This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlop.pod: Slight clarification
[perl5.git] / numeric.c
CommitLineData
98994639
HS
1/* numeric.c
2 *
663f364b 3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
1129b882 4 * 2002, 2003, 2004, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
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/*
4ac71550
TC
12 * "That only makes eleven (plus one mislaid) and not fourteen,
13 * unless wizards count differently to other people." --Beorn
14 *
15 * [p.115 of _The Hobbit_: "Queer Lodgings"]
98994639
HS
16 */
17
ccfc67b7
JH
18/*
19=head1 Numeric functions
166f8a29 20
7fefc6c1
KW
21=cut
22
166f8a29
DM
23This file contains all the stuff needed by perl for manipulating numeric
24values, including such things as replacements for the OS's atof() function
25
ccfc67b7
JH
26*/
27
98994639
HS
28#include "EXTERN.h"
29#define PERL_IN_NUMERIC_C
30#include "perl.h"
31
9ec8aea5
KW
32#ifdef Perl_strtod
33
34PERL_STATIC_INLINE NV
35S_strtod(pTHX_ const char * const s, char ** e)
36{
37 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
38 NV result;
39
40 STORE_LC_NUMERIC_SET_TO_NEEDED();
41
42# ifdef USE_QUADMATH
43
44 result = strtoflt128(s, e);
45
46# elif defined(HAS_STRTOLD) && defined(HAS_LONG_DOUBLE) \
47 && defined(USE_LONG_DOUBLE)
48# if defined(__MINGW64_VERSION_MAJOR)
49 /***********************************************
50 We are unable to use strtold because of
51 https://sourceforge.net/p/mingw-w64/bugs/711/
52 &
53 https://sourceforge.net/p/mingw-w64/bugs/725/
54
55 but __mingw_strtold is fine.
56 ***********************************************/
57
58 result = __mingw_strtold(s, e);
59
60# else
61
62 result = strtold(s, e);
63
64# endif
65# elif defined(HAS_STRTOD)
66
67 result = strtod(s, e);
68
02fd9d54
KW
69# else
70# error No strtod() equivalent found
9ec8aea5
KW
71# endif
72
73 RESTORE_LC_NUMERIC();
74
75 return result;
76}
77
78#endif /* #ifdef Perl_strtod */
79
80/*
81
82=for apidoc my_strtod
83
84This function is equivalent to the libc strtod() function, and is available
85even on platforms that lack plain strtod(). Its return value is the best
86available precision depending on platform capabilities and F<Configure>
87options.
88
89It properly handles the locale radix character, meaning it expects a dot except
90when called from within the scope of S<C<use locale>>, in which case the radix
91character should be that specified by the current locale.
92
e7a3fd45 93The synonym Strtod() may be used instead.
9ec8aea5
KW
94
95=cut
96
97*/
98
99NV
0f3d8cd4 100Perl_my_strtod(const char * const s, char **e)
9ec8aea5
KW
101{
102 dTHX;
103
104 PERL_ARGS_ASSERT_MY_STRTOD;
105
106#ifdef Perl_strtod
107
108 return S_strtod(aTHX_ s, e);
109
110#else
111
112 {
113 NV result;
114 char ** end_ptr = NULL;
115
116 *end_ptr = my_atof2(s, &result);
117 if (e) {
118 *e = *end_ptr;
119 }
120
121 if (! *end_ptr) {
122 result = 0.0;
123 }
124
125 return result;
126 }
127
128#endif
129
130}
131
132
98994639 133U32
ddeaf645 134Perl_cast_ulong(NV f)
98994639
HS
135{
136 if (f < 0.0)
137 return f < I32_MIN ? (U32) I32_MIN : (U32)(I32) f;
138 if (f < U32_MAX_P1) {
139#if CASTFLAGS & 2
140 if (f < U32_MAX_P1_HALF)
141 return (U32) f;
142 f -= U32_MAX_P1_HALF;
071db91b 143 return ((U32) f) | (1 + (U32_MAX >> 1));
98994639
HS
144#else
145 return (U32) f;
146#endif
147 }
148 return f > 0 ? U32_MAX : 0 /* NaN */;
149}
150
151I32
ddeaf645 152Perl_cast_i32(NV f)
98994639
HS
153{
154 if (f < I32_MAX_P1)
155 return f < I32_MIN ? I32_MIN : (I32) f;
156 if (f < U32_MAX_P1) {
157#if CASTFLAGS & 2
158 if (f < U32_MAX_P1_HALF)
159 return (I32)(U32) f;
160 f -= U32_MAX_P1_HALF;
071db91b 161 return (I32)(((U32) f) | (1 + (U32_MAX >> 1)));
98994639
HS
162#else
163 return (I32)(U32) f;
164#endif
165 }
166 return f > 0 ? (I32)U32_MAX : 0 /* NaN */;
167}
168
169IV
ddeaf645 170Perl_cast_iv(NV f)
98994639
HS
171{
172 if (f < IV_MAX_P1)
173 return f < IV_MIN ? IV_MIN : (IV) f;
174 if (f < UV_MAX_P1) {
175#if CASTFLAGS & 2
176 /* For future flexibility allowing for sizeof(UV) >= sizeof(IV) */
177 if (f < UV_MAX_P1_HALF)
178 return (IV)(UV) f;
179 f -= UV_MAX_P1_HALF;
071db91b 180 return (IV)(((UV) f) | (1 + (UV_MAX >> 1)));
98994639
HS
181#else
182 return (IV)(UV) f;
183#endif
184 }
185 return f > 0 ? (IV)UV_MAX : 0 /* NaN */;
186}
187
188UV
ddeaf645 189Perl_cast_uv(NV f)
98994639
HS
190{
191 if (f < 0.0)
192 return f < IV_MIN ? (UV) IV_MIN : (UV)(IV) f;
193 if (f < UV_MAX_P1) {
194#if CASTFLAGS & 2
195 if (f < UV_MAX_P1_HALF)
196 return (UV) f;
197 f -= UV_MAX_P1_HALF;
071db91b 198 return ((UV) f) | (1 + (UV_MAX >> 1));
98994639
HS
199#else
200 return (UV) f;
201#endif
202 }
203 return f > 0 ? UV_MAX : 0 /* NaN */;
204}
205
53305cf1
NC
206/*
207=for apidoc grok_bin
98994639 208
53305cf1
NC
209converts a string representing a binary number to numeric form.
210
2d7f6611 211On entry C<start> and C<*len> give the string to scan, C<*flags> gives
796b6530 212conversion flags, and C<result> should be C<NULL> or a pointer to an NV.
53305cf1 213The scan stops at the end of the string, or the first invalid character.
2d7f6611 214Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in C<*flags>, encountering an
7b667b5f 215invalid character will also trigger a warning.
2d7f6611
KW
216On return C<*len> is set to the length of the scanned string,
217and C<*flags> gives output flags.
53305cf1 218
7fc63493 219If the value is <= C<UV_MAX> it is returned as a UV, the output flags are clear,
796b6530
KW
220and nothing is written to C<*result>. If the value is > C<UV_MAX>, C<grok_bin>
221returns C<UV_MAX>, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
2d7f6611 222and writes the value to C<*result> (or the value is discarded if C<result>
53305cf1
NC
223is NULL).
224
796b6530 225The binary number may optionally be prefixed with C<"0b"> or C<"b"> unless
2d7f6611
KW
226C<PERL_SCAN_DISALLOW_PREFIX> is set in C<*flags> on entry. If
227C<PERL_SCAN_ALLOW_UNDERSCORES> is set in C<*flags> then the binary
796b6530 228number may use C<"_"> characters to separate digits.
53305cf1 229
ceea512d
KW
230=for apidoc Amnh||PERL_SCAN_ALLOW_UNDERSCORES
231=for apidoc Amnh||PERL_SCAN_DISALLOW_PREFIX
232=for apidoc Amnh||PERL_SCAN_GREATER_THAN_UV_MAX
233=for apidoc Amnh||PERL_SCAN_SILENT_ILLDIGIT
234=for apidoc Amnh||PERL_SCAN_TRAILING
235
53305cf1 236=cut
02470786
KW
237
238Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE
239which suppresses any message for non-portable numbers that are still valid
240on this platform.
53305cf1
NC
241 */
242
243UV
7918f24d
NC
244Perl_grok_bin(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result)
245{
53305cf1
NC
246 const char *s = start;
247 STRLEN len = *len_p;
248 UV value = 0;
249 NV value_nv = 0;
250
251 const UV max_div_2 = UV_MAX / 2;
f2338a2e 252 const bool allow_underscores = cBOOL(*flags & PERL_SCAN_ALLOW_UNDERSCORES);
53305cf1 253 bool overflowed = FALSE;
7fc63493 254 char bit;
53305cf1 255
7918f24d
NC
256 PERL_ARGS_ASSERT_GROK_BIN;
257
a4c04bdc
NC
258 if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
259 /* strip off leading b or 0b.
260 for compatibility silently suffer "b" and "0b" as valid binary
261 numbers. */
262 if (len >= 1) {
305b8651 263 if (isALPHA_FOLD_EQ(s[0], 'b')) {
a4c04bdc
NC
264 s++;
265 len--;
266 }
305b8651 267 else if (len >= 2 && s[0] == '0' && (isALPHA_FOLD_EQ(s[1], 'b'))) {
a4c04bdc
NC
268 s+=2;
269 len-=2;
270 }
271 }
53305cf1
NC
272 }
273
7fc63493 274 for (; len-- && (bit = *s); s++) {
53305cf1
NC
275 if (bit == '0' || bit == '1') {
276 /* Write it in this wonky order with a goto to attempt to get the
277 compiler to make the common case integer-only loop pretty tight.
278 With gcc seems to be much straighter code than old scan_bin. */
279 redo:
280 if (!overflowed) {
281 if (value <= max_div_2) {
282 value = (value << 1) | (bit - '0');
283 continue;
284 }
285 /* Bah. We're just overflowed. */
dcbac5bb 286 /* diag_listed_as: Integer overflow in %s number */
9b387841
NC
287 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
288 "Integer overflow in binary number");
53305cf1
NC
289 overflowed = TRUE;
290 value_nv = (NV) value;
291 }
292 value_nv *= 2.0;
98994639 293 /* If an NV has not enough bits in its mantissa to
d1be9408 294 * represent a UV this summing of small low-order numbers
98994639
HS
295 * is a waste of time (because the NV cannot preserve
296 * the low-order bits anyway): we could just remember when
53305cf1 297 * did we overflow and in the end just multiply value_nv by the
98994639 298 * right amount. */
53305cf1
NC
299 value_nv += (NV)(bit - '0');
300 continue;
301 }
302 if (bit == '_' && len && allow_underscores && (bit = s[1])
303 && (bit == '0' || bit == '1'))
98994639
HS
304 {
305 --len;
306 ++s;
53305cf1 307 goto redo;
98994639 308 }
a2a5de95
NC
309 if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT))
310 Perl_ck_warner(aTHX_ packWARN(WARN_DIGIT),
311 "Illegal binary digit '%c' ignored", *s);
53305cf1 312 break;
98994639 313 }
19c1206d 314
53305cf1 315 if ( ( overflowed && value_nv > 4294967295.0)
98994639 316#if UVSIZE > 4
02470786
KW
317 || (!overflowed && value > 0xffffffff
318 && ! (*flags & PERL_SCAN_SILENT_NON_PORTABLE))
98994639
HS
319#endif
320 ) {
a2a5de95
NC
321 Perl_ck_warner(aTHX_ packWARN(WARN_PORTABLE),
322 "Binary number > 0b11111111111111111111111111111111 non-portable");
53305cf1
NC
323 }
324 *len_p = s - start;
325 if (!overflowed) {
326 *flags = 0;
327 return value;
98994639 328 }
53305cf1
NC
329 *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
330 if (result)
331 *result = value_nv;
332 return UV_MAX;
98994639
HS
333}
334
53305cf1
NC
335/*
336=for apidoc grok_hex
337
338converts a string representing a hex number to numeric form.
339
2d7f6611 340On entry C<start> and C<*len_p> give the string to scan, C<*flags> gives
796b6530 341conversion flags, and C<result> should be C<NULL> or a pointer to an NV.
7b667b5f 342The scan stops at the end of the string, or the first invalid character.
2d7f6611 343Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in C<*flags>, encountering an
7b667b5f 344invalid character will also trigger a warning.
2d7f6611
KW
345On return C<*len> is set to the length of the scanned string,
346and C<*flags> gives output flags.
53305cf1 347
796b6530
KW
348If the value is <= C<UV_MAX> it is returned as a UV, the output flags are clear,
349and nothing is written to C<*result>. If the value is > C<UV_MAX>, C<grok_hex>
350returns C<UV_MAX>, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
2d7f6611 351and writes the value to C<*result> (or the value is discarded if C<result>
796b6530 352is C<NULL>).
53305cf1 353
796b6530 354The hex number may optionally be prefixed with C<"0x"> or C<"x"> unless
2d7f6611
KW
355C<PERL_SCAN_DISALLOW_PREFIX> is set in C<*flags> on entry. If
356C<PERL_SCAN_ALLOW_UNDERSCORES> is set in C<*flags> then the hex
796b6530 357number may use C<"_"> characters to separate digits.
53305cf1
NC
358
359=cut
02470786
KW
360
361Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE
baf48926 362which suppresses any message for non-portable numbers, but which are valid
02470786 363on this platform.
53305cf1
NC
364 */
365
366UV
7918f24d
NC
367Perl_grok_hex(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result)
368{
53305cf1
NC
369 const char *s = start;
370 STRLEN len = *len_p;
371 UV value = 0;
372 NV value_nv = 0;
53305cf1 373 const UV max_div_16 = UV_MAX / 16;
f2338a2e 374 const bool allow_underscores = cBOOL(*flags & PERL_SCAN_ALLOW_UNDERSCORES);
53305cf1 375 bool overflowed = FALSE;
98994639 376
7918f24d
NC
377 PERL_ARGS_ASSERT_GROK_HEX;
378
a4c04bdc
NC
379 if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
380 /* strip off leading x or 0x.
381 for compatibility silently suffer "x" and "0x" as valid hex numbers.
382 */
383 if (len >= 1) {
305b8651 384 if (isALPHA_FOLD_EQ(s[0], 'x')) {
a4c04bdc
NC
385 s++;
386 len--;
387 }
305b8651 388 else if (len >= 2 && s[0] == '0' && (isALPHA_FOLD_EQ(s[1], 'x'))) {
a4c04bdc
NC
389 s+=2;
390 len-=2;
391 }
392 }
98994639
HS
393 }
394
395 for (; len-- && *s; s++) {
626ef089 396 if (isXDIGIT(*s)) {
53305cf1
NC
397 /* Write it in this wonky order with a goto to attempt to get the
398 compiler to make the common case integer-only loop pretty tight.
399 With gcc seems to be much straighter code than old scan_hex. */
400 redo:
401 if (!overflowed) {
402 if (value <= max_div_16) {
626ef089 403 value = (value << 4) | XDIGIT_VALUE(*s);
53305cf1
NC
404 continue;
405 }
406 /* Bah. We're just overflowed. */
dcbac5bb 407 /* diag_listed_as: Integer overflow in %s number */
9b387841
NC
408 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
409 "Integer overflow in hexadecimal number");
53305cf1
NC
410 overflowed = TRUE;
411 value_nv = (NV) value;
412 }
413 value_nv *= 16.0;
414 /* If an NV has not enough bits in its mantissa to
d1be9408 415 * represent a UV this summing of small low-order numbers
53305cf1
NC
416 * is a waste of time (because the NV cannot preserve
417 * the low-order bits anyway): we could just remember when
418 * did we overflow and in the end just multiply value_nv by the
419 * right amount of 16-tuples. */
626ef089 420 value_nv += (NV) XDIGIT_VALUE(*s);
53305cf1
NC
421 continue;
422 }
423 if (*s == '_' && len && allow_underscores && s[1]
626ef089 424 && isXDIGIT(s[1]))
98994639
HS
425 {
426 --len;
427 ++s;
53305cf1 428 goto redo;
98994639 429 }
a2a5de95
NC
430 if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT))
431 Perl_ck_warner(aTHX_ packWARN(WARN_DIGIT),
53305cf1
NC
432 "Illegal hexadecimal digit '%c' ignored", *s);
433 break;
434 }
19c1206d 435
53305cf1
NC
436 if ( ( overflowed && value_nv > 4294967295.0)
437#if UVSIZE > 4
02470786
KW
438 || (!overflowed && value > 0xffffffff
439 && ! (*flags & PERL_SCAN_SILENT_NON_PORTABLE))
53305cf1
NC
440#endif
441 ) {
a2a5de95
NC
442 Perl_ck_warner(aTHX_ packWARN(WARN_PORTABLE),
443 "Hexadecimal number > 0xffffffff non-portable");
53305cf1
NC
444 }
445 *len_p = s - start;
446 if (!overflowed) {
447 *flags = 0;
448 return value;
449 }
450 *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
451 if (result)
452 *result = value_nv;
453 return UV_MAX;
454}
455
456/*
457=for apidoc grok_oct
458
7b667b5f
MHM
459converts a string representing an octal number to numeric form.
460
2d7f6611 461On entry C<start> and C<*len> give the string to scan, C<*flags> gives
796b6530 462conversion flags, and C<result> should be C<NULL> or a pointer to an NV.
7b667b5f 463The scan stops at the end of the string, or the first invalid character.
2d7f6611 464Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in C<*flags>, encountering an
154bd527 4658 or 9 will also trigger a warning.
2d7f6611
KW
466On return C<*len> is set to the length of the scanned string,
467and C<*flags> gives output flags.
7b667b5f 468
796b6530
KW
469If the value is <= C<UV_MAX> it is returned as a UV, the output flags are clear,
470and nothing is written to C<*result>. If the value is > C<UV_MAX>, C<grok_oct>
471returns C<UV_MAX>, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
2d7f6611 472and writes the value to C<*result> (or the value is discarded if C<result>
796b6530 473is C<NULL>).
7b667b5f 474
2d7f6611 475If C<PERL_SCAN_ALLOW_UNDERSCORES> is set in C<*flags> then the octal
796b6530 476number may use C<"_"> characters to separate digits.
53305cf1
NC
477
478=cut
02470786 479
333ae27c
KW
480Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE>
481which suppresses any message for non-portable numbers, but which are valid
02470786 482on this platform.
53305cf1
NC
483 */
484
485UV
7918f24d
NC
486Perl_grok_oct(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result)
487{
53305cf1
NC
488 const char *s = start;
489 STRLEN len = *len_p;
490 UV value = 0;
491 NV value_nv = 0;
53305cf1 492 const UV max_div_8 = UV_MAX / 8;
f2338a2e 493 const bool allow_underscores = cBOOL(*flags & PERL_SCAN_ALLOW_UNDERSCORES);
53305cf1
NC
494 bool overflowed = FALSE;
495
7918f24d
NC
496 PERL_ARGS_ASSERT_GROK_OCT;
497
53305cf1 498 for (; len-- && *s; s++) {
626ef089 499 if (isOCTAL(*s)) {
53305cf1
NC
500 /* Write it in this wonky order with a goto to attempt to get the
501 compiler to make the common case integer-only loop pretty tight.
502 */
503 redo:
504 if (!overflowed) {
505 if (value <= max_div_8) {
626ef089 506 value = (value << 3) | OCTAL_VALUE(*s);
53305cf1
NC
507 continue;
508 }
509 /* Bah. We're just overflowed. */
dcbac5bb 510 /* diag_listed_as: Integer overflow in %s number */
9b387841
NC
511 Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW),
512 "Integer overflow in octal number");
53305cf1
NC
513 overflowed = TRUE;
514 value_nv = (NV) value;
515 }
516 value_nv *= 8.0;
98994639 517 /* If an NV has not enough bits in its mantissa to
d1be9408 518 * represent a UV this summing of small low-order numbers
98994639
HS
519 * is a waste of time (because the NV cannot preserve
520 * the low-order bits anyway): we could just remember when
53305cf1
NC
521 * did we overflow and in the end just multiply value_nv by the
522 * right amount of 8-tuples. */
626ef089 523 value_nv += (NV) OCTAL_VALUE(*s);
53305cf1
NC
524 continue;
525 }
626ef089
KW
526 if (*s == '_' && len && allow_underscores && isOCTAL(s[1])) {
527 --len;
528 ++s;
529 goto redo;
530 }
53305cf1 531 /* Allow \octal to work the DWIM way (that is, stop scanning
7b667b5f 532 * as soon as non-octal characters are seen, complain only if
626ef089
KW
533 * someone seems to want to use the digits eight and nine. Since we
534 * know it is not octal, then if isDIGIT, must be an 8 or 9). */
535 if (isDIGIT(*s)) {
a2a5de95
NC
536 if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT))
537 Perl_ck_warner(aTHX_ packWARN(WARN_DIGIT),
538 "Illegal octal digit '%c' ignored", *s);
53305cf1
NC
539 }
540 break;
98994639 541 }
19c1206d 542
53305cf1 543 if ( ( overflowed && value_nv > 4294967295.0)
98994639 544#if UVSIZE > 4
02470786
KW
545 || (!overflowed && value > 0xffffffff
546 && ! (*flags & PERL_SCAN_SILENT_NON_PORTABLE))
98994639
HS
547#endif
548 ) {
a2a5de95
NC
549 Perl_ck_warner(aTHX_ packWARN(WARN_PORTABLE),
550 "Octal number > 037777777777 non-portable");
53305cf1
NC
551 }
552 *len_p = s - start;
553 if (!overflowed) {
554 *flags = 0;
555 return value;
98994639 556 }
53305cf1
NC
557 *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
558 if (result)
559 *result = value_nv;
560 return UV_MAX;
561}
562
563/*
564=for apidoc scan_bin
565
72d33970 566For backwards compatibility. Use C<grok_bin> instead.
53305cf1
NC
567
568=for apidoc scan_hex
569
72d33970 570For backwards compatibility. Use C<grok_hex> instead.
53305cf1
NC
571
572=for apidoc scan_oct
573
72d33970 574For backwards compatibility. Use C<grok_oct> instead.
53305cf1
NC
575
576=cut
577 */
578
579NV
73d840c0 580Perl_scan_bin(pTHX_ const char *start, STRLEN len, STRLEN *retlen)
53305cf1
NC
581{
582 NV rnv;
583 I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
73d840c0 584 const UV ruv = grok_bin (start, &len, &flags, &rnv);
53305cf1 585
7918f24d
NC
586 PERL_ARGS_ASSERT_SCAN_BIN;
587
53305cf1
NC
588 *retlen = len;
589 return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
590}
591
592NV
73d840c0 593Perl_scan_oct(pTHX_ const char *start, STRLEN len, STRLEN *retlen)
53305cf1
NC
594{
595 NV rnv;
596 I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
73d840c0 597 const UV ruv = grok_oct (start, &len, &flags, &rnv);
53305cf1 598
7918f24d
NC
599 PERL_ARGS_ASSERT_SCAN_OCT;
600
53305cf1
NC
601 *retlen = len;
602 return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
603}
604
605NV
73d840c0 606Perl_scan_hex(pTHX_ const char *start, STRLEN len, STRLEN *retlen)
53305cf1
NC
607{
608 NV rnv;
609 I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
73d840c0 610 const UV ruv = grok_hex (start, &len, &flags, &rnv);
53305cf1 611
7918f24d
NC
612 PERL_ARGS_ASSERT_SCAN_HEX;
613
53305cf1
NC
614 *retlen = len;
615 return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
98994639
HS
616}
617
618/*
619=for apidoc grok_numeric_radix
620
621Scan and skip for a numeric decimal separator (radix).
622
623=cut
624 */
625bool
626Perl_grok_numeric_radix(pTHX_ const char **sp, const char *send)
627{
7918f24d
NC
628 PERL_ARGS_ASSERT_GROK_NUMERIC_RADIX;
629
7ea85fa8
KW
630#ifdef USE_LOCALE_NUMERIC
631
d6ded950 632 if (IN_LC(LC_NUMERIC)) {
f0dafd73
KW
633 STRLEN len;
634 char * radix;
635 bool matches_radix = FALSE;
67d796ae 636 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f0dafd73 637
a1395eaf 638 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
f0dafd73
KW
639
640 radix = SvPV(PL_numeric_radix_sv, len);
641 radix = savepvn(radix, len);
642
21431899 643 RESTORE_LC_NUMERIC();
f0dafd73
KW
644
645 if (*sp + len <= send) {
646 matches_radix = memEQ(*sp, radix, len);
647 }
648
649 Safefree(radix);
650
651 if (matches_radix) {
652 *sp += len;
653 return TRUE;
654 }
98994639 655 }
f0dafd73 656
98994639 657#endif
7918f24d 658
f0dafd73
KW
659 /* always try "." if numeric radix didn't match because
660 * we may have data from different locales mixed */
98994639
HS
661 if (*sp < send && **sp == '.') {
662 ++*sp;
663 return TRUE;
664 }
f0dafd73 665
98994639
HS
666 return FALSE;
667}
668
569f27e5 669/*
ff4eb398
JH
670=for apidoc grok_infnan
671
796b6530 672Helper for C<grok_number()>, accepts various ways of spelling "infinity"
ff4eb398
JH
673or "not a number", and returns one of the following flag combinations:
674
5962c2f6 675 IS_NUMBER_INFINITY
ff4eb398 676 IS_NUMBER_NAN
5962c2f6 677 IS_NUMBER_INFINITY | IS_NUMBER_NEG
ff4eb398
JH
678 IS_NUMBER_NAN | IS_NUMBER_NEG
679 0
680
796b6530 681possibly |-ed with C<IS_NUMBER_TRAILING>.
b489e20f 682
796b6530 683If an infinity or a not-a-number is recognized, C<*sp> will point to
62bdc035 684one byte past the end of the recognized string. If the recognition fails,
796b6530 685zero is returned, and C<*sp> will not move.
ff4eb398 686
ceea512d
KW
687=for apidoc Amn|bool|IS_NUMBER_GREATER_THAN_UV_MAX
688=for apidoc Amn|bool|IS_NUMBER_INFINITY
689=for apidoc Amn|bool|IS_NUMBER_IN_UV
690=for apidoc Amn|bool|IS_NUMBER_NAN
691=for apidoc Amn|bool|IS_NUMBER_NEG
692=for apidoc Amn|bool|IS_NUMBER_NOT_INT
693
ff4eb398
JH
694=cut
695*/
696
697int
3823048b 698Perl_grok_infnan(pTHX_ const char** sp, const char* send)
ff4eb398
JH
699{
700 const char* s = *sp;
701 int flags = 0;
a5dc2484 702#if defined(NV_INF) || defined(NV_NAN)
62bdc035 703 bool odh = FALSE; /* one-dot-hash: 1.#INF */
ff4eb398
JH
704
705 PERL_ARGS_ASSERT_GROK_INFNAN;
706
8c12dc63
JH
707 if (*s == '+') {
708 s++; if (s == send) return 0;
709 }
710 else if (*s == '-') {
ff4eb398
JH
711 flags |= IS_NUMBER_NEG; /* Yes, -NaN happens. Incorrect but happens. */
712 s++; if (s == send) return 0;
713 }
714
715 if (*s == '1') {
62bdc035
JH
716 /* Visual C: 1.#SNAN, -1.#QNAN, 1#INF, 1.#IND (maybe also 1.#NAN)
717 * Let's keep the dot optional. */
ff4eb398
JH
718 s++; if (s == send) return 0;
719 if (*s == '.') {
720 s++; if (s == send) return 0;
721 }
722 if (*s == '#') {
723 s++; if (s == send) return 0;
724 } else
725 return 0;
e855f543 726 odh = TRUE;
ff4eb398
JH
727 }
728
305b8651 729 if (isALPHA_FOLD_EQ(*s, 'I')) {
62bdc035
JH
730 /* INF or IND (1.#IND is "indeterminate", a certain type of NAN) */
731
305b8651 732 s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0;
ff4eb398 733 s++; if (s == send) return 0;
305b8651 734 if (isALPHA_FOLD_EQ(*s, 'F')) {
ff4eb398 735 s++;
b8974fcb
JH
736 if (s < send && (isALPHA_FOLD_EQ(*s, 'I'))) {
737 int fail =
738 flags | IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT | IS_NUMBER_TRAILING;
739 s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return fail;
740 s++; if (s == send || isALPHA_FOLD_NE(*s, 'I')) return fail;
741 s++; if (s == send || isALPHA_FOLD_NE(*s, 'T')) return fail;
742 s++; if (s == send || isALPHA_FOLD_NE(*s, 'Y')) return fail;
3396ed30 743 s++;
b8974fcb
JH
744 } else if (odh) {
745 while (*s == '0') { /* 1.#INF00 */
746 s++;
747 }
3396ed30 748 }
b489e20f
JH
749 while (s < send && isSPACE(*s))
750 s++;
751 if (s < send && *s) {
3396ed30 752 flags |= IS_NUMBER_TRAILING;
fae4db12 753 }
ff4eb398
JH
754 flags |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT;
755 }
e855f543 756 else if (isALPHA_FOLD_EQ(*s, 'D') && odh) { /* 1.#IND */
ff4eb398
JH
757 s++;
758 flags |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT;
fae4db12
JH
759 while (*s == '0') { /* 1.#IND00 */
760 s++;
761 }
1e9aa12f
JH
762 if (*s) {
763 flags |= IS_NUMBER_TRAILING;
764 }
ff4eb398
JH
765 } else
766 return 0;
ff4eb398
JH
767 }
768 else {
62bdc035 769 /* Maybe NAN of some sort */
3823048b
JH
770
771 if (isALPHA_FOLD_EQ(*s, 'S') || isALPHA_FOLD_EQ(*s, 'Q')) {
772 /* snan, qNaN */
773 /* XXX do something with the snan/qnan difference */
774 s++; if (s == send) return 0;
775 }
776
777 if (isALPHA_FOLD_EQ(*s, 'N')) {
778 s++; if (s == send || isALPHA_FOLD_NE(*s, 'A')) return 0;
779 s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0;
780 s++;
781
782 flags |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT;
783
784 /* NaN can be followed by various stuff (NaNQ, NaNS), but
785 * there are also multiple different NaN values, and some
786 * implementations output the "payload" values,
787 * e.g. NaN123, NAN(abc), while some legacy implementations
788 * have weird stuff like NaN%. */
789 if (isALPHA_FOLD_EQ(*s, 'q') ||
790 isALPHA_FOLD_EQ(*s, 's')) {
791 /* "nanq" or "nans" are ok, though generating
792 * these portably is tricky. */
793 s++;
794 }
795 if (*s == '(') {
796 /* C99 style "nan(123)" or Perlish equivalent "nan($uv)". */
797 const char *t;
798 s++;
799 if (s == send) {
800 return flags | IS_NUMBER_TRAILING;
801 }
802 t = s + 1;
803 while (t < send && *t && *t != ')') {
804 t++;
805 }
806 if (t == send) {
807 return flags | IS_NUMBER_TRAILING;
808 }
809 if (*t == ')') {
810 int nantype;
811 UV nanval;
812 if (s[0] == '0' && s + 2 < t &&
813 isALPHA_FOLD_EQ(s[1], 'x') &&
814 isXDIGIT(s[2])) {
815 STRLEN len = t - s;
816 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
817 nanval = grok_hex(s, &len, &flags, NULL);
818 if ((flags & PERL_SCAN_GREATER_THAN_UV_MAX)) {
819 nantype = 0;
820 } else {
821 nantype = IS_NUMBER_IN_UV;
822 }
823 s += len;
824 } else if (s[0] == '0' && s + 2 < t &&
825 isALPHA_FOLD_EQ(s[1], 'b') &&
826 (s[2] == '0' || s[2] == '1')) {
827 STRLEN len = t - s;
828 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES;
829 nanval = grok_bin(s, &len, &flags, NULL);
830 if ((flags & PERL_SCAN_GREATER_THAN_UV_MAX)) {
831 nantype = 0;
832 } else {
833 nantype = IS_NUMBER_IN_UV;
834 }
835 s += len;
836 } else {
837 const char *u;
838 nantype =
839 grok_number_flags(s, t - s, &nanval,
840 PERL_SCAN_TRAILING |
841 PERL_SCAN_ALLOW_UNDERSCORES);
842 /* Unfortunately grok_number_flags() doesn't
843 * tell how far we got and the ')' will always
844 * be "trailing", so we need to double-check
845 * whether we had something dubious. */
846 for (u = s; u < t; u++) {
847 if (!isDIGIT(*u)) {
848 flags |= IS_NUMBER_TRAILING;
849 break;
850 }
851 }
852 s = u;
853 }
854
855 /* XXX Doesn't do octal: nan("0123").
856 * Probably not a big loss. */
857
858 if ((nantype & IS_NUMBER_NOT_INT) ||
859 !(nantype && IS_NUMBER_IN_UV)) {
860 /* XXX the nanval is currently unused, that is,
861 * not inserted as the NaN payload of the NV.
862 * But the above code already parses the C99
863 * nan(...) format. See below, and see also
864 * the nan() in POSIX.xs.
865 *
866 * Certain configuration combinations where
867 * NVSIZE is greater than UVSIZE mean that
868 * a single UV cannot contain all the possible
869 * NaN payload bits. There would need to be
870 * some more generic syntax than "nan($uv)".
871 *
872 * Issues to keep in mind:
873 *
874 * (1) In most common cases there would
875 * not be an integral number of bytes that
876 * could be set, only a certain number of bits.
877 * For example for the common case of
878 * NVSIZE == UVSIZE == 8 there is room for 52
879 * bits in the payload, but the most significant
880 * bit is commonly reserved for the
881 * signaling/quiet bit, leaving 51 bits.
882 * Furthermore, the C99 nan() is supposed
883 * to generate quiet NaNs, so it is doubtful
884 * whether it should be able to generate
885 * signaling NaNs. For the x86 80-bit doubles
886 * (if building a long double Perl) there would
887 * be 62 bits (s/q bit being the 63rd).
888 *
889 * (2) Endianness of the payload bits. If the
890 * payload is specified as an UV, the low-order
891 * bits of the UV are naturally little-endianed
892 * (rightmost) bits of the payload. The endianness
893 * of UVs and NVs can be different. */
894 return 0;
895 }
896 if (s < t) {
897 flags |= IS_NUMBER_TRAILING;
898 }
899 } else {
900 /* Looked like nan(...), but no close paren. */
901 flags |= IS_NUMBER_TRAILING;
902 }
903 } else {
904 while (s < send && isSPACE(*s))
905 s++;
906 if (s < send && *s) {
907 /* Note that we here implicitly accept (parse as
908 * "nan", but with warnings) also any other weird
909 * trailing stuff for "nan". In the above we just
910 * check that if we got the C99-style "nan(...)",
911 * the "..." looks sane.
912 * If in future we accept more ways of specifying
913 * the nan payload, the accepting would happen around
914 * here. */
915 flags |= IS_NUMBER_TRAILING;
916 }
917 }
918 s = send;
919 }
920 else
921 return 0;
ff4eb398
JH
922 }
923
b489e20f
JH
924 while (s < send && isSPACE(*s))
925 s++;
926
a5dc2484
JH
927#else
928 PERL_UNUSED_ARG(send);
929#endif /* #if defined(NV_INF) || defined(NV_NAN) */
a1fe7cea
JH
930 *sp = s;
931 return flags;
ff4eb398
JH
932}
933
13393a5e 934/*
3823048b 935=for apidoc grok_number_flags
13393a5e
JH
936
937Recognise (or not) a number. The type of the number is returned
938(0 if unrecognised), otherwise it is a bit-ORed combination of
796b6530
KW
939C<IS_NUMBER_IN_UV>, C<IS_NUMBER_GREATER_THAN_UV_MAX>, C<IS_NUMBER_NOT_INT>,
940C<IS_NUMBER_NEG>, C<IS_NUMBER_INFINITY>, C<IS_NUMBER_NAN> (defined in perl.h).
941
942If the value of the number can fit in a UV, it is returned in C<*valuep>.
943C<IS_NUMBER_IN_UV> will be set to indicate that C<*valuep> is valid, C<IS_NUMBER_IN_UV>
944will never be set unless C<*valuep> is valid, but C<*valuep> may have been assigned
945to during processing even though C<IS_NUMBER_IN_UV> is not set on return.
946If C<valuep> is C<NULL>, C<IS_NUMBER_IN_UV> will be set for the same cases as when
947C<valuep> is non-C<NULL>, but no actual assignment (or SEGV) will occur.
948
949C<IS_NUMBER_NOT_INT> will be set with C<IS_NUMBER_IN_UV> if trailing decimals were
950seen (in which case C<*valuep> gives the true value truncated to an integer), and
951C<IS_NUMBER_NEG> if the number is negative (in which case C<*valuep> holds the
952absolute value). C<IS_NUMBER_IN_UV> is not set if e notation was used or the
13393a5e
JH
953number is larger than a UV.
954
955C<flags> allows only C<PERL_SCAN_TRAILING>, which allows for trailing
956non-numeric text on an otherwise successful I<grok>, setting
957C<IS_NUMBER_TRAILING> on the result.
958
959=for apidoc grok_number
960
796b6530 961Identical to C<grok_number_flags()> with C<flags> set to zero.
13393a5e
JH
962
963=cut
964 */
965int
966Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep)
967{
968 PERL_ARGS_ASSERT_GROK_NUMBER;
969
970 return grok_number_flags(pv, len, valuep, 0);
971}
972
945b524a
JH
973static const UV uv_max_div_10 = UV_MAX / 10;
974static const U8 uv_max_mod_10 = UV_MAX % 10;
975
3f7602fa 976int
3823048b 977Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags)
3f7602fa 978{
60939fb8 979 const char *s = pv;
c4420975 980 const char * const send = pv + len;
ae776a2c 981 const char *d;
60939fb8 982 int numtype = 0;
60939fb8 983
3823048b 984 PERL_ARGS_ASSERT_GROK_NUMBER_FLAGS;
7918f24d 985
60939fb8
NC
986 while (s < send && isSPACE(*s))
987 s++;
988 if (s == send) {
989 return 0;
990 } else if (*s == '-') {
991 s++;
992 numtype = IS_NUMBER_NEG;
993 }
994 else if (*s == '+')
aa42a541 995 s++;
60939fb8
NC
996
997 if (s == send)
998 return 0;
999
ae776a2c 1000 /* The first digit (after optional sign): note that might
8c12dc63 1001 * also point to "infinity" or "nan", or "1.#INF". */
ae776a2c
JH
1002 d = s;
1003
8c12dc63 1004 /* next must be digit or the radix separator or beginning of infinity/nan */
60939fb8
NC
1005 if (isDIGIT(*s)) {
1006 /* UVs are at least 32 bits, so the first 9 decimal digits cannot
1007 overflow. */
1008 UV value = *s - '0';
1009 /* This construction seems to be more optimiser friendly.
1010 (without it gcc does the isDIGIT test and the *s - '0' separately)
1011 With it gcc on arm is managing 6 instructions (6 cycles) per digit.
1012 In theory the optimiser could deduce how far to unroll the loop
1013 before checking for overflow. */
58bb9ec3
NC
1014 if (++s < send) {
1015 int digit = *s - '0';
f85b645f 1016 if (inRANGE(digit, 0, 9)) {
60939fb8 1017 value = value * 10 + digit;
58bb9ec3
NC
1018 if (++s < send) {
1019 digit = *s - '0';
f85b645f 1020 if (inRANGE(digit, 0, 9)) {
60939fb8 1021 value = value * 10 + digit;
58bb9ec3
NC
1022 if (++s < send) {
1023 digit = *s - '0';
f85b645f 1024 if (inRANGE(digit, 0, 9)) {
60939fb8 1025 value = value * 10 + digit;
58bb9ec3
NC
1026 if (++s < send) {
1027 digit = *s - '0';
f85b645f 1028 if (inRANGE(digit, 0, 9)) {
60939fb8 1029 value = value * 10 + digit;
58bb9ec3
NC
1030 if (++s < send) {
1031 digit = *s - '0';
f85b645f 1032 if (inRANGE(digit, 0, 9)) {
60939fb8 1033 value = value * 10 + digit;
58bb9ec3
NC
1034 if (++s < send) {
1035 digit = *s - '0';
f85b645f 1036 if (inRANGE(digit, 0, 9)) {
60939fb8 1037 value = value * 10 + digit;
58bb9ec3
NC
1038 if (++s < send) {
1039 digit = *s - '0';
f85b645f 1040 if (inRANGE(digit, 0, 9)) {
60939fb8 1041 value = value * 10 + digit;
58bb9ec3
NC
1042 if (++s < send) {
1043 digit = *s - '0';
f85b645f 1044 if (inRANGE(digit, 0, 9)) {
60939fb8 1045 value = value * 10 + digit;
58bb9ec3 1046 if (++s < send) {
60939fb8
NC
1047 /* Now got 9 digits, so need to check
1048 each time for overflow. */
58bb9ec3 1049 digit = *s - '0';
f85b645f 1050 while ( inRANGE(digit, 0, 9)
945b524a
JH
1051 && (value < uv_max_div_10
1052 || (value == uv_max_div_10
1053 && digit <= uv_max_mod_10))) {
60939fb8 1054 value = value * 10 + digit;
58bb9ec3
NC
1055 if (++s < send)
1056 digit = *s - '0';
60939fb8
NC
1057 else
1058 break;
1059 }
f85b645f 1060 if (inRANGE(digit, 0, 9)
51bd16da 1061 && (s < send)) {
60939fb8
NC
1062 /* value overflowed.
1063 skip the remaining digits, don't
1064 worry about setting *valuep. */
1065 do {
1066 s++;
1067 } while (s < send && isDIGIT(*s));
1068 numtype |=
1069 IS_NUMBER_GREATER_THAN_UV_MAX;
1070 goto skip_value;
1071 }
1072 }
1073 }
98994639 1074 }
60939fb8
NC
1075 }
1076 }
1077 }
1078 }
1079 }
1080 }
1081 }
1082 }
1083 }
1084 }
1085 }
98994639 1086 }
60939fb8 1087 }
98994639 1088 }
60939fb8
NC
1089 numtype |= IS_NUMBER_IN_UV;
1090 if (valuep)
1091 *valuep = value;
1092
1093 skip_value:
1094 if (GROK_NUMERIC_RADIX(&s, send)) {
1095 numtype |= IS_NUMBER_NOT_INT;
1096 while (s < send && isDIGIT(*s)) /* optional digits after the radix */
1097 s++;
98994639 1098 }
60939fb8
NC
1099 }
1100 else if (GROK_NUMERIC_RADIX(&s, send)) {
1101 numtype |= IS_NUMBER_NOT_INT | IS_NUMBER_IN_UV; /* valuep assigned below */
1102 /* no digits before the radix means we need digits after it */
1103 if (s < send && isDIGIT(*s)) {
1104 do {
1105 s++;
1106 } while (s < send && isDIGIT(*s));
1107 if (valuep) {
1108 /* integer approximation is valid - it's 0. */
1109 *valuep = 0;
1110 }
98994639 1111 }
60939fb8 1112 else
ae776a2c 1113 return 0;
ff4eb398 1114 }
60939fb8 1115
926f5fc6 1116 if (s > d && s < send) {
60939fb8 1117 /* we can have an optional exponent part */
305b8651 1118 if (isALPHA_FOLD_EQ(*s, 'e')) {
60939fb8
NC
1119 s++;
1120 if (s < send && (*s == '-' || *s == '+'))
1121 s++;
1122 if (s < send && isDIGIT(*s)) {
1123 do {
1124 s++;
1125 } while (s < send && isDIGIT(*s));
1126 }
3f7602fa
TC
1127 else if (flags & PERL_SCAN_TRAILING)
1128 return numtype | IS_NUMBER_TRAILING;
60939fb8 1129 else
3f7602fa
TC
1130 return 0;
1131
1132 /* The only flag we keep is sign. Blow away any "it's UV" */
1133 numtype &= IS_NUMBER_NEG;
1134 numtype |= IS_NUMBER_NOT_INT;
60939fb8
NC
1135 }
1136 }
1137 while (s < send && isSPACE(*s))
1138 s++;
1139 if (s >= send)
aa8b85de 1140 return numtype;
b59bf0b2 1141 if (memEQs(pv, len, "0 but true")) {
60939fb8
NC
1142 if (valuep)
1143 *valuep = 0;
1144 return IS_NUMBER_IN_UV;
1145 }
8c12dc63
JH
1146 /* We could be e.g. at "Inf" or "NaN", or at the "#" of "1.#INF". */
1147 if ((s + 2 < send) && strchr("inqs#", toFOLD(*s))) {
1148 /* Really detect inf/nan. Start at d, not s, since the above
1149 * code might have already consumed the "1." or "1". */
7eff3d39 1150 const int infnan = Perl_grok_infnan(aTHX_ &d, send);
8c12dc63
JH
1151 if ((infnan & IS_NUMBER_INFINITY)) {
1152 return (numtype | infnan); /* Keep sign for infinity. */
1153 }
1154 else if ((infnan & IS_NUMBER_NAN)) {
1155 return (numtype | infnan) & ~IS_NUMBER_NEG; /* Clear sign for nan. */
1156 }
1157 }
3f7602fa
TC
1158 else if (flags & PERL_SCAN_TRAILING) {
1159 return numtype | IS_NUMBER_TRAILING;
1160 }
1161
60939fb8 1162 return 0;
98994639
HS
1163}
1164
6313e544 1165/*
5d4a52b5 1166=for apidoc grok_atoUV
6313e544 1167
5d4a52b5 1168parse a string, looking for a decimal unsigned integer.
338aa8b0 1169
5d4a52b5
KW
1170On entry, C<pv> points to the beginning of the string;
1171C<valptr> points to a UV that will receive the converted value, if found;
1172C<endptr> is either NULL or points to a variable that points to one byte
1173beyond the point in C<pv> that this routine should examine.
1174If C<endptr> is NULL, C<pv> is assumed to be NUL-terminated.
f4379102 1175
5d4a52b5
KW
1176Returns FALSE if C<pv> doesn't represent a valid unsigned integer value (with
1177no leading zeros). Otherwise it returns TRUE, and sets C<*valptr> to that
1178value.
6313e544 1179
5d4a52b5
KW
1180If you constrain the portion of C<pv> that is looked at by this function (by
1181passing a non-NULL C<endptr>), and if the intial bytes of that portion form a
1182valid value, it will return TRUE, setting C<*endptr> to the byte following the
1183final digit of the value. But if there is no constraint at what's looked at,
1184all of C<pv> must be valid in order for TRUE to be returned.
6313e544 1185
5d4a52b5 1186The only characters this accepts are the decimal digits '0'..'9'.
338aa8b0 1187
5d4a52b5
KW
1188As opposed to L<atoi(3)> or L<strtol(3)>, C<grok_atoUV> does NOT allow optional
1189leading whitespace, nor negative inputs. If such features are required, the
1190calling code needs to explicitly implement those.
6313e544 1191
5d4a52b5
KW
1192Note that this function returns FALSE for inputs that would overflow a UV,
1193or have leading zeros. Thus a single C<0> is accepted, but not C<00> nor
1194C<01>, C<002>, I<etc>.
1195
1196Background: C<atoi> has severe problems with illegal inputs, it cannot be
d62b8c6a 1197used for incremental parsing, and therefore should be avoided
5d4a52b5 1198C<atoi> and C<strtol> are also affected by locale settings, which can also be
d62b8c6a
JH
1199seen as a bug (global state controlled by user environment).
1200
238217e5
JK
1201=cut
1202
6313e544
JH
1203*/
1204
22ff3130
HS
1205bool
1206Perl_grok_atoUV(const char *pv, UV *valptr, const char** endptr)
6313e544
JH
1207{
1208 const char* s = pv;
1209 const char** eptr;
1210 const char* end2; /* Used in case endptr is NULL. */
22ff3130 1211 UV val = 0; /* The parsed value. */
6313e544 1212
22ff3130 1213 PERL_ARGS_ASSERT_GROK_ATOUV;
6313e544 1214
5d4a52b5
KW
1215 if (endptr) {
1216 eptr = endptr;
1217 }
1218 else {
1219 end2 = s + strlen(s);
1220 eptr = &end2;
1221 }
1222
1223 if ( *eptr <= s
1224 || ! isDIGIT(*s))
1225 {
1226 return FALSE;
1227 }
1228
97d95d46
KW
1229 /* Single-digit inputs are quite common. */
1230 val = *s++ - '0';
1231 if (s < *eptr && isDIGIT(*s)) {
1232 /* Fail on extra leading zeros. */
1233 if (val == 0)
1234 return FALSE;
1235 while (s < *eptr && isDIGIT(*s)) {
1236 /* This could be unrolled like in grok_number(), but
1237 * the expected uses of this are not speed-needy, and
1238 * unlikely to need full 64-bitness. */
1239 const U8 digit = *s++ - '0';
1240 if (val < uv_max_div_10 ||
1241 (val == uv_max_div_10 && digit <= uv_max_mod_10)) {
1242 val = val * 10 + digit;
1243 } else {
22ff3130 1244 return FALSE;
6313e544
JH
1245 }
1246 }
97d95d46
KW
1247 }
1248
5d4a52b5
KW
1249 if (endptr == NULL) {
1250 if (*s) {
1251 return FALSE; /* If endptr is NULL, no trailing non-digits allowed. */
1252 }
1253 }
1254 else {
1255 *endptr = s;
75feedba 1256 }
97d95d46 1257
22ff3130
HS
1258 *valptr = val;
1259 return TRUE;
6313e544
JH
1260}
1261
ce6f496d 1262#ifndef Perl_strtod
4801ca72 1263STATIC NV
98994639
HS
1264S_mulexp10(NV value, I32 exponent)
1265{
1266 NV result = 1.0;
1267 NV power = 10.0;
1268 bool negative = 0;
1269 I32 bit;
1270
1271 if (exponent == 0)
1272 return value;
659c4b96
DM
1273 if (value == 0)
1274 return (NV)0;
87032ba1 1275
24866caa 1276 /* On OpenVMS VAX we by default use the D_FLOAT double format,
67597c89 1277 * and that format does not have *easy* capabilities [1] for
19c1206d
KW
1278 * overflowing doubles 'silently' as IEEE fp does. We also need
1279 * to support G_FLOAT on both VAX and Alpha, and though the exponent
1280 * range is much larger than D_FLOAT it still doesn't do silent
1281 * overflow. Therefore we need to detect early whether we would
1282 * overflow (this is the behaviour of the native string-to-float
24866caa 1283 * conversion routines, and therefore of native applications, too).
67597c89 1284 *
24866caa
CB
1285 * [1] Trying to establish a condition handler to trap floating point
1286 * exceptions is not a good idea. */
87032ba1
JH
1287
1288 /* In UNICOS and in certain Cray models (such as T90) there is no
1289 * IEEE fp, and no way at all from C to catch fp overflows gracefully.
1290 * There is something you can do if you are willing to use some
1291 * inline assembler: the instruction is called DFI-- but that will
1292 * disable *all* floating point interrupts, a little bit too large
1293 * a hammer. Therefore we need to catch potential overflows before
1294 * it's too late. */
353813d9 1295
a7157111 1296#if ((defined(VMS) && !defined(_IEEE_FP)) || defined(_UNICOS) || defined(DOUBLE_IS_VAX_FLOAT)) && defined(NV_MAX_10_EXP)
353813d9 1297 STMT_START {
c4420975 1298 const NV exp_v = log10(value);
353813d9
HS
1299 if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP)
1300 return NV_MAX;
1301 if (exponent < 0) {
1302 if (-(exponent + exp_v) >= NV_MAX_10_EXP)
1303 return 0.0;
1304 while (-exponent >= NV_MAX_10_EXP) {
1305 /* combination does not overflow, but 10^(-exponent) does */
1306 value /= 10;
1307 ++exponent;
1308 }
1309 }
1310 } STMT_END;
87032ba1
JH
1311#endif
1312
353813d9
HS
1313 if (exponent < 0) {
1314 negative = 1;
1315 exponent = -exponent;
b27804d8
DM
1316#ifdef NV_MAX_10_EXP
1317 /* for something like 1234 x 10^-309, the action of calculating
1318 * the intermediate value 10^309 then returning 1234 / (10^309)
1319 * will fail, since 10^309 becomes infinity. In this case try to
1320 * refactor it as 123 / (10^308) etc.
1321 */
1322 while (value && exponent > NV_MAX_10_EXP) {
1323 exponent--;
1324 value /= 10;
1325 }
48853916
JH
1326 if (value == 0.0)
1327 return value;
b27804d8 1328#endif
353813d9 1329 }
c62e754c
JH
1330#if defined(__osf__)
1331 /* Even with cc -ieee + ieee_set_fp_control(IEEE_TRAP_ENABLE_INV)
1332 * Tru64 fp behavior on inf/nan is somewhat broken. Another way
1333 * to do this would be ieee_set_fp_control(IEEE_TRAP_ENABLE_OVF)
1334 * but that breaks another set of infnan.t tests. */
1335# define FP_OVERFLOWS_TO_ZERO
1336#endif
98994639
HS
1337 for (bit = 1; exponent; bit <<= 1) {
1338 if (exponent & bit) {
1339 exponent ^= bit;
1340 result *= power;
c62e754c
JH
1341#ifdef FP_OVERFLOWS_TO_ZERO
1342 if (result == 0)
a7157111 1343# ifdef NV_INF
c62e754c 1344 return value < 0 ? -NV_INF : NV_INF;
a7157111
JH
1345# else
1346 return value < 0 ? -FLT_MAX : FLT_MAX;
1347# endif
c62e754c 1348#endif
236f0012 1349 /* Floating point exceptions are supposed to be turned off,
19c1206d 1350 * but if we're obviously done, don't risk another iteration.
236f0012
CB
1351 */
1352 if (exponent == 0) break;
98994639
HS
1353 }
1354 power *= power;
1355 }
1356 return negative ? value / result : value * result;
1357}
ce6f496d 1358#endif /* #ifndef Perl_strtod */
98994639 1359
ce6f496d 1360#ifdef Perl_strtod
b93d1309 1361# define ATOF(s, x) my_atof2(s, &x)
f7b64c80 1362#else
b93d1309 1363# define ATOF(s, x) Perl_atof2(s, x)
f7b64c80 1364#endif
b93d1309 1365
98994639
HS
1366NV
1367Perl_my_atof(pTHX_ const char* s)
1368{
f720c878
KW
1369 /* 's' must be NUL terminated */
1370
98994639 1371 NV x = 0.0;
9eda1ea6
KW
1372
1373 PERL_ARGS_ASSERT_MY_ATOF;
1374
b93d1309 1375#if ! defined(USE_LOCALE_NUMERIC)
9eda1ea6 1376
b93d1309 1377 ATOF(s, x);
9eda1ea6
KW
1378
1379#else
7918f24d 1380
a2287a13 1381 {
67d796ae
KW
1382 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
1383 STORE_LC_NUMERIC_SET_TO_NEEDED();
fdf55d20
KW
1384 if (! (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))) {
1385 ATOF(s,x);
1386 }
1387 else {
19c1206d 1388
e4850248
KW
1389 /* Look through the string for the first thing that looks like a
1390 * decimal point: either the value in the current locale or the
1391 * standard fallback of '.'. The one which appears earliest in the
1392 * input string is the one that we should have atof look for. Note
1393 * that we have to determine this beforehand because on some
1394 * systems, Perl_atof2 is just a wrapper around the system's atof.
1395 * */
1ae85f6c
KW
1396 const char * const standard_pos = strchr(s, '.');
1397 const char * const local_pos
1398 = strstr(s, SvPV_nolen(PL_numeric_radix_sv));
1399 const bool use_standard_radix
1400 = standard_pos && (!local_pos || standard_pos < local_pos);
78787052 1401
665873e9 1402 if (use_standard_radix) {
e4850248 1403 SET_NUMERIC_STANDARD();
665873e9
KW
1404 LOCK_LC_NUMERIC_STANDARD();
1405 }
78787052 1406
b93d1309 1407 ATOF(s,x);
78787052 1408
665873e9
KW
1409 if (use_standard_radix) {
1410 UNLOCK_LC_NUMERIC_STANDARD();
67d796ae 1411 SET_NUMERIC_UNDERLYING();
665873e9 1412 }
e4850248 1413 }
a2287a13
KW
1414 RESTORE_LC_NUMERIC();
1415 }
9eda1ea6 1416
98994639 1417#endif
9eda1ea6 1418
98994639
HS
1419 return x;
1420}
1421
a7157111 1422#if defined(NV_INF) || defined(NV_NAN)
3c81f0b3 1423
829757a4 1424static char*
5563f457 1425S_my_atof_infnan(pTHX_ const char* s, bool negative, const char* send, NV* value)
829757a4
JH
1426{
1427 const char *p0 = negative ? s - 1 : s;
1428 const char *p = p0;
7eff3d39 1429 const int infnan = grok_infnan(&p, send);
829757a4
JH
1430 if (infnan && p != p0) {
1431 /* If we can generate inf/nan directly, let's do so. */
1432#ifdef NV_INF
1433 if ((infnan & IS_NUMBER_INFINITY)) {
3823048b 1434 *value = (infnan & IS_NUMBER_NEG) ? -NV_INF: NV_INF;
829757a4
JH
1435 return (char*)p;
1436 }
1437#endif
1438#ifdef NV_NAN
1439 if ((infnan & IS_NUMBER_NAN)) {
3823048b 1440 *value = NV_NAN;
829757a4
JH
1441 return (char*)p;
1442 }
1443#endif
1444#ifdef Perl_strtod
68611e6f 1445 /* If still here, we didn't have either NV_INF or NV_NAN,
829757a4
JH
1446 * and can try falling back to native strtod/strtold.
1447 *
1448 * The native interface might not recognize all the possible
1449 * inf/nan strings Perl recognizes. What we can try
1450 * is to try faking the input. We will try inf/-inf/nan
1451 * as the most promising/portable input. */
1452 {
6d37e916 1453 const char* fake = "silence compiler warning";
829757a4
JH
1454 char* endp;
1455 NV nv;
a7157111 1456#ifdef NV_INF
829757a4
JH
1457 if ((infnan & IS_NUMBER_INFINITY)) {
1458 fake = ((infnan & IS_NUMBER_NEG)) ? "-inf" : "inf";
1459 }
a7157111
JH
1460#endif
1461#ifdef NV_NAN
1462 if ((infnan & IS_NUMBER_NAN)) {
829757a4
JH
1463 fake = "nan";
1464 }
a7157111 1465#endif
6d37e916 1466 assert(strNE(fake, "silence compiler warning"));
9ec8aea5 1467 nv = S_strtod(aTHX_ fake, &endp);
829757a4 1468 if (fake != endp) {
a7157111 1469#ifdef NV_INF
829757a4 1470 if ((infnan & IS_NUMBER_INFINITY)) {
a7157111 1471# ifdef Perl_isinf
829757a4
JH
1472 if (Perl_isinf(nv))
1473 *value = nv;
a7157111 1474# else
829757a4
JH
1475 /* last resort, may generate SIGFPE */
1476 *value = Perl_exp((NV)1e9);
1477 if ((infnan & IS_NUMBER_NEG))
1478 *value = -*value;
a7157111 1479# endif
829757a4
JH
1480 return (char*)p; /* p, not endp */
1481 }
a7157111
JH
1482#endif
1483#ifdef NV_NAN
1484 if ((infnan & IS_NUMBER_NAN)) {
1485# ifdef Perl_isnan
829757a4
JH
1486 if (Perl_isnan(nv))
1487 *value = nv;
a7157111 1488# else
829757a4
JH
1489 /* last resort, may generate SIGFPE */
1490 *value = Perl_log((NV)-1.0);
a7157111 1491# endif
829757a4 1492 return (char*)p; /* p, not endp */
a7157111 1493#endif
829757a4
JH
1494 }
1495 }
1496 }
1497#endif /* #ifdef Perl_strtod */
1498 }
1499 return NULL;
1500}
1501
a7157111
JH
1502#endif /* if defined(NV_INF) || defined(NV_NAN) */
1503
98994639
HS
1504char*
1505Perl_my_atof2(pTHX_ const char* orig, NV* value)
1506{
6928bedc
KW
1507 PERL_ARGS_ASSERT_MY_ATOF2;
1508 return my_atof3(orig, value, 0);
1509}
1510
1511char*
16411967 1512Perl_my_atof3(pTHX_ const char* orig, NV* value, const STRLEN len)
6928bedc 1513{
e1ec3a88 1514 const char* s = orig;
a4eca1d4 1515 NV result[3] = {0.0, 0.0, 0.0};
ce6f496d 1516#if defined(USE_PERL_ATOF) || defined(Perl_strtod)
6928bedc
KW
1517 const char* send = s + ((len != 0)
1518 ? len
1519 : strlen(orig)); /* one past the last */
a4eca1d4
JH
1520 bool negative = 0;
1521#endif
ce6f496d 1522#if defined(USE_PERL_ATOF) && !defined(Perl_strtod)
a4eca1d4 1523 UV accumulator[2] = {0,0}; /* before/after dp */
8194bf88 1524 bool seen_digit = 0;
20f6aaab
AS
1525 I32 exp_adjust[2] = {0,0};
1526 I32 exp_acc[2] = {-1, -1};
1527 /* the current exponent adjust for the accumulators */
98994639 1528 I32 exponent = 0;
8194bf88 1529 I32 seen_dp = 0;
20f6aaab
AS
1530 I32 digit = 0;
1531 I32 old_digit = 0;
8194bf88 1532 I32 sig_digits = 0; /* noof significant digits seen so far */
a4eca1d4 1533#endif
8194bf88 1534
ce6f496d 1535#if defined(USE_PERL_ATOF) || defined(Perl_strtod)
6928bedc 1536 PERL_ARGS_ASSERT_MY_ATOF3;
7918f24d 1537
a4eca1d4 1538 /* leading whitespace */
6928bedc 1539 while (s < send && isSPACE(*s))
a4eca1d4
JH
1540 ++s;
1541
1542 /* sign */
1543 switch (*s) {
1544 case '-':
1545 negative = 1;
1546 /* FALLTHROUGH */
1547 case '+':
1548 ++s;
1549 }
1550#endif
1551
ce6f496d 1552#ifdef Perl_strtod
a4eca1d4
JH
1553 {
1554 char* endp;
d94e901a
KW
1555 char* copy = NULL;
1556
adc55e02 1557 if ((endp = S_my_atof_infnan(aTHX_ s, negative, send, value)))
a4eca1d4 1558 return endp;
d94e901a 1559
14d26b44
TC
1560 /* strtold() accepts 0x-prefixed hex and in POSIX implementations,
1561 0b-prefixed binary numbers, which is backward incompatible
1562 */
1563 if ((len == 0 || len >= 2) && *s == '0' &&
1564 (isALPHA_FOLD_EQ(s[1], 'x') || isALPHA_FOLD_EQ(s[1], 'b'))) {
1565 *value = 0;
1566 return (char *)s+1;
1567 }
1568
d94e901a
KW
1569 /* If the length is passed in, the input string isn't NUL-terminated,
1570 * and in it turns out the function below assumes it is; therefore we
1571 * create a copy and NUL-terminate that */
1572 if (len) {
1573 Newx(copy, len + 1, char);
1574 Copy(orig, copy, len, char);
1575 copy[len] = '\0';
1576 s = copy + (s - orig);
1577 }
1578
9ec8aea5 1579 result[2] = S_strtod(aTHX_ s, &endp);
d94e901a
KW
1580
1581 /* If we created a copy, 'endp' is in terms of that. Convert back to
1582 * the original */
1583 if (copy) {
aac39b03 1584 s = (s - copy) + (char *) orig;
d94e901a
KW
1585 endp = (endp - copy) + (char *) orig;
1586 Safefree(copy);
1587 }
1588
a4eca1d4
JH
1589 if (s != endp) {
1590 *value = negative ? -result[2] : result[2];
1591 return endp;
1592 }
1593 return NULL;
1594 }
1595#elif defined(USE_PERL_ATOF)
1596
8194bf88
DM
1597/* There is no point in processing more significant digits
1598 * than the NV can hold. Note that NV_DIG is a lower-bound value,
1599 * while we need an upper-bound value. We add 2 to account for this;
1600 * since it will have been conservative on both the first and last digit.
1601 * For example a 32-bit mantissa with an exponent of 4 would have
1602 * exact values in the set
1603 * 4
1604 * 8
1605 * ..
1606 * 17179869172
1607 * 17179869176
1608 * 17179869180
1609 *
1610 * where for the purposes of calculating NV_DIG we would have to discount
1611 * both the first and last digit, since neither can hold all values from
1612 * 0..9; but for calculating the value we must examine those two digits.
1613 */
ffa277e5
AS
1614#ifdef MAX_SIG_DIG_PLUS
1615 /* It is not necessarily the case that adding 2 to NV_DIG gets all the
1616 possible digits in a NV, especially if NVs are not IEEE compliant
1617 (e.g., long doubles on IRIX) - Allen <allens@cpan.org> */
1618# define MAX_SIG_DIGITS (NV_DIG+MAX_SIG_DIG_PLUS)
1619#else
1620# define MAX_SIG_DIGITS (NV_DIG+2)
1621#endif
8194bf88
DM
1622
1623/* the max number we can accumulate in a UV, and still safely do 10*N+9 */
1624#define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10))
98994639 1625
a5dc2484 1626#if defined(NV_INF) || defined(NV_NAN)
ae776a2c 1627 {
7eff3d39 1628 char* endp;
5563f457 1629 if ((endp = S_my_atof_infnan(aTHX_ s, negative, send, value)))
7eff3d39 1630 return endp;
ae776a2c 1631 }
a5dc2484 1632#endif
2b54f59f 1633
8194bf88
DM
1634 /* we accumulate digits into an integer; when this becomes too
1635 * large, we add the total to NV and start again */
98994639 1636
6928bedc 1637 while (s < send) {
8194bf88
DM
1638 if (isDIGIT(*s)) {
1639 seen_digit = 1;
20f6aaab 1640 old_digit = digit;
8194bf88 1641 digit = *s++ - '0';
20f6aaab
AS
1642 if (seen_dp)
1643 exp_adjust[1]++;
98994639 1644
8194bf88
DM
1645 /* don't start counting until we see the first significant
1646 * digit, eg the 5 in 0.00005... */
1647 if (!sig_digits && digit == 0)
1648 continue;
1649
1650 if (++sig_digits > MAX_SIG_DIGITS) {
98994639 1651 /* limits of precision reached */
20f6aaab
AS
1652 if (digit > 5) {
1653 ++accumulator[seen_dp];
1654 } else if (digit == 5) {
1655 if (old_digit % 2) { /* round to even - Allen */
1656 ++accumulator[seen_dp];
1657 }
1658 }
1659 if (seen_dp) {
1660 exp_adjust[1]--;
1661 } else {
1662 exp_adjust[0]++;
1663 }
8194bf88 1664 /* skip remaining digits */
6928bedc 1665 while (s < send && isDIGIT(*s)) {
98994639 1666 ++s;
20f6aaab
AS
1667 if (! seen_dp) {
1668 exp_adjust[0]++;
1669 }
98994639
HS
1670 }
1671 /* warn of loss of precision? */
98994639 1672 }
8194bf88 1673 else {
20f6aaab 1674 if (accumulator[seen_dp] > MAX_ACCUMULATE) {
8194bf88 1675 /* add accumulator to result and start again */
20f6aaab
AS
1676 result[seen_dp] = S_mulexp10(result[seen_dp],
1677 exp_acc[seen_dp])
1678 + (NV)accumulator[seen_dp];
1679 accumulator[seen_dp] = 0;
1680 exp_acc[seen_dp] = 0;
98994639 1681 }
20f6aaab
AS
1682 accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit;
1683 ++exp_acc[seen_dp];
98994639 1684 }
8194bf88 1685 }
e1ec3a88 1686 else if (!seen_dp && GROK_NUMERIC_RADIX(&s, send)) {
8194bf88 1687 seen_dp = 1;
20f6aaab 1688 if (sig_digits > MAX_SIG_DIGITS) {
6928bedc 1689 while (s < send && isDIGIT(*s)) {
20f6aaab 1690 ++s;
9604fbf0 1691 }
20f6aaab
AS
1692 break;
1693 }
8194bf88
DM
1694 }
1695 else {
1696 break;
98994639
HS
1697 }
1698 }
1699
20f6aaab
AS
1700 result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0];
1701 if (seen_dp) {
1702 result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1];
1703 }
98994639 1704
6928bedc 1705 if (s < send && seen_digit && (isALPHA_FOLD_EQ(*s, 'e'))) {
98994639
HS
1706 bool expnegative = 0;
1707
1708 ++s;
1709 switch (*s) {
1710 case '-':
1711 expnegative = 1;
924ba076 1712 /* FALLTHROUGH */
98994639
HS
1713 case '+':
1714 ++s;
1715 }
6928bedc 1716 while (s < send && isDIGIT(*s))
98994639
HS
1717 exponent = exponent * 10 + (*s++ - '0');
1718 if (expnegative)
1719 exponent = -exponent;
1720 }
1721
1722 /* now apply the exponent */
20f6aaab
AS
1723
1724 if (seen_dp) {
1725 result[2] = S_mulexp10(result[0],exponent+exp_adjust[0])
1726 + S_mulexp10(result[1],exponent-exp_adjust[1]);
1727 } else {
1728 result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]);
1729 }
98994639
HS
1730
1731 /* now apply the sign */
1732 if (negative)
20f6aaab 1733 result[2] = -result[2];
a36244b7 1734#endif /* USE_PERL_ATOF */
20f6aaab 1735 *value = result[2];
73d840c0 1736 return (char *)s;
98994639
HS
1737}
1738
5d34af89 1739/*
3d9d9213 1740=for apidoc isinfnan
5d34af89 1741
796b6530
KW
1742C<Perl_isinfnan()> is utility function that returns true if the NV
1743argument is either an infinity or a C<NaN>, false otherwise. To test
1744in more detail, use C<Perl_isinf()> and C<Perl_isnan()>.
5d34af89 1745
68611e6f
JH
1746This is also the logical inverse of Perl_isfinite().
1747
5d34af89
JH
1748=cut
1749*/
1cd88304
JH
1750bool
1751Perl_isinfnan(NV nv)
1752{
a5dc2484 1753 PERL_UNUSED_ARG(nv);
1cd88304
JH
1754#ifdef Perl_isinf
1755 if (Perl_isinf(nv))
1756 return TRUE;
1757#endif
1758#ifdef Perl_isnan
1759 if (Perl_isnan(nv))
1760 return TRUE;
1761#endif
1762 return FALSE;
1763}
1764
354b74ae
FC
1765/*
1766=for apidoc
1767
796b6530 1768Checks whether the argument would be either an infinity or C<NaN> when used
354b74ae 1769as a number, but is careful not to trigger non-numeric or uninitialized
796b6530 1770warnings. it assumes the caller has done C<SvGETMAGIC(sv)> already.
354b74ae
FC
1771
1772=cut
1773*/
1774
1775bool
1776Perl_isinfnansv(pTHX_ SV *sv)
1777{
1778 PERL_ARGS_ASSERT_ISINFNANSV;
1779 if (!SvOK(sv))
1780 return FALSE;
1781 if (SvNOKp(sv))
1782 return Perl_isinfnan(SvNVX(sv));
1783 if (SvIOKp(sv))
1784 return FALSE;
1785 {
1786 STRLEN len;
1787 const char *s = SvPV_nomg_const(sv, len);
3823048b 1788 return cBOOL(grok_infnan(&s, s+len));
354b74ae
FC
1789 }
1790}
1791
d67dac15 1792#ifndef HAS_MODFL
68611e6f
JH
1793/* C99 has truncl, pre-C99 Solaris had aintl. We can use either with
1794 * copysignl to emulate modfl, which is in some platforms missing or
1795 * broken. */
d67dac15
JH
1796# if defined(HAS_TRUNCL) && defined(HAS_COPYSIGNL)
1797long double
1798Perl_my_modfl(long double x, long double *ip)
1799{
68611e6f
JH
1800 *ip = truncl(x);
1801 return (x == *ip ? copysignl(0.0L, x) : x - *ip);
d67dac15
JH
1802}
1803# elif defined(HAS_AINTL) && defined(HAS_COPYSIGNL)
55954f19
JH
1804long double
1805Perl_my_modfl(long double x, long double *ip)
1806{
68611e6f
JH
1807 *ip = aintl(x);
1808 return (x == *ip ? copysignl(0.0L, x) : x - *ip);
55954f19 1809}
d67dac15 1810# endif
55954f19
JH
1811#endif
1812
7b9b7dff 1813/* Similarly, with ilogbl and scalbnl we can emulate frexpl. */
55954f19
JH
1814#if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL)
1815long double
1816Perl_my_frexpl(long double x, int *e) {
68611e6f
JH
1817 *e = x == 0.0L ? 0 : ilogbl(x) + 1;
1818 return (scalbnl(x, -*e));
55954f19
JH
1819}
1820#endif
66610fdd
RGS
1821
1822/*
ed140128
AD
1823=for apidoc Perl_signbit
1824
1825Return a non-zero integer if the sign bit on an NV is set, and 0 if
19c1206d 1826it is not.
ed140128 1827
796b6530
KW
1828If F<Configure> detects this system has a C<signbit()> that will work with
1829our NVs, then we just use it via the C<#define> in F<perl.h>. Otherwise,
8b7fad81 1830fall back on this implementation. The main use of this function
796b6530 1831is catching C<-0.0>.
ed140128 1832
796b6530
KW
1833C<Configure> notes: This function is called C<'Perl_signbit'> instead of a
1834plain C<'signbit'> because it is easy to imagine a system having a C<signbit()>
ed140128 1835function or macro that doesn't happen to work with our particular choice
796b6530 1836of NVs. We shouldn't just re-C<#define> C<signbit> as C<Perl_signbit> and expect
ed140128 1837the standard system headers to be happy. Also, this is a no-context
796b6530
KW
1838function (no C<pTHX_>) because C<Perl_signbit()> is usually re-C<#defined> in
1839F<perl.h> as a simple macro call to the system's C<signbit()>.
1840Users should just always call C<Perl_signbit()>.
ed140128
AD
1841
1842=cut
1843*/
1844#if !defined(HAS_SIGNBIT)
1845int
1846Perl_signbit(NV x) {
8b7fad81 1847# ifdef Perl_fp_class_nzero
406d5545
JH
1848 return Perl_fp_class_nzero(x);
1849 /* Try finding the high byte, and assume it's highest bit
1850 * is the sign. This assumption is probably wrong somewhere. */
572cd850
JH
1851# elif defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN
1852 return (((unsigned char *)&x)[9] & 0x80);
1853# elif defined(NV_LITTLE_ENDIAN)
1854 /* Note that NVSIZE is sizeof(NV), which would make the below be
1855 * wrong if the end bytes are unused, which happens with the x86
1856 * 80-bit long doubles, which is why take care of that above. */
1857 return (((unsigned char *)&x)[NVSIZE - 1] & 0x80);
1858# elif defined(NV_BIG_ENDIAN)
1859 return (((unsigned char *)&x)[0] & 0x80);
1860# else
406d5545 1861 /* This last resort fallback is wrong for the negative zero. */
3585840c 1862 return (x < 0.0) ? 1 : 0;
572cd850 1863# endif
ed140128
AD
1864}
1865#endif
1866
1867/*
14d04a33 1868 * ex: set ts=8 sts=4 sw=4 et:
37442d52 1869 */