X-Git-Url: https://perl5.git.perl.org/perl5.git/blobdiff_plain/6640aa2c3b93d7ac78e4e86983fe5948b3ca55f2..20fbb8c297b88a2b4ad0fb679a190392299d875b:/numeric.c diff --git a/numeric.c b/numeric.c index c20c85e..f5eadc8 100644 --- a/numeric.c +++ b/numeric.c @@ -29,6 +29,107 @@ values, including such things as replacements for the OS's atof() function #define PERL_IN_NUMERIC_C #include "perl.h" +#ifdef Perl_strtod + +PERL_STATIC_INLINE NV +S_strtod(pTHX_ const char * const s, char ** e) +{ + DECLARATION_FOR_LC_NUMERIC_MANIPULATION; + NV result; + + STORE_LC_NUMERIC_SET_TO_NEEDED(); + +# ifdef USE_QUADMATH + + result = strtoflt128(s, e); + +# elif defined(HAS_STRTOLD) && defined(HAS_LONG_DOUBLE) \ + && defined(USE_LONG_DOUBLE) +# if defined(__MINGW64_VERSION_MAJOR) + /*********************************************** + We are unable to use strtold because of + https://sourceforge.net/p/mingw-w64/bugs/711/ + & + https://sourceforge.net/p/mingw-w64/bugs/725/ + + but __mingw_strtold is fine. + ***********************************************/ + + result = __mingw_strtold(s, e); + +# else + + result = strtold(s, e); + +# endif +# elif defined(HAS_STRTOD) + + result = strtod(s, e); + +# else +# error No strtod() equivalent found +# endif + + RESTORE_LC_NUMERIC(); + + return result; +} + +#endif /* #ifdef Perl_strtod */ + +/* + +=for apidoc my_strtod + +This function is equivalent to the libc strtod() function, and is available +even on platforms that lack plain strtod(). Its return value is the best +available precision depending on platform capabilities and F +options. + +It properly handles the locale radix character, meaning it expects a dot except +when called from within the scope of S>, in which case the radix +character should be that specified by the current locale. + +The synonym Strtod() may be used instead. + +=cut + +*/ + +NV +Perl_my_strtod(const char * const s, char **e) +{ + dTHX; + + PERL_ARGS_ASSERT_MY_STRTOD; + +#ifdef Perl_strtod + + return S_strtod(aTHX_ s, e); + +#else + + { + NV result; + char ** end_ptr = NULL; + + *end_ptr = my_atof2(s, &result); + if (e) { + *e = *end_ptr; + } + + if (! *end_ptr) { + result = 0.0; + } + + return result; + } + +#endif + +} + + U32 Perl_cast_ulong(NV f) { @@ -39,7 +140,7 @@ Perl_cast_ulong(NV f) if (f < U32_MAX_P1_HALF) return (U32) f; f -= U32_MAX_P1_HALF; - return ((U32) f) | (1 + U32_MAX >> 1); + return ((U32) f) | (1 + (U32_MAX >> 1)); #else return (U32) f; #endif @@ -57,7 +158,7 @@ Perl_cast_i32(NV f) if (f < U32_MAX_P1_HALF) return (I32)(U32) f; f -= U32_MAX_P1_HALF; - return (I32)(((U32) f) | (1 + U32_MAX >> 1)); + return (I32)(((U32) f) | (1 + (U32_MAX >> 1))); #else return (I32)(U32) f; #endif @@ -76,7 +177,7 @@ Perl_cast_iv(NV f) if (f < UV_MAX_P1_HALF) return (IV)(UV) f; f -= UV_MAX_P1_HALF; - return (IV)(((UV) f) | (1 + UV_MAX >> 1)); + return (IV)(((UV) f) | (1 + (UV_MAX >> 1))); #else return (IV)(UV) f; #endif @@ -94,7 +195,7 @@ Perl_cast_uv(NV f) if (f < UV_MAX_P1_HALF) return (UV) f; f -= UV_MAX_P1_HALF; - return ((UV) f) | (1 + UV_MAX >> 1); + return ((UV) f) | (1 + (UV_MAX >> 1)); #else return (UV) f; #endif @@ -107,24 +208,24 @@ Perl_cast_uv(NV f) converts a string representing a binary number to numeric form. -On entry I and I<*len> give the string to scan, I<*flags> gives -conversion flags, and I should be NULL or a pointer to an NV. +On entry C and C<*len> give the string to scan, C<*flags> gives +conversion flags, and C should be C or a pointer to an NV. The scan stops at the end of the string, or the first invalid character. -Unless C is set in I<*flags>, encountering an +Unless C is set in C<*flags>, encountering an invalid character will also trigger a warning. -On return I<*len> is set to the length of the scanned string, -and I<*flags> gives output flags. +On return C<*len> is set to the length of the scanned string, +and C<*flags> gives output flags. If the value is <= C it is returned as a UV, the output flags are clear, -and nothing is written to I<*result>. If the value is > UV_MAX C -returns UV_MAX, sets C in the output flags, -and writes the value to I<*result> (or the value is discarded if I +and nothing is written to C<*result>. If the value is > C, C +returns C, sets C in the output flags, +and writes the value to C<*result> (or the value is discarded if C is NULL). -The binary number may optionally be prefixed with "0b" or "b" unless -C is set in I<*flags> on entry. If -C is set in I<*flags> then the binary -number may use '_' characters to separate digits. +The binary number may optionally be prefixed with C<"0b"> or C<"b"> unless +C is set in C<*flags> on entry. If +C is set in C<*flags> then the binary +number may use C<"_"> characters to separate digits. =cut @@ -204,7 +305,7 @@ Perl_grok_bin(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) "Illegal binary digit '%c' ignored", *s); break; } - + if ( ( overflowed && value_nv > 4294967295.0) #if UVSIZE > 4 || (!overflowed && value > 0xffffffff @@ -230,24 +331,24 @@ Perl_grok_bin(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) converts a string representing a hex number to numeric form. -On entry I and I<*len_p> give the string to scan, I<*flags> gives -conversion flags, and I should be NULL or a pointer to an NV. +On entry C and C<*len_p> give the string to scan, C<*flags> gives +conversion flags, and C should be C or a pointer to an NV. The scan stops at the end of the string, or the first invalid character. -Unless C is set in I<*flags>, encountering an +Unless C is set in C<*flags>, encountering an invalid character will also trigger a warning. -On return I<*len> is set to the length of the scanned string, -and I<*flags> gives output flags. +On return C<*len> is set to the length of the scanned string, +and C<*flags> gives output flags. -If the value is <= UV_MAX it is returned as a UV, the output flags are clear, -and nothing is written to I<*result>. If the value is > UV_MAX C -returns UV_MAX, sets C in the output flags, -and writes the value to I<*result> (or the value is discarded if I -is NULL). +If the value is <= C it is returned as a UV, the output flags are clear, +and nothing is written to C<*result>. If the value is > C, C +returns C, sets C in the output flags, +and writes the value to C<*result> (or the value is discarded if C +is C). -The hex number may optionally be prefixed with "0x" or "x" unless -C is set in I<*flags> on entry. If -C is set in I<*flags> then the hex -number may use '_' characters to separate digits. +The hex number may optionally be prefixed with C<"0x"> or C<"x"> unless +C is set in C<*flags> on entry. If +C is set in C<*flags> then the hex +number may use C<"_"> characters to separate digits. =cut @@ -325,7 +426,7 @@ Perl_grok_hex(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) "Illegal hexadecimal digit '%c' ignored", *s); break; } - + if ( ( overflowed && value_nv > 4294967295.0) #if UVSIZE > 4 || (!overflowed && value > 0xffffffff @@ -351,22 +452,22 @@ Perl_grok_hex(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) converts a string representing an octal number to numeric form. -On entry I and I<*len> give the string to scan, I<*flags> gives -conversion flags, and I should be NULL or a pointer to an NV. +On entry C and C<*len> give the string to scan, C<*flags> gives +conversion flags, and C should be C or a pointer to an NV. The scan stops at the end of the string, or the first invalid character. -Unless C is set in I<*flags>, encountering an +Unless C is set in C<*flags>, encountering an 8 or 9 will also trigger a warning. -On return I<*len> is set to the length of the scanned string, -and I<*flags> gives output flags. +On return C<*len> is set to the length of the scanned string, +and C<*flags> gives output flags. -If the value is <= UV_MAX it is returned as a UV, the output flags are clear, -and nothing is written to I<*result>. If the value is > UV_MAX C -returns UV_MAX, sets C in the output flags, -and writes the value to I<*result> (or the value is discarded if I -is NULL). +If the value is <= C it is returned as a UV, the output flags are clear, +and nothing is written to C<*result>. If the value is > C, C +returns C, sets C in the output flags, +and writes the value to C<*result> (or the value is discarded if C +is C). -If C is set in I<*flags> then the octal -number may use '_' characters to separate digits. +If C is set in C<*flags> then the octal +number may use C<"_"> characters to separate digits. =cut @@ -432,7 +533,7 @@ Perl_grok_oct(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) } break; } - + if ( ( overflowed && value_nv > 4294967295.0) #if UVSIZE > 4 || (!overflowed && value > 0xffffffff @@ -518,133 +619,64 @@ Scan and skip for a numeric decimal separator (radix). bool Perl_grok_numeric_radix(pTHX_ const char **sp, const char *send) { -#ifdef USE_LOCALE_NUMERIC - PERL_ARGS_ASSERT_GROK_NUMERIC_RADIX; - - if (IN_LC(LC_NUMERIC)) { - DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED(); - if (PL_numeric_radix_sv) { - STRLEN len; - const char * const radix = SvPV(PL_numeric_radix_sv, len); - if (*sp + len <= send && memEQ(*sp, radix, len)) { - *sp += len; - RESTORE_LC_NUMERIC(); - return TRUE; - } - } - RESTORE_LC_NUMERIC(); - } - /* always try "." if numeric radix didn't match because - * we may have data from different locales mixed */ -#endif - PERL_ARGS_ASSERT_GROK_NUMERIC_RADIX; - if (*sp < send && **sp == '.') { - ++*sp; - return TRUE; - } - return FALSE; -} +#ifdef USE_LOCALE_NUMERIC -/* -=for apidoc nan_hibyte + if (IN_LC(LC_NUMERIC)) { + STRLEN len; + char * radix; + bool matches_radix = FALSE; + DECLARATION_FOR_LC_NUMERIC_MANIPULATION; -Given an NV, returns pointer to the byte containing the most -significant bit of the NaN, this bit is most commonly the -quiet/signaling bit of the NaN. The mask will contain a mask -appropriate for manipulating the most significant bit. -Note that this bit may not be the highest bit of the byte. + STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); -If the NV is not a NaN, returns NULL. + radix = SvPV(PL_numeric_radix_sv, len); + radix = savepvn(radix, len); -Most platforms have "high bit is one" -> quiet nan. -The known opposite exceptions are older MIPS and HPPA platforms. + RESTORE_LC_NUMERIC(); -Some platforms do not differentiate between quiet and signaling NaNs. + if (*sp + len <= send) { + matches_radix = memEQ(*sp, radix, len); + } -=cut -*/ -U8* -Perl_nan_hibyte(NV *nvp, U8* mask) -{ - STRLEN i = (NV_MANT_REAL_DIG - 1) / 8; - STRLEN j = (NV_MANT_REAL_DIG - 1) % 8; + Safefree(radix); - PERL_ARGS_ASSERT_NAN_HIBYTE; + if (matches_radix) { + *sp += len; + return TRUE; + } + } - *mask = 1 << j; -#ifdef NV_BIG_ENDIAN - return (U8*) nvp + NVSIZE - 1 - i; -#endif -#ifdef NV_LITTLE_ENDIAN - return (U8*) nvp + i; #endif -} - -/* -=for apidoc nan_signaling_set - -Set or unset the NaN signaling-ness. - -Of those platforms that differentiate between quiet and signaling -platforms the majority has the semantics of the most significant bit -being on meaning quiet NaN, so for signaling we need to clear the bit. - -Some platforms (older MIPS, and HPPA) have the opposite -semantics, and we set the bit for a signaling NaN. -=cut -*/ -void -Perl_nan_signaling_set(NV *nvp, bool signaling) -{ - U8 mask; - U8* hibyte; - - PERL_ARGS_ASSERT_NAN_SIGNALING_SET; - - hibyte = nan_hibyte(nvp, &mask); - if (hibyte) { - const NV nan = NV_NAN; - /* Decent optimizers should make the irrelevant branch to disappear. */ - if ((((U8*)&nan)[hibyte - (U8*)nvp] & mask)) { - /* x86 style: the most significant bit of the NaN is off - * for a signaling NaN, and on for a quiet NaN. */ - if (signaling) { - *hibyte &= ~mask; - } else { - *hibyte |= mask; - } - } else { - /* MIPS/HPPA style: the most significant bit of the NaN is on - * for a signaling NaN, and off for a quiet NaN. */ - if (signaling) { - *hibyte |= mask; - } else { - *hibyte &= ~mask; - } - } + /* always try "." if numeric radix didn't match because + * we may have data from different locales mixed */ + if (*sp < send && **sp == '.') { + ++*sp; + return TRUE; } + + return FALSE; } /* =for apidoc grok_infnan -Helper for grok_number(), accepts various ways of spelling "infinity" +Helper for C, accepts various ways of spelling "infinity" or "not a number", and returns one of the following flag combinations: - IS_NUMBER_INFINITE + IS_NUMBER_INFINITY IS_NUMBER_NAN - IS_NUMBER_INFINITE | IS_NUMBER_NEG + IS_NUMBER_INFINITY | IS_NUMBER_NEG IS_NUMBER_NAN | IS_NUMBER_NEG 0 -possibly |-ed with IS_NUMBER_TRAILING. +possibly |-ed with C. -If an infinity or a not-a-number is recognized, the *sp will point to +If an infinity or a not-a-number is recognized, C<*sp> will point to one byte past the end of the recognized string. If the recognition fails, -zero is returned, and the *sp will not move. +zero is returned, and C<*sp> will not move. =cut */ @@ -654,6 +686,7 @@ Perl_grok_infnan(pTHX_ const char** sp, const char* send) { const char* s = *sp; int flags = 0; +#if defined(NV_INF) || defined(NV_NAN) bool odh = FALSE; /* one-dot-hash: 1.#INF */ PERL_ARGS_ASSERT_GROK_INFNAN; @@ -878,6 +911,9 @@ Perl_grok_infnan(pTHX_ const char** sp, const char* send) while (s < send && isSPACE(*s)) s++; +#else + PERL_UNUSED_ARG(send); +#endif /* #if defined(NV_INF) || defined(NV_NAN) */ *sp = s; return flags; } @@ -887,20 +923,20 @@ Perl_grok_infnan(pTHX_ const char** sp, const char* send) Recognise (or not) a number. The type of the number is returned (0 if unrecognised), otherwise it is a bit-ORed combination of -IS_NUMBER_IN_UV, IS_NUMBER_GREATER_THAN_UV_MAX, IS_NUMBER_NOT_INT, -IS_NUMBER_NEG, IS_NUMBER_INFINITY, IS_NUMBER_NAN (defined in perl.h). - -If the value of the number can fit in a UV, it is returned in the *valuep -IS_NUMBER_IN_UV will be set to indicate that *valuep is valid, IS_NUMBER_IN_UV -will never be set unless *valuep is valid, but *valuep may have been assigned -to during processing even though IS_NUMBER_IN_UV is not set on return. -If valuep is NULL, IS_NUMBER_IN_UV will be set for the same cases as when -valuep is non-NULL, but no actual assignment (or SEGV) will occur. - -IS_NUMBER_NOT_INT will be set with IS_NUMBER_IN_UV if trailing decimals were -seen (in which case *valuep gives the true value truncated to an integer), and -IS_NUMBER_NEG if the number is negative (in which case *valuep holds the -absolute value). IS_NUMBER_IN_UV is not set if e notation was used or the +C, C, C, +C, C, C (defined in perl.h). + +If the value of the number can fit in a UV, it is returned in C<*valuep>. +C will be set to indicate that C<*valuep> is valid, C +will never be set unless C<*valuep> is valid, but C<*valuep> may have been assigned +to during processing even though C is not set on return. +If C is C, C will be set for the same cases as when +C is non-C, but no actual assignment (or SEGV) will occur. + +C will be set with C if trailing decimals were +seen (in which case C<*valuep> gives the true value truncated to an integer), and +C if the number is negative (in which case C<*valuep> holds the +absolute value). C is not set if e notation was used or the number is larger than a UV. C allows only C, which allows for trailing @@ -909,7 +945,7 @@ C on the result. =for apidoc grok_number -Identical to grok_number_flags() with flags set to zero. +Identical to C with C set to zero. =cut */ @@ -964,41 +1000,41 @@ Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) before checking for overflow. */ if (++s < send) { int digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { digit = *s - '0'; - if (digit >= 0 && digit <= 9) { + if (inRANGE(digit, 0, 9)) { value = value * 10 + digit; if (++s < send) { /* Now got 9 digits, so need to check each time for overflow. */ digit = *s - '0'; - while (digit >= 0 && digit <= 9 + while ( inRANGE(digit, 0, 9) && (value < uv_max_div_10 || (value == uv_max_div_10 && digit <= uv_max_mod_10))) { @@ -1008,7 +1044,7 @@ Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) else break; } - if (digit >= 0 && digit <= 9 + if (inRANGE(digit, 0, 9) && (s < send)) { /* value overflowed. skip the remaining digits, don't @@ -1089,7 +1125,7 @@ Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) s++; if (s >= send) return numtype; - if (len == 10 && memEQ(pv, "0 but true", 10)) { + if (memEQs(pv, len, "0 but true")) { if (valuep) *valuep = 0; return IS_NUMBER_IN_UV; @@ -1098,7 +1134,7 @@ Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) if ((s + 2 < send) && strchr("inqs#", toFOLD(*s))) { /* Really detect inf/nan. Start at d, not s, since the above * code might have already consumed the "1." or "1". */ - int infnan = Perl_grok_infnan(aTHX_ &d, send); + const int infnan = Perl_grok_infnan(aTHX_ &d, send); if ((infnan & IS_NUMBER_INFINITY)) { return (numtype | infnan); /* Keep sign for infinity. */ } @@ -1114,91 +1150,103 @@ Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) } /* -=for apidoc grok_atou - -grok_atou is a safer replacement for atoi and strtol. - -grok_atou parses a C-style zero-byte terminated string, looking for -a decimal unsigned integer. - -Returns the unsigned integer, if a valid value can be parsed -from the beginning of the string. +=for apidoc grok_atoUV -Accepts only the decimal digits '0'..'9'. +parse a string, looking for a decimal unsigned integer. -As opposed to atoi or strtol, grok_atou does NOT allow optional -leading whitespace, or negative inputs. If such features are -required, the calling code needs to explicitly implement those. +On entry, C points to the beginning of the string; +C points to a UV that will receive the converted value, if found; +C is either NULL or points to a variable that points to one byte +beyond the point in C that this routine should examine. +If C is NULL, C is assumed to be NUL-terminated. -If a valid value cannot be parsed, returns either zero (if non-digits -are met before any digits) or UV_MAX (if the value overflows). +Returns FALSE if C doesn't represent a valid unsigned integer value (with +no leading zeros). Otherwise it returns TRUE, and sets C<*valptr> to that +value. -Note that extraneous leading zeros also count as an overflow -(meaning that only "0" is the zero). +If you constrain the portion of C that is looked at by this function (by +passing a non-NULL C), and if the intial bytes of that portion form a +valid value, it will return TRUE, setting C<*endptr> to the byte following the +final digit of the value. But if there is no constraint at what's looked at, +all of C must be valid in order for TRUE to be returned. -On failure, the *endptr is also set to NULL, unless endptr is NULL. +The only characters this accepts are the decimal digits '0'..'9'. -Trailing non-digit bytes are allowed if the endptr is non-NULL. -On return the *endptr will contain the pointer to the first non-digit byte. +As opposed to L or L, C does NOT allow optional +leading whitespace, nor negative inputs. If such features are required, the +calling code needs to explicitly implement those. -If the endptr is NULL, the first non-digit byte MUST be -the zero byte terminating the pv, or zero will be returned. +Note that this function returns FALSE for inputs that would overflow a UV, +or have leading zeros. Thus a single C<0> is accepted, but not C<00> nor +C<01>, C<002>, I. -Background: atoi has severe problems with illegal inputs, it cannot be +Background: C has severe problems with illegal inputs, it cannot be used for incremental parsing, and therefore should be avoided -atoi and strtol are also affected by locale settings, which can also be +C and C are also affected by locale settings, which can also be seen as a bug (global state controlled by user environment). =cut + */ -UV -Perl_grok_atou(const char *pv, const char** endptr) +bool +Perl_grok_atoUV(const char *pv, UV *valptr, const char** endptr) { const char* s = pv; const char** eptr; const char* end2; /* Used in case endptr is NULL. */ - UV val = 0; /* The return value. */ + UV val = 0; /* The parsed value. */ - PERL_ARGS_ASSERT_GROK_ATOU; + PERL_ARGS_ASSERT_GROK_ATOUV; - eptr = endptr ? endptr : &end2; - if (isDIGIT(*s)) { - /* Single-digit inputs are quite common. */ - val = *s++ - '0'; - if (isDIGIT(*s)) { - /* Extra leading zeros cause overflow. */ - if (val == 0) { - *eptr = NULL; - return UV_MAX; - } - while (isDIGIT(*s)) { - /* This could be unrolled like in grok_number(), but - * the expected uses of this are not speed-needy, and - * unlikely to need full 64-bitness. */ - U8 digit = *s++ - '0'; - if (val < uv_max_div_10 || - (val == uv_max_div_10 && digit <= uv_max_mod_10)) { - val = val * 10 + digit; - } else { - *eptr = NULL; - return UV_MAX; - } + if (endptr) { + eptr = endptr; + } + else { + end2 = s + strlen(s); + eptr = &end2; + } + + if ( *eptr <= s + || ! isDIGIT(*s)) + { + return FALSE; + } + + /* Single-digit inputs are quite common. */ + val = *s++ - '0'; + if (s < *eptr && isDIGIT(*s)) { + /* Fail on extra leading zeros. */ + if (val == 0) + return FALSE; + while (s < *eptr && isDIGIT(*s)) { + /* This could be unrolled like in grok_number(), but + * the expected uses of this are not speed-needy, and + * unlikely to need full 64-bitness. */ + const U8 digit = *s++ - '0'; + if (val < uv_max_div_10 || + (val == uv_max_div_10 && digit <= uv_max_mod_10)) { + val = val * 10 + digit; + } else { + return FALSE; } } } - if (s == pv) { - *eptr = NULL; /* If no progress, failed to parse anything. */ - return 0; + + if (endptr == NULL) { + if (*s) { + return FALSE; /* If endptr is NULL, no trailing non-digits allowed. */ + } } - if (endptr == NULL && *s) { - return 0; /* If endptr is NULL, no trailing non-digits allowed. */ + else { + *endptr = s; } - *eptr = s; - return val; + + *valptr = val; + return TRUE; } -#ifndef USE_QUADMATH +#ifndef Perl_strtod STATIC NV S_mulexp10(NV value, I32 exponent) { @@ -1214,11 +1262,11 @@ S_mulexp10(NV value, I32 exponent) /* On OpenVMS VAX we by default use the D_FLOAT double format, * and that format does not have *easy* capabilities [1] for - * overflowing doubles 'silently' as IEEE fp does. We also need - * to support G_FLOAT on both VAX and Alpha, and though the exponent - * range is much larger than D_FLOAT it still doesn't do silent - * overflow. Therefore we need to detect early whether we would - * overflow (this is the behaviour of the native string-to-float + * overflowing doubles 'silently' as IEEE fp does. We also need + * to support G_FLOAT on both VAX and Alpha, and though the exponent + * range is much larger than D_FLOAT it still doesn't do silent + * overflow. Therefore we need to detect early whether we would + * overflow (this is the behaviour of the native string-to-float * conversion routines, and therefore of native applications, too). * * [1] Trying to establish a condition handler to trap floating point @@ -1232,7 +1280,7 @@ S_mulexp10(NV value, I32 exponent) * a hammer. Therefore we need to catch potential overflows before * it's too late. */ -#if ((defined(VMS) && !defined(_IEEE_FP)) || defined(_UNICOS)) && defined(NV_MAX_10_EXP) +#if ((defined(VMS) && !defined(_IEEE_FP)) || defined(_UNICOS) || defined(DOUBLE_IS_VAX_FLOAT)) && defined(NV_MAX_10_EXP) STMT_START { const NV exp_v = log10(value); if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP) @@ -1279,10 +1327,14 @@ S_mulexp10(NV value, I32 exponent) result *= power; #ifdef FP_OVERFLOWS_TO_ZERO if (result == 0) +# ifdef NV_INF return value < 0 ? -NV_INF : NV_INF; +# else + return value < 0 ? -FLT_MAX : FLT_MAX; +# endif #endif /* Floating point exceptions are supposed to be turned off, - * but if we're obviously done, don't risk another iteration. + * but if we're obviously done, don't risk another iteration. */ if (exponent == 0) break; } @@ -1290,24 +1342,36 @@ S_mulexp10(NV value, I32 exponent) } return negative ? value / result : value * result; } -#endif /* #ifndef USE_QUADMATH */ +#endif /* #ifndef Perl_strtod */ + +#ifdef Perl_strtod +# define ATOF(s, x) my_atof2(s, &x) +#else +# define ATOF(s, x) Perl_atof2(s, x) +#endif NV Perl_my_atof(pTHX_ const char* s) { + /* 's' must be NUL terminated */ + NV x = 0.0; -#ifdef USE_QUADMATH - Perl_my_atof2(aTHX_ s, &x); - return x; -#else -# ifdef USE_LOCALE_NUMERIC + PERL_ARGS_ASSERT_MY_ATOF; +#if ! defined(USE_LOCALE_NUMERIC) + + ATOF(s, x); + +#else + { - DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED(); - if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) { - const char *standard = NULL, *local = NULL; - bool use_standard_radix; + DECLARATION_FOR_LC_NUMERIC_MANIPULATION; + STORE_LC_NUMERIC_SET_TO_NEEDED(); + if (! (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))) { + ATOF(s,x); + } + else { /* Look through the string for the first thing that looks like a * decimal point: either the value in the current locale or the @@ -1316,30 +1380,33 @@ Perl_my_atof(pTHX_ const char* s) * that we have to determine this beforehand because on some * systems, Perl_atof2 is just a wrapper around the system's atof. * */ - standard = strchr(s, '.'); - local = strstr(s, SvPV_nolen(PL_numeric_radix_sv)); - - use_standard_radix = standard && (!local || standard < local); + const char * const standard_pos = strchr(s, '.'); + const char * const local_pos + = strstr(s, SvPV_nolen(PL_numeric_radix_sv)); + const bool use_standard_radix + = standard_pos && (!local_pos || standard_pos < local_pos); - if (use_standard_radix) + if (use_standard_radix) { SET_NUMERIC_STANDARD(); + LOCK_LC_NUMERIC_STANDARD(); + } - Perl_atof2(s, x); + ATOF(s,x); - if (use_standard_radix) - SET_NUMERIC_LOCAL(); + if (use_standard_radix) { + UNLOCK_LC_NUMERIC_STANDARD(); + SET_NUMERIC_UNDERLYING(); + } } - else - Perl_atof2(s, x); RESTORE_LC_NUMERIC(); } -# else - Perl_atof2(s, x); -# endif + #endif + return x; } +#if defined(NV_INF) || defined(NV_NAN) #ifdef USING_MSVC6 # pragma warning(push) @@ -1350,7 +1417,7 @@ S_my_atof_infnan(pTHX_ const char* s, bool negative, const char* send, NV* value { const char *p0 = negative ? s - 1 : s; const char *p = p0; - int infnan = grok_infnan(&p, send); + const int infnan = grok_infnan(&p, send); if (infnan && p != p0) { /* If we can generate inf/nan directly, let's do so. */ #ifdef NV_INF @@ -1369,46 +1436,52 @@ S_my_atof_infnan(pTHX_ const char* s, bool negative, const char* send, NV* value /* If still here, we didn't have either NV_INF or NV_NAN, * and can try falling back to native strtod/strtold. * - * (Though, are our NV_INF or NV_NAN ever not defined?) - * * The native interface might not recognize all the possible * inf/nan strings Perl recognizes. What we can try * is to try faking the input. We will try inf/-inf/nan * as the most promising/portable input. */ { - const char* fake = NULL; + const char* fake = "silence compiler warning"; char* endp; NV nv; +#ifdef NV_INF if ((infnan & IS_NUMBER_INFINITY)) { fake = ((infnan & IS_NUMBER_NEG)) ? "-inf" : "inf"; } - else if ((infnan & IS_NUMBER_NAN)) { +#endif +#ifdef NV_NAN + if ((infnan & IS_NUMBER_NAN)) { fake = "nan"; } - assert(fake); - nv = Perl_strtod(fake, &endp); +#endif + assert(strNE(fake, "silence compiler warning")); + nv = S_strtod(aTHX_ fake, &endp); if (fake != endp) { +#ifdef NV_INF if ((infnan & IS_NUMBER_INFINITY)) { -#ifdef Perl_isinf +# ifdef Perl_isinf if (Perl_isinf(nv)) *value = nv; -#else +# else /* last resort, may generate SIGFPE */ *value = Perl_exp((NV)1e9); if ((infnan & IS_NUMBER_NEG)) *value = -*value; -#endif +# endif return (char*)p; /* p, not endp */ } - else if ((infnan & IS_NUMBER_NAN)) { -#ifdef Perl_isnan +#endif +#ifdef NV_NAN + if ((infnan & IS_NUMBER_NAN)) { +# ifdef Perl_isnan if (Perl_isnan(nv)) *value = nv; -#else +# else /* last resort, may generate SIGFPE */ *value = Perl_log((NV)-1.0); -#endif +# endif return (char*)p; /* p, not endp */ +#endif } } } @@ -1420,16 +1493,27 @@ S_my_atof_infnan(pTHX_ const char* s, bool negative, const char* send, NV* value # pragma warning(pop) #endif +#endif /* if defined(NV_INF) || defined(NV_NAN) */ + char* Perl_my_atof2(pTHX_ const char* orig, NV* value) { + PERL_ARGS_ASSERT_MY_ATOF2; + return my_atof3(orig, value, 0); +} + +char* +Perl_my_atof3(pTHX_ const char* orig, NV* value, const STRLEN len) +{ const char* s = orig; NV result[3] = {0.0, 0.0, 0.0}; -#if defined(USE_PERL_ATOF) || defined(USE_QUADMATH) - const char* send = s + strlen(orig); /* one past the last */ +#if defined(USE_PERL_ATOF) || defined(Perl_strtod) + const char* send = s + ((len != 0) + ? len + : strlen(orig)); /* one past the last */ bool negative = 0; #endif -#if defined(USE_PERL_ATOF) && !defined(USE_QUADMATH) +#if defined(USE_PERL_ATOF) && !defined(Perl_strtod) UV accumulator[2] = {0,0}; /* before/after dp */ bool seen_digit = 0; I32 exp_adjust[2] = {0,0}; @@ -1442,11 +1526,11 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) I32 sig_digits = 0; /* noof significant digits seen so far */ #endif -#if defined(USE_PERL_ATOF) || defined(USE_QUADMATH) - PERL_ARGS_ASSERT_MY_ATOF2; +#if defined(USE_PERL_ATOF) || defined(Perl_strtod) + PERL_ARGS_ASSERT_MY_ATOF3; /* leading whitespace */ - while (isSPACE(*s)) + while (s < send && isSPACE(*s)) ++s; /* sign */ @@ -1459,12 +1543,34 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) } #endif -#ifdef USE_QUADMATH +#ifdef Perl_strtod { char* endp; - if ((endp = S_my_atof_infnan(s, negative, send, value))) + char* copy = NULL; + + if ((endp = S_my_atof_infnan(aTHX_ s, negative, send, value))) return endp; - result[2] = strtoflt128(s, &endp); + + /* If the length is passed in, the input string isn't NUL-terminated, + * and in it turns out the function below assumes it is; therefore we + * create a copy and NUL-terminate that */ + if (len) { + Newx(copy, len + 1, char); + Copy(orig, copy, len, char); + copy[len] = '\0'; + s = copy + (s - orig); + } + + result[2] = S_strtod(aTHX_ s, &endp); + + /* If we created a copy, 'endp' is in terms of that. Convert back to + * the original */ + if (copy) { + s = (s - copy) + (char *) orig; + endp = (endp - copy) + (char *) orig; + Safefree(copy); + } + if (s != endp) { *value = negative ? -result[2] : result[2]; return endp; @@ -1502,16 +1608,18 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) /* the max number we can accumulate in a UV, and still safely do 10*N+9 */ #define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10)) +#if defined(NV_INF) || defined(NV_NAN) { - const char* endp; + char* endp; if ((endp = S_my_atof_infnan(aTHX_ s, negative, send, value))) - return (char*)endp; + return endp; } +#endif /* we accumulate digits into an integer; when this becomes too * large, we add the total to NV and start again */ - while (1) { + while (s < send) { if (isDIGIT(*s)) { seen_digit = 1; old_digit = digit; @@ -1539,7 +1647,7 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) exp_adjust[0]++; } /* skip remaining digits */ - while (isDIGIT(*s)) { + while (s < send && isDIGIT(*s)) { ++s; if (! seen_dp) { exp_adjust[0]++; @@ -1563,9 +1671,9 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) else if (!seen_dp && GROK_NUMERIC_RADIX(&s, send)) { seen_dp = 1; if (sig_digits > MAX_SIG_DIGITS) { - do { + while (s < send && isDIGIT(*s)) { ++s; - } while (isDIGIT(*s)); + } break; } } @@ -1579,7 +1687,7 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1]; } - if (seen_digit && (isALPHA_FOLD_EQ(*s, 'e'))) { + if (s < send && seen_digit && (isALPHA_FOLD_EQ(*s, 'e'))) { bool expnegative = 0; ++s; @@ -1590,14 +1698,12 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) case '+': ++s; } - while (isDIGIT(*s)) + while (s < send && isDIGIT(*s)) exponent = exponent * 10 + (*s++ - '0'); if (expnegative) exponent = -exponent; } - - /* now apply the exponent */ if (seen_dp) { @@ -1618,9 +1724,9 @@ Perl_my_atof2(pTHX_ const char* orig, NV* value) /* =for apidoc isinfnan -Perl_isinfnan() is utility function that returns true if the NV -argument is either an infinity or a NaN, false otherwise. To test -in more detail, use Perl_isinf() and Perl_isnan(). +C is utility function that returns true if the NV +argument is either an infinity or a C, false otherwise. To test +in more detail, use C and C. This is also the logical inverse of Perl_isfinite(). @@ -1629,6 +1735,7 @@ This is also the logical inverse of Perl_isfinite(). bool Perl_isinfnan(NV nv) { + PERL_UNUSED_ARG(nv); #ifdef Perl_isinf if (Perl_isinf(nv)) return TRUE; @@ -1643,9 +1750,9 @@ Perl_isinfnan(NV nv) /* =for apidoc -Checks whether the argument would be either an infinity or NaN when used +Checks whether the argument would be either an infinity or C when used as a number, but is careful not to trigger non-numeric or uninitialized -warnings. it assumes the caller has done SvGETMAGIC(sv) already. +warnings. it assumes the caller has done C already. =cut */ @@ -1701,21 +1808,21 @@ Perl_my_frexpl(long double x, int *e) { =for apidoc Perl_signbit Return a non-zero integer if the sign bit on an NV is set, and 0 if -it is not. +it is not. -If Configure detects this system has a signbit() that will work with -our NVs, then we just use it via the #define in perl.h. Otherwise, +If F detects this system has a C that will work with +our NVs, then we just use it via the C<#define> in F. Otherwise, fall back on this implementation. The main use of this function -is catching -0.0. +is catching C<-0.0>. -Configure notes: This function is called 'Perl_signbit' instead of a -plain 'signbit' because it is easy to imagine a system having a signbit() +C notes: This function is called C<'Perl_signbit'> instead of a +plain C<'signbit'> because it is easy to imagine a system having a C function or macro that doesn't happen to work with our particular choice -of NVs. We shouldn't just re-#define signbit as Perl_signbit and expect +of NVs. We shouldn't just re-C<#define> C as C and expect the standard system headers to be happy. Also, this is a no-context -function (no pTHX_) because Perl_signbit() is usually re-#defined in -perl.h as a simple macro call to the system's signbit(). -Users should just always call Perl_signbit(). +function (no C) because C is usually re-C<#defined> in +F as a simple macro call to the system's C. +Users should just always call C. =cut */ @@ -1723,19 +1830,25 @@ Users should just always call Perl_signbit(). int Perl_signbit(NV x) { # ifdef Perl_fp_class_nzero - if (x == 0) - return Perl_fp_class_nzero(x); -# endif + return Perl_fp_class_nzero(x); + /* Try finding the high byte, and assume it's highest bit + * is the sign. This assumption is probably wrong somewhere. */ +# elif defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN + return (((unsigned char *)&x)[9] & 0x80); +# elif defined(NV_LITTLE_ENDIAN) + /* Note that NVSIZE is sizeof(NV), which would make the below be + * wrong if the end bytes are unused, which happens with the x86 + * 80-bit long doubles, which is why take care of that above. */ + return (((unsigned char *)&x)[NVSIZE - 1] & 0x80); +# elif defined(NV_BIG_ENDIAN) + return (((unsigned char *)&x)[0] & 0x80); +# else + /* This last resort fallback is wrong for the negative zero. */ return (x < 0.0) ? 1 : 0; +# endif } #endif /* - * Local variables: - * c-indentation-style: bsd - * c-basic-offset: 4 - * indent-tabs-mode: nil - * End: - * * ex: set ts=8 sts=4 sw=4 et: */