/* numeric.c
*
- * Copyright (c) 2001, Larry Wall
+ * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
+ * 2002, 2003, 2004, 2005, 2006, 2007, by Larry Wall and others
*
* You may distribute under the terms of either the GNU General Public
* License or the Artistic License, as specified in the README file.
* wizards count differently to other people."
*/
+/*
+=head1 Numeric functions
+
+This file contains all the stuff needed by perl for manipulating numeric
+values, including such things as replacements for the OS's atof() function
+
+=cut
+
+*/
+
#include "EXTERN.h"
#define PERL_IN_NUMERIC_C
#include "perl.h"
U32
Perl_cast_ulong(pTHX_ NV f)
{
+ PERL_UNUSED_CONTEXT;
if (f < 0.0)
return f < I32_MIN ? (U32) I32_MIN : (U32)(I32) f;
if (f < U32_MAX_P1) {
I32
Perl_cast_i32(pTHX_ NV f)
{
+ PERL_UNUSED_CONTEXT;
if (f < I32_MAX_P1)
return f < I32_MIN ? I32_MIN : (I32) f;
if (f < U32_MAX_P1) {
IV
Perl_cast_iv(pTHX_ NV f)
{
+ PERL_UNUSED_CONTEXT;
if (f < IV_MAX_P1)
return f < IV_MIN ? IV_MIN : (IV) f;
if (f < UV_MAX_P1) {
UV
Perl_cast_uv(pTHX_ NV f)
{
+ PERL_UNUSED_CONTEXT;
if (f < 0.0)
return f < IV_MIN ? (UV) IV_MIN : (UV)(IV) f;
if (f < UV_MAX_P1) {
return f > 0 ? UV_MAX : 0 /* NaN */;
}
-#if defined(HUGE_VAL) || (defined(USE_LONG_DOUBLE) && defined(HUGE_VALL))
/*
- * This hack is to force load of "huge" support from libm.a
- * So it is in perl for (say) POSIX to use.
- * Needed for SunOS with Sun's 'acc' for example.
+=for apidoc grok_bin
+
+converts a string representing a binary number to numeric form.
+
+On entry I<start> and I<*len> give the string to scan, I<*flags> gives
+conversion flags, and I<result> should be NULL or a pointer to an NV.
+The scan stops at the end of the string, or the first invalid character.
+Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*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.
+
+If the value is <= C<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<grok_bin>
+returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
+and writes the value to I<*result> (or the value is discarded if I<result>
+is NULL).
+
+The binary number may optionally be prefixed with "0b" or "b" unless
+C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If
+C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the binary
+number may use '_' characters to separate digits.
+
+=cut
*/
-NV
-Perl_huge(void)
-{
-# if defined(USE_LONG_DOUBLE) && defined(HUGE_VALL)
- return HUGE_VALL;
-# endif
- return HUGE_VAL;
-}
-#endif
-NV
-Perl_scan_bin(pTHX_ char *start, STRLEN len, STRLEN *retlen)
-{
- register char *s = start;
- register NV rnv = 0.0;
- register UV ruv = 0;
- register bool seenb = FALSE;
- register bool overflowed = FALSE;
+UV
+Perl_grok_bin(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) {
+ const char *s = start;
+ STRLEN len = *len_p;
+ UV value = 0;
+ NV value_nv = 0;
- for (; len-- && *s; s++) {
- if (!(*s == '0' || *s == '1')) {
- if (*s == '_' && len && *retlen
- && (s[1] == '0' || s[1] == '1'))
- {
- --len;
- ++s;
- }
- else if (seenb == FALSE && *s == 'b' && ruv == 0) {
- /* Disallow 0bbb0b0bbb... */
- seenb = TRUE;
- continue;
- }
- else {
- if (ckWARN(WARN_DIGIT))
- Perl_warner(aTHX_ WARN_DIGIT,
- "Illegal binary digit '%c' ignored", *s);
- break;
- }
- }
- if (!overflowed) {
- register UV xuv = ruv << 1;
-
- if ((xuv >> 1) != ruv) {
- overflowed = TRUE;
- rnv = (NV) ruv;
- if (ckWARN_d(WARN_OVERFLOW))
- Perl_warner(aTHX_ WARN_OVERFLOW,
- "Integer overflow in binary number");
- }
- else
- ruv = xuv | (*s - '0');
- }
- if (overflowed) {
- rnv *= 2;
+ const UV max_div_2 = UV_MAX / 2;
+ const bool allow_underscores = (bool)(*flags & PERL_SCAN_ALLOW_UNDERSCORES);
+ bool overflowed = FALSE;
+ char bit;
+
+ if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
+ /* strip off leading b or 0b.
+ for compatibility silently suffer "b" and "0b" as valid binary
+ numbers. */
+ if (len >= 1) {
+ if (s[0] == 'b') {
+ s++;
+ len--;
+ }
+ else if (len >= 2 && s[0] == '0' && s[1] == 'b') {
+ s+=2;
+ len-=2;
+ }
+ }
+ }
+
+ for (; len-- && (bit = *s); s++) {
+ if (bit == '0' || bit == '1') {
+ /* Write it in this wonky order with a goto to attempt to get the
+ compiler to make the common case integer-only loop pretty tight.
+ With gcc seems to be much straighter code than old scan_bin. */
+ redo:
+ if (!overflowed) {
+ if (value <= max_div_2) {
+ value = (value << 1) | (bit - '0');
+ continue;
+ }
+ /* Bah. We're just overflowed. */
+ if (ckWARN_d(WARN_OVERFLOW))
+ Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
+ "Integer overflow in binary number");
+ overflowed = TRUE;
+ value_nv = (NV) value;
+ }
+ value_nv *= 2.0;
/* If an NV has not enough bits in its mantissa to
- * represent an UV this summing of small low-order numbers
+ * represent a UV this summing of small low-order numbers
* is a waste of time (because the NV cannot preserve
* the low-order bits anyway): we could just remember when
- * did we overflow and in the end just multiply rnv by the
+ * did we overflow and in the end just multiply value_nv by the
* right amount. */
- rnv += (*s - '0');
- }
+ value_nv += (NV)(bit - '0');
+ continue;
+ }
+ if (bit == '_' && len && allow_underscores && (bit = s[1])
+ && (bit == '0' || bit == '1'))
+ {
+ --len;
+ ++s;
+ goto redo;
+ }
+ if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
+ Perl_warner(aTHX_ packWARN(WARN_DIGIT),
+ "Illegal binary digit '%c' ignored", *s);
+ break;
}
- if (!overflowed)
- rnv = (NV) ruv;
- if ( ( overflowed && rnv > 4294967295.0)
+
+ if ( ( overflowed && value_nv > 4294967295.0)
#if UVSIZE > 4
- || (!overflowed && ruv > 0xffffffff )
+ || (!overflowed && value > 0xffffffff )
#endif
) {
if (ckWARN(WARN_PORTABLE))
- Perl_warner(aTHX_ WARN_PORTABLE,
+ Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
"Binary number > 0b11111111111111111111111111111111 non-portable");
}
- *retlen = s - start;
- return rnv;
+ *len_p = s - start;
+ if (!overflowed) {
+ *flags = 0;
+ return value;
+ }
+ *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
+ if (result)
+ *result = value_nv;
+ return UV_MAX;
}
-NV
-Perl_scan_oct(pTHX_ char *start, STRLEN len, STRLEN *retlen)
-{
- register char *s = start;
- register NV rnv = 0.0;
- register UV ruv = 0;
- register bool overflowed = FALSE;
+/*
+=for apidoc grok_hex
+
+converts a string representing a hex number to numeric form.
+
+On entry I<start> and I<*len> give the string to scan, I<*flags> gives
+conversion flags, and I<result> should be NULL or a pointer to an NV.
+The scan stops at the end of the string, or the first invalid character.
+Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*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.
+
+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<grok_hex>
+returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
+and writes the value to I<*result> (or the value is discarded if I<result>
+is NULL).
+
+The hex number may optionally be prefixed with "0x" or "x" unless
+C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If
+C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the hex
+number may use '_' characters to separate digits.
+
+=cut
+ */
+
+UV
+Perl_grok_hex(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) {
+ dVAR;
+ const char *s = start;
+ STRLEN len = *len_p;
+ UV value = 0;
+ NV value_nv = 0;
+
+ const UV max_div_16 = UV_MAX / 16;
+ const bool allow_underscores = (bool)(*flags & PERL_SCAN_ALLOW_UNDERSCORES);
+ bool overflowed = FALSE;
+
+ if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
+ /* strip off leading x or 0x.
+ for compatibility silently suffer "x" and "0x" as valid hex numbers.
+ */
+ if (len >= 1) {
+ if (s[0] == 'x') {
+ s++;
+ len--;
+ }
+ else if (len >= 2 && s[0] == '0' && s[1] == 'x') {
+ s+=2;
+ len-=2;
+ }
+ }
+ }
for (; len-- && *s; s++) {
- if (!(*s >= '0' && *s <= '7')) {
- if (*s == '_' && len && *retlen
- && (s[1] >= '0' && s[1] <= '7'))
+ const char *hexdigit = strchr(PL_hexdigit, *s);
+ if (hexdigit) {
+ /* Write it in this wonky order with a goto to attempt to get the
+ compiler to make the common case integer-only loop pretty tight.
+ With gcc seems to be much straighter code than old scan_hex. */
+ redo:
+ if (!overflowed) {
+ if (value <= max_div_16) {
+ value = (value << 4) | ((hexdigit - PL_hexdigit) & 15);
+ continue;
+ }
+ /* Bah. We're just overflowed. */
+ if (ckWARN_d(WARN_OVERFLOW))
+ Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
+ "Integer overflow in hexadecimal number");
+ overflowed = TRUE;
+ value_nv = (NV) value;
+ }
+ value_nv *= 16.0;
+ /* If an NV has not enough bits in its mantissa to
+ * represent a UV this summing of small low-order numbers
+ * is a waste of time (because the NV cannot preserve
+ * the low-order bits anyway): we could just remember when
+ * did we overflow and in the end just multiply value_nv by the
+ * right amount of 16-tuples. */
+ value_nv += (NV)((hexdigit - PL_hexdigit) & 15);
+ continue;
+ }
+ if (*s == '_' && len && allow_underscores && s[1]
+ && (hexdigit = strchr(PL_hexdigit, s[1])))
{
--len;
++s;
+ goto redo;
}
- else {
- /* Allow \octal to work the DWIM way (that is, stop scanning
- * as soon as non-octal characters are seen, complain only iff
- * someone seems to want to use the digits eight and nine). */
- if (*s == '8' || *s == '9') {
- if (ckWARN(WARN_DIGIT))
- Perl_warner(aTHX_ WARN_DIGIT,
- "Illegal octal digit '%c' ignored", *s);
- }
- break;
- }
- }
- if (!overflowed) {
- register UV xuv = ruv << 3;
-
- if ((xuv >> 3) != ruv) {
- overflowed = TRUE;
- rnv = (NV) ruv;
- if (ckWARN_d(WARN_OVERFLOW))
- Perl_warner(aTHX_ WARN_OVERFLOW,
- "Integer overflow in octal number");
- }
- else
- ruv = xuv | (*s - '0');
- }
- if (overflowed) {
- rnv *= 8.0;
- /* If an NV has not enough bits in its mantissa to
- * represent an UV this summing of small low-order numbers
- * is a waste of time (because the NV cannot preserve
- * the low-order bits anyway): we could just remember when
- * did we overflow and in the end just multiply rnv by the
- * right amount of 8-tuples. */
- rnv += (NV)(*s - '0');
- }
+ if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
+ Perl_warner(aTHX_ packWARN(WARN_DIGIT),
+ "Illegal hexadecimal digit '%c' ignored", *s);
+ break;
}
- if (!overflowed)
- rnv = (NV) ruv;
- if ( ( overflowed && rnv > 4294967295.0)
+
+ if ( ( overflowed && value_nv > 4294967295.0)
#if UVSIZE > 4
- || (!overflowed && ruv > 0xffffffff )
+ || (!overflowed && value > 0xffffffff )
#endif
) {
if (ckWARN(WARN_PORTABLE))
- Perl_warner(aTHX_ WARN_PORTABLE,
- "Octal number > 037777777777 non-portable");
+ Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
+ "Hexadecimal number > 0xffffffff non-portable");
}
- *retlen = s - start;
- return rnv;
+ *len_p = s - start;
+ if (!overflowed) {
+ *flags = 0;
+ return value;
+ }
+ *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
+ if (result)
+ *result = value_nv;
+ return UV_MAX;
}
-NV
-Perl_scan_hex(pTHX_ char *start, STRLEN len, STRLEN *retlen)
-{
- register char *s = start;
- register NV rnv = 0.0;
- register UV ruv = 0;
- register bool overflowed = FALSE;
- char *hexdigit;
-
- if (len > 2) {
- if (s[0] == 'x') {
- s++;
- len--;
- }
- else if (len > 3 && s[0] == '0' && s[1] == 'x') {
- s+=2;
- len-=2;
- }
- }
+/*
+=for apidoc grok_oct
+
+converts a string representing an octal number to numeric form.
+
+On entry I<start> and I<*len> give the string to scan, I<*flags> gives
+conversion flags, and I<result> should be NULL or a pointer to an NV.
+The scan stops at the end of the string, or the first invalid character.
+Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*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.
+
+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<grok_oct>
+returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
+and writes the value to I<*result> (or the value is discarded if I<result>
+is NULL).
+
+If C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the octal
+number may use '_' characters to separate digits.
+
+=cut
+ */
+
+UV
+Perl_grok_oct(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) {
+ const char *s = start;
+ STRLEN len = *len_p;
+ UV value = 0;
+ NV value_nv = 0;
+
+ const UV max_div_8 = UV_MAX / 8;
+ const bool allow_underscores = (bool)(*flags & PERL_SCAN_ALLOW_UNDERSCORES);
+ bool overflowed = FALSE;
for (; len-- && *s; s++) {
- hexdigit = strchr((char *) PL_hexdigit, *s);
- if (!hexdigit) {
- if (*s == '_' && len && *retlen && s[1]
- && (hexdigit = strchr((char *) PL_hexdigit, s[1])))
+ /* gcc 2.95 optimiser not smart enough to figure that this subtraction
+ out front allows slicker code. */
+ int digit = *s - '0';
+ if (digit >= 0 && digit <= 7) {
+ /* Write it in this wonky order with a goto to attempt to get the
+ compiler to make the common case integer-only loop pretty tight.
+ */
+ redo:
+ if (!overflowed) {
+ if (value <= max_div_8) {
+ value = (value << 3) | digit;
+ continue;
+ }
+ /* Bah. We're just overflowed. */
+ if (ckWARN_d(WARN_OVERFLOW))
+ Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
+ "Integer overflow in octal number");
+ overflowed = TRUE;
+ value_nv = (NV) value;
+ }
+ value_nv *= 8.0;
+ /* If an NV has not enough bits in its mantissa to
+ * represent a UV this summing of small low-order numbers
+ * is a waste of time (because the NV cannot preserve
+ * the low-order bits anyway): we could just remember when
+ * did we overflow and in the end just multiply value_nv by the
+ * right amount of 8-tuples. */
+ value_nv += (NV)digit;
+ continue;
+ }
+ if (digit == ('_' - '0') && len && allow_underscores
+ && (digit = s[1] - '0') && (digit >= 0 && digit <= 7))
{
--len;
++s;
+ goto redo;
}
- else {
- if (ckWARN(WARN_DIGIT))
- Perl_warner(aTHX_ WARN_DIGIT,
- "Illegal hexadecimal digit '%c' ignored", *s);
- break;
- }
- }
- if (!overflowed) {
- register UV xuv = ruv << 4;
-
- if ((xuv >> 4) != ruv) {
- overflowed = TRUE;
- rnv = (NV) ruv;
- if (ckWARN_d(WARN_OVERFLOW))
- Perl_warner(aTHX_ WARN_OVERFLOW,
- "Integer overflow in hexadecimal number");
- }
- else
- ruv = xuv | ((hexdigit - PL_hexdigit) & 15);
- }
- if (overflowed) {
- rnv *= 16.0;
- /* If an NV has not enough bits in its mantissa to
- * represent an UV this summing of small low-order numbers
- * is a waste of time (because the NV cannot preserve
- * the low-order bits anyway): we could just remember when
- * did we overflow and in the end just multiply rnv by the
- * right amount of 16-tuples. */
- rnv += (NV)((hexdigit - PL_hexdigit) & 15);
- }
+ /* Allow \octal to work the DWIM way (that is, stop scanning
+ * as soon as non-octal characters are seen, complain only if
+ * someone seems to want to use the digits eight and nine). */
+ if (digit == 8 || digit == 9) {
+ if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
+ Perl_warner(aTHX_ packWARN(WARN_DIGIT),
+ "Illegal octal digit '%c' ignored", *s);
+ }
+ break;
}
- if (!overflowed)
- rnv = (NV) ruv;
- if ( ( overflowed && rnv > 4294967295.0)
+
+ if ( ( overflowed && value_nv > 4294967295.0)
#if UVSIZE > 4
- || (!overflowed && ruv > 0xffffffff )
+ || (!overflowed && value > 0xffffffff )
#endif
) {
if (ckWARN(WARN_PORTABLE))
- Perl_warner(aTHX_ WARN_PORTABLE,
- "Hexadecimal number > 0xffffffff non-portable");
+ Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
+ "Octal number > 037777777777 non-portable");
+ }
+ *len_p = s - start;
+ if (!overflowed) {
+ *flags = 0;
+ return value;
}
- *retlen = s - start;
- return rnv;
+ *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
+ if (result)
+ *result = value_nv;
+ return UV_MAX;
+}
+
+/*
+=for apidoc scan_bin
+
+For backwards compatibility. Use C<grok_bin> instead.
+
+=for apidoc scan_hex
+
+For backwards compatibility. Use C<grok_hex> instead.
+
+=for apidoc scan_oct
+
+For backwards compatibility. Use C<grok_oct> instead.
+
+=cut
+ */
+
+NV
+Perl_scan_bin(pTHX_ const char *start, STRLEN len, STRLEN *retlen)
+{
+ NV rnv;
+ I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
+ const UV ruv = grok_bin (start, &len, &flags, &rnv);
+
+ *retlen = len;
+ return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
+}
+
+NV
+Perl_scan_oct(pTHX_ const char *start, STRLEN len, STRLEN *retlen)
+{
+ NV rnv;
+ I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
+ const UV ruv = grok_oct (start, &len, &flags, &rnv);
+
+ *retlen = len;
+ return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
+}
+
+NV
+Perl_scan_hex(pTHX_ const char *start, STRLEN len, STRLEN *retlen)
+{
+ NV rnv;
+ I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
+ const UV ruv = grok_hex (start, &len, &flags, &rnv);
+
+ *retlen = len;
+ return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
}
/*
Perl_grok_numeric_radix(pTHX_ const char **sp, const char *send)
{
#ifdef USE_LOCALE_NUMERIC
+ dVAR;
if (PL_numeric_radix_sv && IN_LOCALE) {
STRLEN len;
- char* radix = SvPV(PL_numeric_radix_sv, len);
+ const char * const radix = SvPV(PL_numeric_radix_sv, len);
if (*sp + len <= send && memEQ(*sp, radix, len)) {
*sp += len;
return TRUE;
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 (defined in perl.h).
+IS_NUMBER_NEG, IS_NUMBER_INFINITY, IS_NUMBER_NAN (defined in perl.h).
If the value of the number can fit an in UV, it is returned in the *valuep
IS_NUMBER_IN_UV will be set to indicate that *valuep is valid, IS_NUMBER_IN_UV
Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep)
{
const char *s = pv;
- const char *send = pv + len;
+ const char * const send = pv + len;
const UV max_div_10 = UV_MAX / 10;
const char max_mod_10 = UV_MAX % 10;
int numtype = 0;
int sawinf = 0;
+ int sawnan = 0;
while (s < send && isSPACE(*s))
s++;
With it gcc on arm is managing 6 instructions (6 cycles) per digit.
In theory the optimiser could deduce how far to unroll the loop
before checking for overflow. */
- if (s < send) {
- int digit = *++s - '0';
+ if (++s < send) {
+ int digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
- digit = *++s - '0';
+ if (++s < send) {
+ digit = *s - '0';
if (digit >= 0 && digit <= 9) {
value = value * 10 + digit;
- if (s < send) {
+ if (++s < send) {
/* Now got 9 digits, so need to check
each time for overflow. */
- digit = *++s - '0';
+ digit = *s - '0';
while (digit >= 0 && digit <= 9
&& (value < max_div_10
|| (value == max_div_10
&& digit <= max_mod_10))) {
value = value * 10 + digit;
- if (s < send)
- digit = *++s - '0';
+ if (++s < send)
+ digit = *s - '0';
else
break;
}
if (digit >= 0 && digit <= 9
- && !(s < send)) {
+ && (s < send)) {
/* value overflowed.
skip the remaining digits, don't
worry about setting *valuep. */
s++;
}
sawinf = 1;
- } else /* Add test for NaN here. */
+ } else if (*s == 'N' || *s == 'n') {
+ /* XXX TODO: There are signaling NaNs and quiet NaNs. */
+ s++; if (s == send || (*s != 'A' && *s != 'a')) return 0;
+ s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
+ s++;
+ sawnan = 1;
+ } else
return 0;
if (sawinf) {
numtype &= IS_NUMBER_NEG; /* Keep track of sign */
numtype |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT;
+ } else if (sawnan) {
+ numtype &= IS_NUMBER_NEG; /* Keep track of sign */
+ numtype |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT;
} else if (s < send) {
/* we can have an optional exponent part */
if (*s == 'e' || *s == 'E') {
while (s < send && isSPACE(*s))
s++;
if (s >= send)
- return numtype;
+ return numtype;
if (len == 10 && memEQ(pv, "0 but true", 10)) {
if (valuep)
*valuep = 0;
return 0;
}
-NV
+STATIC NV
S_mulexp10(NV value, I32 exponent)
{
NV result = 1.0;
if (exponent == 0)
return value;
- else if (exponent < 0) {
+ if (value == 0)
+ return (NV)0;
+
+ /* 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
+ * conversion routines, and therefore of native applications, too).
+ *
+ * [1] Trying to establish a condition handler to trap floating point
+ * exceptions is not a good idea. */
+
+ /* In UNICOS and in certain Cray models (such as T90) there is no
+ * IEEE fp, and no way at all from C to catch fp overflows gracefully.
+ * There is something you can do if you are willing to use some
+ * inline assembler: the instruction is called DFI-- but that will
+ * disable *all* floating point interrupts, a little bit too large
+ * 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)
+ STMT_START {
+ const NV exp_v = log10(value);
+ if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP)
+ return NV_MAX;
+ if (exponent < 0) {
+ if (-(exponent + exp_v) >= NV_MAX_10_EXP)
+ return 0.0;
+ while (-exponent >= NV_MAX_10_EXP) {
+ /* combination does not overflow, but 10^(-exponent) does */
+ value /= 10;
+ ++exponent;
+ }
+ }
+ } STMT_END;
+#endif
+
+ if (exponent < 0) {
negative = 1;
exponent = -exponent;
}
if (exponent & bit) {
exponent ^= bit;
result *= power;
+ /* Floating point exceptions are supposed to be turned off,
+ * but if we're obviously done, don't risk another iteration.
+ */
+ if (exponent == 0) break;
}
power *= power;
}
{
NV x = 0.0;
#ifdef USE_LOCALE_NUMERIC
+ dVAR;
if (PL_numeric_local && IN_LOCALE) {
NV y;
/* Scan the number twice; once using locale and once without;
* choose the larger result (in absolute value). */
- Perl_atof2(aTHX_ s, &x);
+ Perl_atof2(s, x);
SET_NUMERIC_STANDARD();
- Perl_atof2(aTHX_ s, &y);
+ Perl_atof2(s, y);
SET_NUMERIC_LOCAL();
if ((y < 0.0 && y < x) || (y > 0.0 && y > x))
return y;
}
else
- Perl_atof2(aTHX_ s, &x);
+ Perl_atof2(s, x);
#else
- Perl_atof2(aTHX_ s, &x);
+ Perl_atof2(s, x);
#endif
return x;
}
char*
Perl_my_atof2(pTHX_ const char* orig, NV* value)
{
- NV result = 0.0;
+ NV result[3] = {0.0, 0.0, 0.0};
+ const char* s = orig;
+#ifdef USE_PERL_ATOF
+ UV accumulator[2] = {0,0}; /* before/after dp */
bool negative = 0;
- char* s = (char*)orig;
- char* send = s + strlen(orig) - 1;
- bool seendigit = 0;
- I32 expextra = 0;
+ const char* send = s + strlen(orig) - 1;
+ bool seen_digit = 0;
+ I32 exp_adjust[2] = {0,0};
+ I32 exp_acc[2] = {-1, -1};
+ /* the current exponent adjust for the accumulators */
I32 exponent = 0;
- I32 i;
-/* this is arbitrary */
-#define PARTLIM 6
-/* we want the largest integers we can usefully use */
-#if defined(HAS_QUAD) && defined(USE_64_BIT_INT)
-# define PARTSIZE ((int)TYPE_DIGITS(U64)-1)
- U64 part[PARTLIM];
-#else
-# define PARTSIZE ((int)TYPE_DIGITS(U32)-1)
- U32 part[PARTLIM];
-#endif
- I32 ipart = 0; /* index into part[] */
- I32 offcount; /* number of digits in least significant part */
+ I32 seen_dp = 0;
+ I32 digit = 0;
+ I32 old_digit = 0;
+ I32 sig_digits = 0; /* noof significant digits seen so far */
+
+/* There is no point in processing more significant digits
+ * than the NV can hold. Note that NV_DIG is a lower-bound value,
+ * while we need an upper-bound value. We add 2 to account for this;
+ * since it will have been conservative on both the first and last digit.
+ * For example a 32-bit mantissa with an exponent of 4 would have
+ * exact values in the set
+ * 4
+ * 8
+ * ..
+ * 17179869172
+ * 17179869176
+ * 17179869180
+ *
+ * where for the purposes of calculating NV_DIG we would have to discount
+ * both the first and last digit, since neither can hold all values from
+ * 0..9; but for calculating the value we must examine those two digits.
+ */
+#define MAX_SIG_DIGITS (NV_DIG+2)
+
+/* the max number we can accumulate in a UV, and still safely do 10*N+9 */
+#define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10))
+
+ /* leading whitespace */
+ while (isSPACE(*s))
+ ++s;
/* sign */
switch (*s) {
++s;
}
- part[0] = offcount = 0;
- if (isDIGIT(*s)) {
- seendigit = 1; /* get this over with */
+ /* punt to strtod for NaN/Inf; if no support for it there, tough luck */
- /* skip leading zeros */
- while (*s == '0')
- ++s;
+#ifdef HAS_STRTOD
+ if (*s == 'n' || *s == 'N' || *s == 'i' || *s == 'I') {
+ const char *p = negative ? s - 1 : s;
+ char *endp;
+ NV rslt;
+ rslt = strtod(p, &endp);
+ if (endp != p) {
+ *value = rslt;
+ return (char *)endp;
+ }
}
+#endif
- /* integer digits */
- while (isDIGIT(*s)) {
- if (++offcount > PARTSIZE) {
- if (++ipart < PARTLIM) {
- part[ipart] = 0;
- offcount = 1; /* ++0 */
- }
- else {
+ /* we accumulate digits into an integer; when this becomes too
+ * large, we add the total to NV and start again */
+
+ while (1) {
+ if (isDIGIT(*s)) {
+ seen_digit = 1;
+ old_digit = digit;
+ digit = *s++ - '0';
+ if (seen_dp)
+ exp_adjust[1]++;
+
+ /* don't start counting until we see the first significant
+ * digit, eg the 5 in 0.00005... */
+ if (!sig_digits && digit == 0)
+ continue;
+
+ if (++sig_digits > MAX_SIG_DIGITS) {
/* limits of precision reached */
- --ipart;
- --offcount;
- if (*s >= '5')
- ++part[ipart];
+ if (digit > 5) {
+ ++accumulator[seen_dp];
+ } else if (digit == 5) {
+ if (old_digit % 2) { /* round to even - Allen */
+ ++accumulator[seen_dp];
+ }
+ }
+ if (seen_dp) {
+ exp_adjust[1]--;
+ } else {
+ exp_adjust[0]++;
+ }
+ /* skip remaining digits */
while (isDIGIT(*s)) {
- ++expextra;
++s;
+ if (! seen_dp) {
+ exp_adjust[0]++;
+ }
}
/* warn of loss of precision? */
- break;
}
- }
- part[ipart] = part[ipart] * 10 + (*s++ - '0');
- }
-
- /* decimal point */
- if (GROK_NUMERIC_RADIX((const char **)&s, send)) {
- if (isDIGIT(*s))
- seendigit = 1; /* get this over with */
-
- /* decimal digits */
- while (isDIGIT(*s)) {
- if (++offcount > PARTSIZE) {
- if (++ipart < PARTLIM) {
- part[ipart] = 0;
- offcount = 1; /* ++0 */
- }
- else {
- /* limits of precision reached */
- --ipart;
- --offcount;
- if (*s >= '5')
- ++part[ipart];
- while (isDIGIT(*s))
- ++s;
- /* warn of loss of precision? */
- break;
+ else {
+ if (accumulator[seen_dp] > MAX_ACCUMULATE) {
+ /* add accumulator to result and start again */
+ result[seen_dp] = S_mulexp10(result[seen_dp],
+ exp_acc[seen_dp])
+ + (NV)accumulator[seen_dp];
+ accumulator[seen_dp] = 0;
+ exp_acc[seen_dp] = 0;
}
+ accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit;
+ ++exp_acc[seen_dp];
+ }
+ }
+ else if (!seen_dp && GROK_NUMERIC_RADIX(&s, send)) {
+ seen_dp = 1;
+ if (sig_digits > MAX_SIG_DIGITS) {
+ do {
+ ++s;
+ } while (isDIGIT(*s));
+ break;
}
- --expextra;
- part[ipart] = part[ipart] * 10 + (*s++ - '0');
+ }
+ else {
+ break;
}
}
- /* combine components of mantissa */
- for (i = 0; i <= ipart; ++i)
- result += S_mulexp10((NV)part[ipart - i],
- i ? offcount + (i - 1) * PARTSIZE : 0);
+ result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0];
+ if (seen_dp) {
+ result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1];
+ }
- if (seendigit && (*s == 'e' || *s == 'E')) {
+ if (seen_digit && (*s == 'e' || *s == 'E')) {
bool expnegative = 0;
++s;
exponent = -exponent;
}
+
+
/* now apply the exponent */
- exponent += expextra;
- result = S_mulexp10(result, exponent);
+
+ if (seen_dp) {
+ result[2] = S_mulexp10(result[0],exponent+exp_adjust[0])
+ + S_mulexp10(result[1],exponent-exp_adjust[1]);
+ } else {
+ result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]);
+ }
/* now apply the sign */
if (negative)
- result = -result;
- *value = result;
- return s;
+ result[2] = -result[2];
+#endif /* USE_PERL_ATOF */
+ *value = result[2];
+ return (char *)s;
+}
+
+#if ! defined(HAS_MODFL) && defined(HAS_AINTL) && defined(HAS_COPYSIGNL)
+long double
+Perl_my_modfl(long double x, long double *ip)
+{
+ *ip = aintl(x);
+ return (x == *ip ? copysignl(0.0L, x) : x - *ip);
}
+#endif
+#if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL)
+long double
+Perl_my_frexpl(long double x, int *e) {
+ *e = x == 0.0L ? 0 : ilogbl(x) + 1;
+ return (scalbnl(x, -*e));
+}
+#endif
+
+/*
+=for apidoc Perl_signbit
+
+Return a non-zero integer if the sign bit on an NV is set, and 0 if
+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,
+fall back on this implementation. As a first pass, this gets everything
+right except -0.0. Alas, catching -0.0 is the main use for this function,
+so this is not too helpful yet. Still, at least we have the scaffolding
+in place to support other systems, should that prove useful.
+
+
+Configure notes: This function is called 'Perl_signbit' instead of a
+plain 'signbit' because it is easy to imagine a system having a signbit()
+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
+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().
+
+=cut
+*/
+#if !defined(HAS_SIGNBIT)
+int
+Perl_signbit(NV x) {
+ return (x < 0.0) ? 1 : 0;
+}
+#endif
+
+/*
+ * Local variables:
+ * c-indentation-style: bsd
+ * c-basic-offset: 4
+ * indent-tabs-mode: t
+ * End:
+ *
+ * ex: set ts=8 sts=4 sw=4 noet:
+ */