *) = (char)inlen; } else if (size == F_HALF) { *va_arg (arglist, short *) = (short)inlen; } else if (size == F_LONG) { *va_arg (arglist, long *) = inlen; } else if (size == F_LONGLONG) { *va_arg (arglist, int64_t *) = inlen; } else if (size == F_BIGI) { *va_arg (arglist, ptrdiff_t *) = inlen; } else { *va_arg (arglist, int *) = inlen; } } else if (type == 'e' || type == 'E' || type == 'f' || type == 'F' || type == 'g' || type == 'G') { // IEEE 754 floating point numbers #ifdef _MSC_VER #define FP_SIGN_MASK (1ui64<<63) #define FP_EXPONENT_MASK (2047ui64<<52) #define FP_FRACTION_MASK ((1ui64<<52)-1) #else #define FP_SIGN_MASK (1llu<<63) #define FP_EXPONENT_MASK (2047llu<<52) #define FP_FRACTION_MASK ((1llu<<52)-1) #endif union { double d; uint64_t i64; } number; number.d = va_arg (arglist, double); if (precision < 0) precision = 6; flags |= F_SIGNED; if ((number.i64 & FP_EXPONENT_MASK) == FP_EXPONENT_MASK) { if (number.i64 & FP_SIGN_MASK) { flags |= F_NEGATIVE; } if ((number.i64 & FP_FRACTION_MASK) == 0) { obuff = "Infinity"; bufflen = 8; } else if ((number.i64 & ((FP_FRACTION_MASK+1)>>1)) == 0) { obuff = "NaN"; bufflen = 3; } else { obuff = "Ind"; } } else { // Converting a binary floating point number to an ASCII decimal // representation is non-trivial, so I'm not going to do it myself. // (At least for now.) len += fmt_fp (output, outputData, flags, precision, width, number.d, type); continue; } } // Check for sign prefix (only for signed numbers) if (flags & F_SIGNED) { if (flags & F_NEGATIVE) { prefix = &minusprefix; prefixlen = 1; } else if (flags & F_PLUS) { prefix = &plusprefix; prefixlen = 1; } else if (flags & F_BLANK) { prefix = spaces; prefixlen = 1; } } // Pad the output to the field width, if needed int fieldlen = prefixlen + postprefixzeros + bufflen; const char *pad = (flags & F_ZERO) ? zeroes : spaces; // If the output is right aligned and zero-padded, then the prefix must come before the padding. if ((flags & (F_ZERO|F_MINUS)) == F_ZERO && prefixlen > 0) { outlen += output (outputData, prefix, prefixlen); prefixlen = 0; } if (!(flags & F_MINUS) && fieldlen < width) { // Field is right-justified, so padding comes first outlen += writepad (output, outputData, pad, sizeof(spaces), width - fieldlen); width = -1; } // Output field: Prefix, post-prefix zeros, buffer text if (prefixlen > 0) { outlen += output (outputData, prefix, prefixlen); } outlen += writepad (output, outputData, zeroes, sizeof(spaces), postprefixzeros); if (bufflen > 0) { outlen += output (outputData, obuff, bufflen); } if ((flags & F_MINUS) && fieldlen < width) { // Field is left-justified, so padding comes last outlen += writepad (output, outputData, pad, sizeof(spaces), width - fieldlen); } len += outlen; } } } };