diff --git a/README.md b/README.md index ede414b..7de4c3e 100644 --- a/README.md +++ b/README.md @@ -13,6 +13,7 @@ Tiny but poweful single file wavefront obj loader written in C++. No dependency What's new ---------- +* Mar 03, 2015 : Replace atof() with hand-written parser for robust reading of numeric value. Thanks skurmedel! * Feb 06, 2015 : Fix parsing multi-material object * Sep 14, 2014 : Add support for multi-material per object/group. Thanks Mykhailo! * Mar 17, 2014 : Fixed trim newline bugs. Thanks ardneran! diff --git a/tiny_obj_loader.cc b/tiny_obj_loader.cc index fb78074..9cd3fd2 100644 --- a/tiny_obj_loader.cc +++ b/tiny_obj_loader.cc @@ -5,6 +5,7 @@ // // +// version 0.9.9: Replace atof() with custom parser. // version 0.9.8: Fix multi-materials(per-face material ID). // version 0.9.7: Support multi-materials(per-face material ID) per // object/group. @@ -22,6 +23,7 @@ #include #include #include +#include #include #include @@ -87,13 +89,165 @@ static inline int parseInt(const char *&token) { return i; } + +// Tries to parse a floating point number located at s. +// +// s_end should be a location in the string where reading should absolutely +// stop. For example at the end of the string, to prevent buffer overflows. +// +// Parses the following EBNF grammar: +// sign = "+" | "-" ; +// END = ? anything not in digit ? +// digit = "0" | "1" | "2" | "3" | "4" | "5" | "6" | "7" | "8" | "9" ; +// integer = [sign] , digit , {digit} ; +// decimal = integer , ["." , integer] ; +// float = ( decimal , END ) | ( decimal , ("E" | "e") , integer , END ) ; +// +// Valid strings are for example: +// -0 +3.1417e+2 -0.0E-3 1.0324 -1.41 11e2 +// +// If the parsing is a success, result is set to the parsed value and true +// is returned. +// +// The function is greedy and will parse until any of the following happens: +// - a non-conforming character is encountered. +// - s_end is reached. +// +// The following situations triggers a failure: +// - s >= s_end. +// - parse failure. +// +static bool tryParseDouble(const char *s, const char *s_end, double *result) +{ + if (s >= s_end) + { + return false; + } + + double mantissa = 0.0; + // This exponent is base 2 rather than 10. + // However the exponent we parse is supposed to be one of ten, + // thus we must take care to convert the exponent/and or the + // mantissa to a * 2^E, where a is the mantissa and E is the + // exponent. + // To get the final double we will use ldexp, it requires the + // exponent to be in base 2. + int exponent = 0; + + // NOTE: THESE MUST BE DECLARED HERE SINCE WE ARE NOT ALLOWED + // TO JUMP OVER DEFINITIONS. + char sign = '+'; + char exp_sign = '+'; + char const *curr = s; + + // How many characters were read in a loop. + int read = 0; + // Tells whether a loop terminated due to reaching s_end. + bool end_not_reached = false; + + /* + BEGIN PARSING. + */ + + // Find out what sign we've got. + if (*curr == '+' || *curr == '-') + { + sign = *curr; + curr++; + } + else if (isdigit(*curr)) { /* Pass through. */ } + else + { + goto fail; + } + + // Read the integer part. + while ((end_not_reached = (curr != s_end)) && isdigit(*curr)) + { + mantissa *= 10; + mantissa += static_cast(*curr - 0x30); + curr++; read++; + } + + // We must make sure we actually got something. + if (read == 0) + goto fail; + // We allow numbers of form "#", "###" etc. + if (!end_not_reached) + goto assemble; + + // Read the decimal part. + if (*curr == '.') + { + curr++; + read = 1; + while ((end_not_reached = (curr != s_end)) && isdigit(*curr)) + { + // NOTE: Don't use powf here, it will absolutely murder precision. + mantissa += static_cast(*curr - 0x30) * pow(10, -read); + read++; curr++; + } + } + else if (*curr == 'e' || *curr == 'E') {} + else + { + goto assemble; + } + + if (!end_not_reached) + goto assemble; + + // Read the exponent part. + if (*curr == 'e' || *curr == 'E') + { + curr++; + // Figure out if a sign is present and if it is. + if ((end_not_reached = (curr != s_end)) && (*curr == '+' || *curr == '-')) + { + exp_sign = *curr; + curr++; + } + else if (isdigit(*curr)) { /* Pass through. */ } + else + { + // Empty E is not allowed. + goto fail; + } + + read = 0; + while ((end_not_reached = (curr != s_end)) && isdigit(*curr)) + { + exponent *= 10; + exponent += static_cast(*curr - 0x30); + curr++; read++; + } + exponent *= (exp_sign == '+'? 1 : -1); + if (read == 0) + goto fail; + } + +assemble: + *result = (sign == '+'? 1 : -1) * ldexp(mantissa * pow(5, exponent), exponent); + return true; +fail: + return false; +} static inline float parseFloat(const char *&token) { token += strspn(token, " \t"); +#ifdef TINY_OBJ_LOADER_OLD_FLOAT_PARSER float f = (float)atof(token); token += strcspn(token, " \t\r"); +#else + const char *end = token + strcspn(token, " \t\r"); + double val = 0.0; + tryParseDouble(token, end, &val); + float f = static_cast(val); + token = end; +#endif return f; } + static inline void parseFloat2(float &x, float &y, const char *&token) { x = parseFloat(token); y = parseFloat(token);