This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
With mallocwrap New() evaluates its arguments more than once, so they
[perl5.git] / numeric.c
... / ...
CommitLineData
1/* numeric.c
2 *
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999,
4 * 2000, 2001, 2002, 2003, by Larry Wall and others
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
12 * "That only makes eleven (plus one mislaid) and not fourteen, unless
13 * wizards count differently to other people."
14 */
15
16/*
17=head1 Numeric functions
18
19This file contains all the stuff needed by perl for manipulating numeric
20values, including such things as replacements for the OS's atof() function
21
22=cut
23
24*/
25
26#include "EXTERN.h"
27#define PERL_IN_NUMERIC_C
28#include "perl.h"
29
30U32
31Perl_cast_ulong(pTHX_ NV f)
32{
33 if (f < 0.0)
34 return f < I32_MIN ? (U32) I32_MIN : (U32)(I32) f;
35 if (f < U32_MAX_P1) {
36#if CASTFLAGS & 2
37 if (f < U32_MAX_P1_HALF)
38 return (U32) f;
39 f -= U32_MAX_P1_HALF;
40 return ((U32) f) | (1 + U32_MAX >> 1);
41#else
42 return (U32) f;
43#endif
44 }
45 return f > 0 ? U32_MAX : 0 /* NaN */;
46}
47
48I32
49Perl_cast_i32(pTHX_ NV f)
50{
51 if (f < I32_MAX_P1)
52 return f < I32_MIN ? I32_MIN : (I32) f;
53 if (f < U32_MAX_P1) {
54#if CASTFLAGS & 2
55 if (f < U32_MAX_P1_HALF)
56 return (I32)(U32) f;
57 f -= U32_MAX_P1_HALF;
58 return (I32)(((U32) f) | (1 + U32_MAX >> 1));
59#else
60 return (I32)(U32) f;
61#endif
62 }
63 return f > 0 ? (I32)U32_MAX : 0 /* NaN */;
64}
65
66IV
67Perl_cast_iv(pTHX_ NV f)
68{
69 if (f < IV_MAX_P1)
70 return f < IV_MIN ? IV_MIN : (IV) f;
71 if (f < UV_MAX_P1) {
72#if CASTFLAGS & 2
73 /* For future flexibility allowing for sizeof(UV) >= sizeof(IV) */
74 if (f < UV_MAX_P1_HALF)
75 return (IV)(UV) f;
76 f -= UV_MAX_P1_HALF;
77 return (IV)(((UV) f) | (1 + UV_MAX >> 1));
78#else
79 return (IV)(UV) f;
80#endif
81 }
82 return f > 0 ? (IV)UV_MAX : 0 /* NaN */;
83}
84
85UV
86Perl_cast_uv(pTHX_ NV f)
87{
88 if (f < 0.0)
89 return f < IV_MIN ? (UV) IV_MIN : (UV)(IV) f;
90 if (f < UV_MAX_P1) {
91#if CASTFLAGS & 2
92 if (f < UV_MAX_P1_HALF)
93 return (UV) f;
94 f -= UV_MAX_P1_HALF;
95 return ((UV) f) | (1 + UV_MAX >> 1);
96#else
97 return (UV) f;
98#endif
99 }
100 return f > 0 ? UV_MAX : 0 /* NaN */;
101}
102
103#if defined(HUGE_VAL) || (defined(USE_LONG_DOUBLE) && defined(HUGE_VALL))
104/*
105 * This hack is to force load of "huge" support from libm.a
106 * So it is in perl for (say) POSIX to use.
107 * Needed for SunOS with Sun's 'acc' for example.
108 */
109NV
110Perl_huge(void)
111{
112# if defined(USE_LONG_DOUBLE) && defined(HUGE_VALL)
113 return HUGE_VALL;
114# endif
115 return HUGE_VAL;
116}
117#endif
118
119/*
120=for apidoc grok_bin
121
122converts a string representing a binary number to numeric form.
123
124On entry I<start> and I<*len> give the string to scan, I<*flags> gives
125conversion flags, and I<result> should be NULL or a pointer to an NV.
126The scan stops at the end of the string, or the first invalid character.
127Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*flags>, encountering an
128invalid character will also trigger a warning.
129On return I<*len> is set to the length of the scanned string,
130and I<*flags> gives output flags.
131
132If the value is <= UV_MAX it is returned as a UV, the output flags are clear,
133and nothing is written to I<*result>. If the value is > UV_MAX C<grok_bin>
134returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
135and writes the value to I<*result> (or the value is discarded if I<result>
136is NULL).
137
138The binary number may optionally be prefixed with "0b" or "b" unless
139C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If
140C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the binary
141number may use '_' characters to separate digits.
142
143=cut
144 */
145
146UV
147Perl_grok_bin(pTHX_ char *start, STRLEN *len_p, I32 *flags, NV *result) {
148 const char *s = start;
149 STRLEN len = *len_p;
150 UV value = 0;
151 NV value_nv = 0;
152
153 const UV max_div_2 = UV_MAX / 2;
154 bool allow_underscores = *flags & PERL_SCAN_ALLOW_UNDERSCORES;
155 bool overflowed = FALSE;
156
157 if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
158 /* strip off leading b or 0b.
159 for compatibility silently suffer "b" and "0b" as valid binary
160 numbers. */
161 if (len >= 1) {
162 if (s[0] == 'b') {
163 s++;
164 len--;
165 }
166 else if (len >= 2 && s[0] == '0' && s[1] == 'b') {
167 s+=2;
168 len-=2;
169 }
170 }
171 }
172
173 for (; len-- && *s; s++) {
174 char bit = *s;
175 if (bit == '0' || bit == '1') {
176 /* Write it in this wonky order with a goto to attempt to get the
177 compiler to make the common case integer-only loop pretty tight.
178 With gcc seems to be much straighter code than old scan_bin. */
179 redo:
180 if (!overflowed) {
181 if (value <= max_div_2) {
182 value = (value << 1) | (bit - '0');
183 continue;
184 }
185 /* Bah. We're just overflowed. */
186 if (ckWARN_d(WARN_OVERFLOW))
187 Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
188 "Integer overflow in binary number");
189 overflowed = TRUE;
190 value_nv = (NV) value;
191 }
192 value_nv *= 2.0;
193 /* If an NV has not enough bits in its mantissa to
194 * represent a UV this summing of small low-order numbers
195 * is a waste of time (because the NV cannot preserve
196 * the low-order bits anyway): we could just remember when
197 * did we overflow and in the end just multiply value_nv by the
198 * right amount. */
199 value_nv += (NV)(bit - '0');
200 continue;
201 }
202 if (bit == '_' && len && allow_underscores && (bit = s[1])
203 && (bit == '0' || bit == '1'))
204 {
205 --len;
206 ++s;
207 goto redo;
208 }
209 if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
210 Perl_warner(aTHX_ packWARN(WARN_DIGIT),
211 "Illegal binary digit '%c' ignored", *s);
212 break;
213 }
214
215 if ( ( overflowed && value_nv > 4294967295.0)
216#if UVSIZE > 4
217 || (!overflowed && value > 0xffffffff )
218#endif
219 ) {
220 if (ckWARN(WARN_PORTABLE))
221 Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
222 "Binary number > 0b11111111111111111111111111111111 non-portable");
223 }
224 *len_p = s - start;
225 if (!overflowed) {
226 *flags = 0;
227 return value;
228 }
229 *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
230 if (result)
231 *result = value_nv;
232 return UV_MAX;
233}
234
235/*
236=for apidoc grok_hex
237
238converts a string representing a hex number to numeric form.
239
240On entry I<start> and I<*len> give the string to scan, I<*flags> gives
241conversion flags, and I<result> should be NULL or a pointer to an NV.
242The scan stops at the end of the string, or the first invalid character.
243Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*flags>, encountering an
244invalid character will also trigger a warning.
245On return I<*len> is set to the length of the scanned string,
246and I<*flags> gives output flags.
247
248If the value is <= UV_MAX it is returned as a UV, the output flags are clear,
249and nothing is written to I<*result>. If the value is > UV_MAX C<grok_hex>
250returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
251and writes the value to I<*result> (or the value is discarded if I<result>
252is NULL).
253
254The hex number may optionally be prefixed with "0x" or "x" unless
255C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If
256C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the hex
257number may use '_' characters to separate digits.
258
259=cut
260 */
261
262UV
263Perl_grok_hex(pTHX_ char *start, STRLEN *len_p, I32 *flags, NV *result) {
264 const char *s = start;
265 STRLEN len = *len_p;
266 UV value = 0;
267 NV value_nv = 0;
268
269 const UV max_div_16 = UV_MAX / 16;
270 bool allow_underscores = *flags & PERL_SCAN_ALLOW_UNDERSCORES;
271 bool overflowed = FALSE;
272 const char *hexdigit;
273
274 if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
275 /* strip off leading x or 0x.
276 for compatibility silently suffer "x" and "0x" as valid hex numbers.
277 */
278 if (len >= 1) {
279 if (s[0] == 'x') {
280 s++;
281 len--;
282 }
283 else if (len >= 2 && s[0] == '0' && s[1] == 'x') {
284 s+=2;
285 len-=2;
286 }
287 }
288 }
289
290 for (; len-- && *s; s++) {
291 hexdigit = strchr((char *) PL_hexdigit, *s);
292 if (hexdigit) {
293 /* Write it in this wonky order with a goto to attempt to get the
294 compiler to make the common case integer-only loop pretty tight.
295 With gcc seems to be much straighter code than old scan_hex. */
296 redo:
297 if (!overflowed) {
298 if (value <= max_div_16) {
299 value = (value << 4) | ((hexdigit - PL_hexdigit) & 15);
300 continue;
301 }
302 /* Bah. We're just overflowed. */
303 if (ckWARN_d(WARN_OVERFLOW))
304 Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
305 "Integer overflow in hexadecimal number");
306 overflowed = TRUE;
307 value_nv = (NV) value;
308 }
309 value_nv *= 16.0;
310 /* If an NV has not enough bits in its mantissa to
311 * represent a UV this summing of small low-order numbers
312 * is a waste of time (because the NV cannot preserve
313 * the low-order bits anyway): we could just remember when
314 * did we overflow and in the end just multiply value_nv by the
315 * right amount of 16-tuples. */
316 value_nv += (NV)((hexdigit - PL_hexdigit) & 15);
317 continue;
318 }
319 if (*s == '_' && len && allow_underscores && s[1]
320 && (hexdigit = strchr((char *) PL_hexdigit, s[1])))
321 {
322 --len;
323 ++s;
324 goto redo;
325 }
326 if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
327 Perl_warner(aTHX_ packWARN(WARN_DIGIT),
328 "Illegal hexadecimal digit '%c' ignored", *s);
329 break;
330 }
331
332 if ( ( overflowed && value_nv > 4294967295.0)
333#if UVSIZE > 4
334 || (!overflowed && value > 0xffffffff )
335#endif
336 ) {
337 if (ckWARN(WARN_PORTABLE))
338 Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
339 "Hexadecimal number > 0xffffffff non-portable");
340 }
341 *len_p = s - start;
342 if (!overflowed) {
343 *flags = 0;
344 return value;
345 }
346 *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
347 if (result)
348 *result = value_nv;
349 return UV_MAX;
350}
351
352/*
353=for apidoc grok_oct
354
355converts a string representing an octal number to numeric form.
356
357On entry I<start> and I<*len> give the string to scan, I<*flags> gives
358conversion flags, and I<result> should be NULL or a pointer to an NV.
359The scan stops at the end of the string, or the first invalid character.
360Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*flags>, encountering an
361invalid character will also trigger a warning.
362On return I<*len> is set to the length of the scanned string,
363and I<*flags> gives output flags.
364
365If the value is <= UV_MAX it is returned as a UV, the output flags are clear,
366and nothing is written to I<*result>. If the value is > UV_MAX C<grok_oct>
367returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
368and writes the value to I<*result> (or the value is discarded if I<result>
369is NULL).
370
371If C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the octal
372number may use '_' characters to separate digits.
373
374=cut
375 */
376
377UV
378Perl_grok_oct(pTHX_ char *start, STRLEN *len_p, I32 *flags, NV *result) {
379 const char *s = start;
380 STRLEN len = *len_p;
381 UV value = 0;
382 NV value_nv = 0;
383
384 const UV max_div_8 = UV_MAX / 8;
385 bool allow_underscores = *flags & PERL_SCAN_ALLOW_UNDERSCORES;
386 bool overflowed = FALSE;
387
388 for (; len-- && *s; s++) {
389 /* gcc 2.95 optimiser not smart enough to figure that this subtraction
390 out front allows slicker code. */
391 int digit = *s - '0';
392 if (digit >= 0 && digit <= 7) {
393 /* Write it in this wonky order with a goto to attempt to get the
394 compiler to make the common case integer-only loop pretty tight.
395 */
396 redo:
397 if (!overflowed) {
398 if (value <= max_div_8) {
399 value = (value << 3) | digit;
400 continue;
401 }
402 /* Bah. We're just overflowed. */
403 if (ckWARN_d(WARN_OVERFLOW))
404 Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
405 "Integer overflow in octal number");
406 overflowed = TRUE;
407 value_nv = (NV) value;
408 }
409 value_nv *= 8.0;
410 /* If an NV has not enough bits in its mantissa to
411 * represent a UV this summing of small low-order numbers
412 * is a waste of time (because the NV cannot preserve
413 * the low-order bits anyway): we could just remember when
414 * did we overflow and in the end just multiply value_nv by the
415 * right amount of 8-tuples. */
416 value_nv += (NV)digit;
417 continue;
418 }
419 if (digit == ('_' - '0') && len && allow_underscores
420 && (digit = s[1] - '0') && (digit >= 0 && digit <= 7))
421 {
422 --len;
423 ++s;
424 goto redo;
425 }
426 /* Allow \octal to work the DWIM way (that is, stop scanning
427 * as soon as non-octal characters are seen, complain only if
428 * someone seems to want to use the digits eight and nine). */
429 if (digit == 8 || digit == 9) {
430 if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
431 Perl_warner(aTHX_ packWARN(WARN_DIGIT),
432 "Illegal octal digit '%c' ignored", *s);
433 }
434 break;
435 }
436
437 if ( ( overflowed && value_nv > 4294967295.0)
438#if UVSIZE > 4
439 || (!overflowed && value > 0xffffffff )
440#endif
441 ) {
442 if (ckWARN(WARN_PORTABLE))
443 Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
444 "Octal number > 037777777777 non-portable");
445 }
446 *len_p = s - start;
447 if (!overflowed) {
448 *flags = 0;
449 return value;
450 }
451 *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
452 if (result)
453 *result = value_nv;
454 return UV_MAX;
455}
456
457/*
458=for apidoc scan_bin
459
460For backwards compatibility. Use C<grok_bin> instead.
461
462=for apidoc scan_hex
463
464For backwards compatibility. Use C<grok_hex> instead.
465
466=for apidoc scan_oct
467
468For backwards compatibility. Use C<grok_oct> instead.
469
470=cut
471 */
472
473NV
474Perl_scan_bin(pTHX_ char *start, STRLEN len, STRLEN *retlen)
475{
476 NV rnv;
477 I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
478 UV ruv = grok_bin (start, &len, &flags, &rnv);
479
480 *retlen = len;
481 return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
482}
483
484NV
485Perl_scan_oct(pTHX_ char *start, STRLEN len, STRLEN *retlen)
486{
487 NV rnv;
488 I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
489 UV ruv = grok_oct (start, &len, &flags, &rnv);
490
491 *retlen = len;
492 return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
493}
494
495NV
496Perl_scan_hex(pTHX_ char *start, STRLEN len, STRLEN *retlen)
497{
498 NV rnv;
499 I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
500 UV ruv = grok_hex (start, &len, &flags, &rnv);
501
502 *retlen = len;
503 return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
504}
505
506/*
507=for apidoc grok_numeric_radix
508
509Scan and skip for a numeric decimal separator (radix).
510
511=cut
512 */
513bool
514Perl_grok_numeric_radix(pTHX_ const char **sp, const char *send)
515{
516#ifdef USE_LOCALE_NUMERIC
517 if (PL_numeric_radix_sv && IN_LOCALE) {
518 STRLEN len;
519 char* radix = SvPV(PL_numeric_radix_sv, len);
520 if (*sp + len <= send && memEQ(*sp, radix, len)) {
521 *sp += len;
522 return TRUE;
523 }
524 }
525 /* always try "." if numeric radix didn't match because
526 * we may have data from different locales mixed */
527#endif
528 if (*sp < send && **sp == '.') {
529 ++*sp;
530 return TRUE;
531 }
532 return FALSE;
533}
534
535/*
536=for apidoc grok_number
537
538Recognise (or not) a number. The type of the number is returned
539(0 if unrecognised), otherwise it is a bit-ORed combination of
540IS_NUMBER_IN_UV, IS_NUMBER_GREATER_THAN_UV_MAX, IS_NUMBER_NOT_INT,
541IS_NUMBER_NEG, IS_NUMBER_INFINITY, IS_NUMBER_NAN (defined in perl.h).
542
543If the value of the number can fit an in UV, it is returned in the *valuep
544IS_NUMBER_IN_UV will be set to indicate that *valuep is valid, IS_NUMBER_IN_UV
545will never be set unless *valuep is valid, but *valuep may have been assigned
546to during processing even though IS_NUMBER_IN_UV is not set on return.
547If valuep is NULL, IS_NUMBER_IN_UV will be set for the same cases as when
548valuep is non-NULL, but no actual assignment (or SEGV) will occur.
549
550IS_NUMBER_NOT_INT will be set with IS_NUMBER_IN_UV if trailing decimals were
551seen (in which case *valuep gives the true value truncated to an integer), and
552IS_NUMBER_NEG if the number is negative (in which case *valuep holds the
553absolute value). IS_NUMBER_IN_UV is not set if e notation was used or the
554number is larger than a UV.
555
556=cut
557 */
558int
559Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep)
560{
561 const char *s = pv;
562 const char *send = pv + len;
563 const UV max_div_10 = UV_MAX / 10;
564 const char max_mod_10 = UV_MAX % 10;
565 int numtype = 0;
566 int sawinf = 0;
567 int sawnan = 0;
568
569 while (s < send && isSPACE(*s))
570 s++;
571 if (s == send) {
572 return 0;
573 } else if (*s == '-') {
574 s++;
575 numtype = IS_NUMBER_NEG;
576 }
577 else if (*s == '+')
578 s++;
579
580 if (s == send)
581 return 0;
582
583 /* next must be digit or the radix separator or beginning of infinity */
584 if (isDIGIT(*s)) {
585 /* UVs are at least 32 bits, so the first 9 decimal digits cannot
586 overflow. */
587 UV value = *s - '0';
588 /* This construction seems to be more optimiser friendly.
589 (without it gcc does the isDIGIT test and the *s - '0' separately)
590 With it gcc on arm is managing 6 instructions (6 cycles) per digit.
591 In theory the optimiser could deduce how far to unroll the loop
592 before checking for overflow. */
593 if (++s < send) {
594 int digit = *s - '0';
595 if (digit >= 0 && digit <= 9) {
596 value = value * 10 + digit;
597 if (++s < send) {
598 digit = *s - '0';
599 if (digit >= 0 && digit <= 9) {
600 value = value * 10 + digit;
601 if (++s < send) {
602 digit = *s - '0';
603 if (digit >= 0 && digit <= 9) {
604 value = value * 10 + digit;
605 if (++s < send) {
606 digit = *s - '0';
607 if (digit >= 0 && digit <= 9) {
608 value = value * 10 + digit;
609 if (++s < send) {
610 digit = *s - '0';
611 if (digit >= 0 && digit <= 9) {
612 value = value * 10 + digit;
613 if (++s < send) {
614 digit = *s - '0';
615 if (digit >= 0 && digit <= 9) {
616 value = value * 10 + digit;
617 if (++s < send) {
618 digit = *s - '0';
619 if (digit >= 0 && digit <= 9) {
620 value = value * 10 + digit;
621 if (++s < send) {
622 digit = *s - '0';
623 if (digit >= 0 && digit <= 9) {
624 value = value * 10 + digit;
625 if (++s < send) {
626 /* Now got 9 digits, so need to check
627 each time for overflow. */
628 digit = *s - '0';
629 while (digit >= 0 && digit <= 9
630 && (value < max_div_10
631 || (value == max_div_10
632 && digit <= max_mod_10))) {
633 value = value * 10 + digit;
634 if (++s < send)
635 digit = *s - '0';
636 else
637 break;
638 }
639 if (digit >= 0 && digit <= 9
640 && (s < send)) {
641 /* value overflowed.
642 skip the remaining digits, don't
643 worry about setting *valuep. */
644 do {
645 s++;
646 } while (s < send && isDIGIT(*s));
647 numtype |=
648 IS_NUMBER_GREATER_THAN_UV_MAX;
649 goto skip_value;
650 }
651 }
652 }
653 }
654 }
655 }
656 }
657 }
658 }
659 }
660 }
661 }
662 }
663 }
664 }
665 }
666 }
667 }
668 numtype |= IS_NUMBER_IN_UV;
669 if (valuep)
670 *valuep = value;
671
672 skip_value:
673 if (GROK_NUMERIC_RADIX(&s, send)) {
674 numtype |= IS_NUMBER_NOT_INT;
675 while (s < send && isDIGIT(*s)) /* optional digits after the radix */
676 s++;
677 }
678 }
679 else if (GROK_NUMERIC_RADIX(&s, send)) {
680 numtype |= IS_NUMBER_NOT_INT | IS_NUMBER_IN_UV; /* valuep assigned below */
681 /* no digits before the radix means we need digits after it */
682 if (s < send && isDIGIT(*s)) {
683 do {
684 s++;
685 } while (s < send && isDIGIT(*s));
686 if (valuep) {
687 /* integer approximation is valid - it's 0. */
688 *valuep = 0;
689 }
690 }
691 else
692 return 0;
693 } else if (*s == 'I' || *s == 'i') {
694 s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
695 s++; if (s == send || (*s != 'F' && *s != 'f')) return 0;
696 s++; if (s < send && (*s == 'I' || *s == 'i')) {
697 s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
698 s++; if (s == send || (*s != 'I' && *s != 'i')) return 0;
699 s++; if (s == send || (*s != 'T' && *s != 't')) return 0;
700 s++; if (s == send || (*s != 'Y' && *s != 'y')) return 0;
701 s++;
702 }
703 sawinf = 1;
704 } else if (*s == 'N' || *s == 'n') {
705 /* XXX TODO: There are signaling NaNs and quiet NaNs. */
706 s++; if (s == send || (*s != 'A' && *s != 'a')) return 0;
707 s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
708 s++;
709 sawnan = 1;
710 } else
711 return 0;
712
713 if (sawinf) {
714 numtype &= IS_NUMBER_NEG; /* Keep track of sign */
715 numtype |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT;
716 } else if (sawnan) {
717 numtype &= IS_NUMBER_NEG; /* Keep track of sign */
718 numtype |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT;
719 } else if (s < send) {
720 /* we can have an optional exponent part */
721 if (*s == 'e' || *s == 'E') {
722 /* The only flag we keep is sign. Blow away any "it's UV" */
723 numtype &= IS_NUMBER_NEG;
724 numtype |= IS_NUMBER_NOT_INT;
725 s++;
726 if (s < send && (*s == '-' || *s == '+'))
727 s++;
728 if (s < send && isDIGIT(*s)) {
729 do {
730 s++;
731 } while (s < send && isDIGIT(*s));
732 }
733 else
734 return 0;
735 }
736 }
737 while (s < send && isSPACE(*s))
738 s++;
739 if (s >= send)
740 return numtype;
741 if (len == 10 && memEQ(pv, "0 but true", 10)) {
742 if (valuep)
743 *valuep = 0;
744 return IS_NUMBER_IN_UV;
745 }
746 return 0;
747}
748
749STATIC NV
750S_mulexp10(NV value, I32 exponent)
751{
752 NV result = 1.0;
753 NV power = 10.0;
754 bool negative = 0;
755 I32 bit;
756
757 if (exponent == 0)
758 return value;
759 if (value == 0)
760 return 0;
761
762 /* On OpenVMS VAX we by default use the D_FLOAT double format,
763 * and that format does not have *easy* capabilities [1] for
764 * overflowing doubles 'silently' as IEEE fp does. We also need
765 * to support G_FLOAT on both VAX and Alpha, and though the exponent
766 * range is much larger than D_FLOAT it still doesn't do silent
767 * overflow. Therefore we need to detect early whether we would
768 * overflow (this is the behaviour of the native string-to-float
769 * conversion routines, and therefore of native applications, too).
770 *
771 * [1] Trying to establish a condition handler to trap floating point
772 * exceptions is not a good idea. */
773
774 /* In UNICOS and in certain Cray models (such as T90) there is no
775 * IEEE fp, and no way at all from C to catch fp overflows gracefully.
776 * There is something you can do if you are willing to use some
777 * inline assembler: the instruction is called DFI-- but that will
778 * disable *all* floating point interrupts, a little bit too large
779 * a hammer. Therefore we need to catch potential overflows before
780 * it's too late. */
781
782#if ((defined(VMS) && !defined(__IEEE_FP)) || defined(_UNICOS)) && defined(NV_MAX_10_EXP)
783 STMT_START {
784 NV exp_v = log10(value);
785 if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP)
786 return NV_MAX;
787 if (exponent < 0) {
788 if (-(exponent + exp_v) >= NV_MAX_10_EXP)
789 return 0.0;
790 while (-exponent >= NV_MAX_10_EXP) {
791 /* combination does not overflow, but 10^(-exponent) does */
792 value /= 10;
793 ++exponent;
794 }
795 }
796 } STMT_END;
797#endif
798
799 if (exponent < 0) {
800 negative = 1;
801 exponent = -exponent;
802 }
803 for (bit = 1; exponent; bit <<= 1) {
804 if (exponent & bit) {
805 exponent ^= bit;
806 result *= power;
807 /* Floating point exceptions are supposed to be turned off,
808 * but if we're obviously done, don't risk another iteration.
809 */
810 if (exponent == 0) break;
811 }
812 power *= power;
813 }
814 return negative ? value / result : value * result;
815}
816
817NV
818Perl_my_atof(pTHX_ const char* s)
819{
820 NV x = 0.0;
821#ifdef USE_LOCALE_NUMERIC
822 if (PL_numeric_local && IN_LOCALE) {
823 NV y;
824
825 /* Scan the number twice; once using locale and once without;
826 * choose the larger result (in absolute value). */
827 Perl_atof2(s, x);
828 SET_NUMERIC_STANDARD();
829 Perl_atof2(s, y);
830 SET_NUMERIC_LOCAL();
831 if ((y < 0.0 && y < x) || (y > 0.0 && y > x))
832 return y;
833 }
834 else
835 Perl_atof2(s, x);
836#else
837 Perl_atof2(s, x);
838#endif
839 return x;
840}
841
842char*
843Perl_my_atof2(pTHX_ const char* orig, NV* value)
844{
845 NV result[3] = {0.0, 0.0, 0.0};
846 char* s = (char*)orig;
847#ifdef USE_PERL_ATOF
848 UV accumulator[2] = {0,0}; /* before/after dp */
849 bool negative = 0;
850 char* send = s + strlen(orig) - 1;
851 bool seen_digit = 0;
852 I32 exp_adjust[2] = {0,0};
853 I32 exp_acc[2] = {-1, -1};
854 /* the current exponent adjust for the accumulators */
855 I32 exponent = 0;
856 I32 seen_dp = 0;
857 I32 digit = 0;
858 I32 old_digit = 0;
859 I32 sig_digits = 0; /* noof significant digits seen so far */
860
861/* There is no point in processing more significant digits
862 * than the NV can hold. Note that NV_DIG is a lower-bound value,
863 * while we need an upper-bound value. We add 2 to account for this;
864 * since it will have been conservative on both the first and last digit.
865 * For example a 32-bit mantissa with an exponent of 4 would have
866 * exact values in the set
867 * 4
868 * 8
869 * ..
870 * 17179869172
871 * 17179869176
872 * 17179869180
873 *
874 * where for the purposes of calculating NV_DIG we would have to discount
875 * both the first and last digit, since neither can hold all values from
876 * 0..9; but for calculating the value we must examine those two digits.
877 */
878#define MAX_SIG_DIGITS (NV_DIG+2)
879
880/* the max number we can accumulate in a UV, and still safely do 10*N+9 */
881#define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10))
882
883 /* leading whitespace */
884 while (isSPACE(*s))
885 ++s;
886
887 /* sign */
888 switch (*s) {
889 case '-':
890 negative = 1;
891 /* fall through */
892 case '+':
893 ++s;
894 }
895
896 /* we accumulate digits into an integer; when this becomes too
897 * large, we add the total to NV and start again */
898
899 while (1) {
900 if (isDIGIT(*s)) {
901 seen_digit = 1;
902 old_digit = digit;
903 digit = *s++ - '0';
904 if (seen_dp)
905 exp_adjust[1]++;
906
907 /* don't start counting until we see the first significant
908 * digit, eg the 5 in 0.00005... */
909 if (!sig_digits && digit == 0)
910 continue;
911
912 if (++sig_digits > MAX_SIG_DIGITS) {
913 /* limits of precision reached */
914 if (digit > 5) {
915 ++accumulator[seen_dp];
916 } else if (digit == 5) {
917 if (old_digit % 2) { /* round to even - Allen */
918 ++accumulator[seen_dp];
919 }
920 }
921 if (seen_dp) {
922 exp_adjust[1]--;
923 } else {
924 exp_adjust[0]++;
925 }
926 /* skip remaining digits */
927 while (isDIGIT(*s)) {
928 ++s;
929 if (! seen_dp) {
930 exp_adjust[0]++;
931 }
932 }
933 /* warn of loss of precision? */
934 }
935 else {
936 if (accumulator[seen_dp] > MAX_ACCUMULATE) {
937 /* add accumulator to result and start again */
938 result[seen_dp] = S_mulexp10(result[seen_dp],
939 exp_acc[seen_dp])
940 + (NV)accumulator[seen_dp];
941 accumulator[seen_dp] = 0;
942 exp_acc[seen_dp] = 0;
943 }
944 accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit;
945 ++exp_acc[seen_dp];
946 }
947 }
948 else if (!seen_dp && GROK_NUMERIC_RADIX((const char **)&s, send)) {
949 seen_dp = 1;
950 if (sig_digits > MAX_SIG_DIGITS) {
951 ++s;
952 while (isDIGIT(*s)) {
953 ++s;
954 }
955 break;
956 }
957 }
958 else {
959 break;
960 }
961 }
962
963 result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0];
964 if (seen_dp) {
965 result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1];
966 }
967
968 if (seen_digit && (*s == 'e' || *s == 'E')) {
969 bool expnegative = 0;
970
971 ++s;
972 switch (*s) {
973 case '-':
974 expnegative = 1;
975 /* fall through */
976 case '+':
977 ++s;
978 }
979 while (isDIGIT(*s))
980 exponent = exponent * 10 + (*s++ - '0');
981 if (expnegative)
982 exponent = -exponent;
983 }
984
985
986
987 /* now apply the exponent */
988
989 if (seen_dp) {
990 result[2] = S_mulexp10(result[0],exponent+exp_adjust[0])
991 + S_mulexp10(result[1],exponent-exp_adjust[1]);
992 } else {
993 result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]);
994 }
995
996 /* now apply the sign */
997 if (negative)
998 result[2] = -result[2];
999#endif /* USE_PERL_ATOF */
1000 *value = result[2];
1001 return s;
1002}
1003
1004#if ! defined(HAS_MODFL) && defined(HAS_AINTL) && defined(HAS_COPYSIGNL)
1005long double
1006Perl_my_modfl(long double x, long double *ip)
1007{
1008 *ip = aintl(x);
1009 return (x == *ip ? copysignl(0.0L, x) : x - *ip);
1010}
1011#endif
1012
1013#if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL)
1014long double
1015Perl_my_frexpl(long double x, int *e) {
1016 *e = x == 0.0L ? 0 : ilogbl(x) + 1;
1017 return (scalbnl(x, -*e));
1018}
1019#endif