xref: /onnv-gate/usr/src/cmd/perl/5.8.4/distrib/numeric.c (revision 0:68f95e015346)
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 */
19 
20 #include "EXTERN.h"
21 #define PERL_IN_NUMERIC_C
22 #include "perl.h"
23 
24 U32
Perl_cast_ulong(pTHX_ NV f)25 Perl_cast_ulong(pTHX_ NV f)
26 {
27   if (f < 0.0)
28     return f < I32_MIN ? (U32) I32_MIN : (U32)(I32) f;
29   if (f < U32_MAX_P1) {
30 #if CASTFLAGS & 2
31     if (f < U32_MAX_P1_HALF)
32       return (U32) f;
33     f -= U32_MAX_P1_HALF;
34     return ((U32) f) | (1 + U32_MAX >> 1);
35 #else
36     return (U32) f;
37 #endif
38   }
39   return f > 0 ? U32_MAX : 0 /* NaN */;
40 }
41 
42 I32
Perl_cast_i32(pTHX_ NV f)43 Perl_cast_i32(pTHX_ NV f)
44 {
45   if (f < I32_MAX_P1)
46     return f < I32_MIN ? I32_MIN : (I32) f;
47   if (f < U32_MAX_P1) {
48 #if CASTFLAGS & 2
49     if (f < U32_MAX_P1_HALF)
50       return (I32)(U32) f;
51     f -= U32_MAX_P1_HALF;
52     return (I32)(((U32) f) | (1 + U32_MAX >> 1));
53 #else
54     return (I32)(U32) f;
55 #endif
56   }
57   return f > 0 ? (I32)U32_MAX : 0 /* NaN */;
58 }
59 
60 IV
Perl_cast_iv(pTHX_ NV f)61 Perl_cast_iv(pTHX_ NV f)
62 {
63   if (f < IV_MAX_P1)
64     return f < IV_MIN ? IV_MIN : (IV) f;
65   if (f < UV_MAX_P1) {
66 #if CASTFLAGS & 2
67     /* For future flexibility allowing for sizeof(UV) >= sizeof(IV)  */
68     if (f < UV_MAX_P1_HALF)
69       return (IV)(UV) f;
70     f -= UV_MAX_P1_HALF;
71     return (IV)(((UV) f) | (1 + UV_MAX >> 1));
72 #else
73     return (IV)(UV) f;
74 #endif
75   }
76   return f > 0 ? (IV)UV_MAX : 0 /* NaN */;
77 }
78 
79 UV
Perl_cast_uv(pTHX_ NV f)80 Perl_cast_uv(pTHX_ NV f)
81 {
82   if (f < 0.0)
83     return f < IV_MIN ? (UV) IV_MIN : (UV)(IV) f;
84   if (f < UV_MAX_P1) {
85 #if CASTFLAGS & 2
86     if (f < UV_MAX_P1_HALF)
87       return (UV) f;
88     f -= UV_MAX_P1_HALF;
89     return ((UV) f) | (1 + UV_MAX >> 1);
90 #else
91     return (UV) f;
92 #endif
93   }
94   return f > 0 ? UV_MAX : 0 /* NaN */;
95 }
96 
97 #if defined(HUGE_VAL) || (defined(USE_LONG_DOUBLE) && defined(HUGE_VALL))
98 /*
99  * This hack is to force load of "huge" support from libm.a
100  * So it is in perl for (say) POSIX to use.
101  * Needed for SunOS with Sun's 'acc' for example.
102  */
103 NV
Perl_huge(void)104 Perl_huge(void)
105 {
106 #   if defined(USE_LONG_DOUBLE) && defined(HUGE_VALL)
107     return HUGE_VALL;
108 #   endif
109     return HUGE_VAL;
110 }
111 #endif
112 
113 /*
114 =for apidoc grok_bin
115 
116 converts a string representing a binary number to numeric form.
117 
118 On entry I<start> and I<*len> give the string to scan, I<*flags> gives
119 conversion flags, and I<result> should be NULL or a pointer to an NV.
120 The scan stops at the end of the string, or the first invalid character.
121 On return I<*len> is set to the length scanned string, and I<*flags> gives
122 output flags.
123 
124 If the value is <= UV_MAX it is returned as a UV, the output flags are clear,
125 and nothing is written to I<*result>. If the value is > UV_MAX C<grok_bin>
126 returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
127 and writes the value to I<*result> (or the value is discarded if I<result>
128 is NULL).
129 
130 The hex number may optionally be prefixed with "0b" or "b" unless
131 C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If
132 C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the binary
133 number may use '_' characters to separate digits.
134 
135 =cut
136  */
137 
138 UV
Perl_grok_bin(pTHX_ char * start,STRLEN * len_p,I32 * flags,NV * result)139 Perl_grok_bin(pTHX_ char *start, STRLEN *len_p, I32 *flags, NV *result) {
140     const char *s = start;
141     STRLEN len = *len_p;
142     UV value = 0;
143     NV value_nv = 0;
144 
145     const UV max_div_2 = UV_MAX / 2;
146     bool allow_underscores = *flags & PERL_SCAN_ALLOW_UNDERSCORES;
147     bool overflowed = FALSE;
148 
149     if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
150         /* strip off leading b or 0b.
151            for compatibility silently suffer "b" and "0b" as valid binary
152            numbers. */
153         if (len >= 1) {
154             if (s[0] == 'b') {
155                 s++;
156                 len--;
157             }
158             else if (len >= 2 && s[0] == '0' && s[1] == 'b') {
159                 s+=2;
160                 len-=2;
161             }
162         }
163     }
164 
165     for (; len-- && *s; s++) {
166         char bit = *s;
167         if (bit == '0' || bit == '1') {
168             /* Write it in this wonky order with a goto to attempt to get the
169                compiler to make the common case integer-only loop pretty tight.
170                With gcc seems to be much straighter code than old scan_bin.  */
171           redo:
172             if (!overflowed) {
173                 if (value <= max_div_2) {
174                     value = (value << 1) | (bit - '0');
175                     continue;
176                 }
177                 /* Bah. We're just overflowed.  */
178                 if (ckWARN_d(WARN_OVERFLOW))
179                     Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
180                                 "Integer overflow in binary number");
181                 overflowed = TRUE;
182                 value_nv = (NV) value;
183             }
184             value_nv *= 2.0;
185 	    /* If an NV has not enough bits in its mantissa to
186 	     * represent a UV this summing of small low-order numbers
187 	     * is a waste of time (because the NV cannot preserve
188 	     * the low-order bits anyway): we could just remember when
189 	     * did we overflow and in the end just multiply value_nv by the
190 	     * right amount. */
191             value_nv += (NV)(bit - '0');
192             continue;
193         }
194         if (bit == '_' && len && allow_underscores && (bit = s[1])
195             && (bit == '0' || bit == '1'))
196 	    {
197 		--len;
198 		++s;
199                 goto redo;
200 	    }
201         if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
202             Perl_warner(aTHX_ packWARN(WARN_DIGIT),
203                         "Illegal binary digit '%c' ignored", *s);
204         break;
205     }
206 
207     if (   ( overflowed && value_nv > 4294967295.0)
208 #if UVSIZE > 4
209 	|| (!overflowed && value > 0xffffffff  )
210 #endif
211 	) {
212 	if (ckWARN(WARN_PORTABLE))
213 	    Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
214 			"Binary number > 0b11111111111111111111111111111111 non-portable");
215     }
216     *len_p = s - start;
217     if (!overflowed) {
218         *flags = 0;
219         return value;
220     }
221     *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
222     if (result)
223         *result = value_nv;
224     return UV_MAX;
225 }
226 
227 /*
228 =for apidoc grok_hex
229 
230 converts a string representing a hex number to numeric form.
231 
232 On entry I<start> and I<*len> give the string to scan, I<*flags> gives
233 conversion flags, and I<result> should be NULL or a pointer to an NV.
234 The scan stops at the end of the string, or the first non-hex-digit character.
235 On return I<*len> is set to the length scanned string, and I<*flags> gives
236 output flags.
237 
238 If the value is <= UV_MAX it is returned as a UV, the output flags are clear,
239 and nothing is written to I<*result>. If the value is > UV_MAX C<grok_hex>
240 returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags,
241 and writes the value to I<*result> (or the value is discarded if I<result>
242 is NULL).
243 
244 The hex number may optionally be prefixed with "0x" or "x" unless
245 C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If
246 C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the hex
247 number may use '_' characters to separate digits.
248 
249 =cut
250  */
251 
252 UV
Perl_grok_hex(pTHX_ char * start,STRLEN * len_p,I32 * flags,NV * result)253 Perl_grok_hex(pTHX_ char *start, STRLEN *len_p, I32 *flags, NV *result) {
254     const char *s = start;
255     STRLEN len = *len_p;
256     UV value = 0;
257     NV value_nv = 0;
258 
259     const UV max_div_16 = UV_MAX / 16;
260     bool allow_underscores = *flags & PERL_SCAN_ALLOW_UNDERSCORES;
261     bool overflowed = FALSE;
262     const char *hexdigit;
263 
264     if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) {
265         /* strip off leading x or 0x.
266            for compatibility silently suffer "x" and "0x" as valid hex numbers.
267         */
268         if (len >= 1) {
269             if (s[0] == 'x') {
270                 s++;
271                 len--;
272             }
273             else if (len >= 2 && s[0] == '0' && s[1] == 'x') {
274                 s+=2;
275                 len-=2;
276             }
277         }
278     }
279 
280     for (; len-- && *s; s++) {
281 	hexdigit = strchr((char *) PL_hexdigit, *s);
282         if (hexdigit) {
283             /* Write it in this wonky order with a goto to attempt to get the
284                compiler to make the common case integer-only loop pretty tight.
285                With gcc seems to be much straighter code than old scan_hex.  */
286           redo:
287             if (!overflowed) {
288                 if (value <= max_div_16) {
289                     value = (value << 4) | ((hexdigit - PL_hexdigit) & 15);
290                     continue;
291                 }
292                 /* Bah. We're just overflowed.  */
293                 if (ckWARN_d(WARN_OVERFLOW))
294                     Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
295                                 "Integer overflow in hexadecimal number");
296                 overflowed = TRUE;
297                 value_nv = (NV) value;
298             }
299             value_nv *= 16.0;
300 	    /* If an NV has not enough bits in its mantissa to
301 	     * represent a UV this summing of small low-order numbers
302 	     * is a waste of time (because the NV cannot preserve
303 	     * the low-order bits anyway): we could just remember when
304 	     * did we overflow and in the end just multiply value_nv by the
305 	     * right amount of 16-tuples. */
306             value_nv += (NV)((hexdigit - PL_hexdigit) & 15);
307             continue;
308         }
309         if (*s == '_' && len && allow_underscores && s[1]
310 		&& (hexdigit = strchr((char *) PL_hexdigit, s[1])))
311 	    {
312 		--len;
313 		++s;
314                 goto redo;
315 	    }
316         if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
317             Perl_warner(aTHX_ packWARN(WARN_DIGIT),
318                         "Illegal hexadecimal digit '%c' ignored", *s);
319         break;
320     }
321 
322     if (   ( overflowed && value_nv > 4294967295.0)
323 #if UVSIZE > 4
324 	|| (!overflowed && value > 0xffffffff  )
325 #endif
326 	) {
327 	if (ckWARN(WARN_PORTABLE))
328 	    Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
329 			"Hexadecimal number > 0xffffffff non-portable");
330     }
331     *len_p = s - start;
332     if (!overflowed) {
333         *flags = 0;
334         return value;
335     }
336     *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
337     if (result)
338         *result = value_nv;
339     return UV_MAX;
340 }
341 
342 /*
343 =for apidoc grok_oct
344 
345 
346 =cut
347  */
348 
349 UV
Perl_grok_oct(pTHX_ char * start,STRLEN * len_p,I32 * flags,NV * result)350 Perl_grok_oct(pTHX_ char *start, STRLEN *len_p, I32 *flags, NV *result) {
351     const char *s = start;
352     STRLEN len = *len_p;
353     UV value = 0;
354     NV value_nv = 0;
355 
356     const UV max_div_8 = UV_MAX / 8;
357     bool allow_underscores = *flags & PERL_SCAN_ALLOW_UNDERSCORES;
358     bool overflowed = FALSE;
359 
360     for (; len-- && *s; s++) {
361          /* gcc 2.95 optimiser not smart enough to figure that this subtraction
362             out front allows slicker code.  */
363         int digit = *s - '0';
364         if (digit >= 0 && digit <= 7) {
365             /* Write it in this wonky order with a goto to attempt to get the
366                compiler to make the common case integer-only loop pretty tight.
367             */
368           redo:
369             if (!overflowed) {
370                 if (value <= max_div_8) {
371                     value = (value << 3) | digit;
372                     continue;
373                 }
374                 /* Bah. We're just overflowed.  */
375                 if (ckWARN_d(WARN_OVERFLOW))
376                     Perl_warner(aTHX_ packWARN(WARN_OVERFLOW),
377                                 "Integer overflow in octal number");
378                 overflowed = TRUE;
379                 value_nv = (NV) value;
380             }
381             value_nv *= 8.0;
382 	    /* If an NV has not enough bits in its mantissa to
383 	     * represent a UV this summing of small low-order numbers
384 	     * is a waste of time (because the NV cannot preserve
385 	     * the low-order bits anyway): we could just remember when
386 	     * did we overflow and in the end just multiply value_nv by the
387 	     * right amount of 8-tuples. */
388             value_nv += (NV)digit;
389             continue;
390         }
391         if (digit == ('_' - '0') && len && allow_underscores
392             && (digit = s[1] - '0') && (digit >= 0 && digit <= 7))
393 	    {
394 		--len;
395 		++s;
396                 goto redo;
397 	    }
398         /* Allow \octal to work the DWIM way (that is, stop scanning
399          * as soon as non-octal characters are seen, complain only iff
400          * someone seems to want to use the digits eight and nine). */
401         if (digit == 8 || digit == 9) {
402             if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT) && ckWARN(WARN_DIGIT))
403                 Perl_warner(aTHX_ packWARN(WARN_DIGIT),
404                             "Illegal octal digit '%c' ignored", *s);
405         }
406         break;
407     }
408 
409     if (   ( overflowed && value_nv > 4294967295.0)
410 #if UVSIZE > 4
411 	|| (!overflowed && value > 0xffffffff  )
412 #endif
413 	) {
414 	if (ckWARN(WARN_PORTABLE))
415 	    Perl_warner(aTHX_ packWARN(WARN_PORTABLE),
416 			"Octal number > 037777777777 non-portable");
417     }
418     *len_p = s - start;
419     if (!overflowed) {
420         *flags = 0;
421         return value;
422     }
423     *flags = PERL_SCAN_GREATER_THAN_UV_MAX;
424     if (result)
425         *result = value_nv;
426     return UV_MAX;
427 }
428 
429 /*
430 =for apidoc scan_bin
431 
432 For backwards compatibility. Use C<grok_bin> instead.
433 
434 =for apidoc scan_hex
435 
436 For backwards compatibility. Use C<grok_hex> instead.
437 
438 =for apidoc scan_oct
439 
440 For backwards compatibility. Use C<grok_oct> instead.
441 
442 =cut
443  */
444 
445 NV
Perl_scan_bin(pTHX_ char * start,STRLEN len,STRLEN * retlen)446 Perl_scan_bin(pTHX_ char *start, STRLEN len, STRLEN *retlen)
447 {
448     NV rnv;
449     I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
450     UV ruv = grok_bin (start, &len, &flags, &rnv);
451 
452     *retlen = len;
453     return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
454 }
455 
456 NV
Perl_scan_oct(pTHX_ char * start,STRLEN len,STRLEN * retlen)457 Perl_scan_oct(pTHX_ char *start, STRLEN len, STRLEN *retlen)
458 {
459     NV rnv;
460     I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
461     UV ruv = grok_oct (start, &len, &flags, &rnv);
462 
463     *retlen = len;
464     return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
465 }
466 
467 NV
Perl_scan_hex(pTHX_ char * start,STRLEN len,STRLEN * retlen)468 Perl_scan_hex(pTHX_ char *start, STRLEN len, STRLEN *retlen)
469 {
470     NV rnv;
471     I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0;
472     UV ruv = grok_hex (start, &len, &flags, &rnv);
473 
474     *retlen = len;
475     return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv;
476 }
477 
478 /*
479 =for apidoc grok_numeric_radix
480 
481 Scan and skip for a numeric decimal separator (radix).
482 
483 =cut
484  */
485 bool
Perl_grok_numeric_radix(pTHX_ const char ** sp,const char * send)486 Perl_grok_numeric_radix(pTHX_ const char **sp, const char *send)
487 {
488 #ifdef USE_LOCALE_NUMERIC
489     if (PL_numeric_radix_sv && IN_LOCALE) {
490         STRLEN len;
491         char* radix = SvPV(PL_numeric_radix_sv, len);
492         if (*sp + len <= send && memEQ(*sp, radix, len)) {
493             *sp += len;
494             return TRUE;
495         }
496     }
497     /* always try "." if numeric radix didn't match because
498      * we may have data from different locales mixed */
499 #endif
500     if (*sp < send && **sp == '.') {
501         ++*sp;
502         return TRUE;
503     }
504     return FALSE;
505 }
506 
507 /*
508 =for apidoc grok_number
509 
510 Recognise (or not) a number.  The type of the number is returned
511 (0 if unrecognised), otherwise it is a bit-ORed combination of
512 IS_NUMBER_IN_UV, IS_NUMBER_GREATER_THAN_UV_MAX, IS_NUMBER_NOT_INT,
513 IS_NUMBER_NEG, IS_NUMBER_INFINITY, IS_NUMBER_NAN (defined in perl.h).
514 
515 If the value of the number can fit an in UV, it is returned in the *valuep
516 IS_NUMBER_IN_UV will be set to indicate that *valuep is valid, IS_NUMBER_IN_UV
517 will never be set unless *valuep is valid, but *valuep may have been assigned
518 to during processing even though IS_NUMBER_IN_UV is not set on return.
519 If valuep is NULL, IS_NUMBER_IN_UV will be set for the same cases as when
520 valuep is non-NULL, but no actual assignment (or SEGV) will occur.
521 
522 IS_NUMBER_NOT_INT will be set with IS_NUMBER_IN_UV if trailing decimals were
523 seen (in which case *valuep gives the true value truncated to an integer), and
524 IS_NUMBER_NEG if the number is negative (in which case *valuep holds the
525 absolute value).  IS_NUMBER_IN_UV is not set if e notation was used or the
526 number is larger than a UV.
527 
528 =cut
529  */
530 int
Perl_grok_number(pTHX_ const char * pv,STRLEN len,UV * valuep)531 Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep)
532 {
533   const char *s = pv;
534   const char *send = pv + len;
535   const UV max_div_10 = UV_MAX / 10;
536   const char max_mod_10 = UV_MAX % 10;
537   int numtype = 0;
538   int sawinf = 0;
539   int sawnan = 0;
540 
541   while (s < send && isSPACE(*s))
542     s++;
543   if (s == send) {
544     return 0;
545   } else if (*s == '-') {
546     s++;
547     numtype = IS_NUMBER_NEG;
548   }
549   else if (*s == '+')
550   s++;
551 
552   if (s == send)
553     return 0;
554 
555   /* next must be digit or the radix separator or beginning of infinity */
556   if (isDIGIT(*s)) {
557     /* UVs are at least 32 bits, so the first 9 decimal digits cannot
558        overflow.  */
559     UV value = *s - '0';
560     /* This construction seems to be more optimiser friendly.
561        (without it gcc does the isDIGIT test and the *s - '0' separately)
562        With it gcc on arm is managing 6 instructions (6 cycles) per digit.
563        In theory the optimiser could deduce how far to unroll the loop
564        before checking for overflow.  */
565     if (++s < send) {
566       int digit = *s - '0';
567       if (digit >= 0 && digit <= 9) {
568         value = value * 10 + digit;
569         if (++s < send) {
570           digit = *s - '0';
571           if (digit >= 0 && digit <= 9) {
572             value = value * 10 + digit;
573             if (++s < send) {
574               digit = *s - '0';
575               if (digit >= 0 && digit <= 9) {
576                 value = value * 10 + digit;
577 		if (++s < send) {
578                   digit = *s - '0';
579                   if (digit >= 0 && digit <= 9) {
580                     value = value * 10 + digit;
581                     if (++s < send) {
582                       digit = *s - '0';
583                       if (digit >= 0 && digit <= 9) {
584                         value = value * 10 + digit;
585                         if (++s < send) {
586                           digit = *s - '0';
587                           if (digit >= 0 && digit <= 9) {
588                             value = value * 10 + digit;
589                             if (++s < send) {
590                               digit = *s - '0';
591                               if (digit >= 0 && digit <= 9) {
592                                 value = value * 10 + digit;
593                                 if (++s < send) {
594                                   digit = *s - '0';
595                                   if (digit >= 0 && digit <= 9) {
596                                     value = value * 10 + digit;
597                                     if (++s < send) {
598                                       /* Now got 9 digits, so need to check
599                                          each time for overflow.  */
600                                       digit = *s - '0';
601                                       while (digit >= 0 && digit <= 9
602                                              && (value < max_div_10
603                                                  || (value == max_div_10
604                                                      && digit <= max_mod_10))) {
605                                         value = value * 10 + digit;
606                                         if (++s < send)
607                                           digit = *s - '0';
608                                         else
609                                           break;
610                                       }
611                                       if (digit >= 0 && digit <= 9
612                                           && (s < send)) {
613                                         /* value overflowed.
614                                            skip the remaining digits, don't
615                                            worry about setting *valuep.  */
616                                         do {
617                                           s++;
618                                         } while (s < send && isDIGIT(*s));
619                                         numtype |=
620                                           IS_NUMBER_GREATER_THAN_UV_MAX;
621                                         goto skip_value;
622                                       }
623                                     }
624                                   }
625 				}
626                               }
627                             }
628                           }
629                         }
630                       }
631                     }
632                   }
633                 }
634               }
635             }
636           }
637 	}
638       }
639     }
640     numtype |= IS_NUMBER_IN_UV;
641     if (valuep)
642       *valuep = value;
643 
644   skip_value:
645     if (GROK_NUMERIC_RADIX(&s, send)) {
646       numtype |= IS_NUMBER_NOT_INT;
647       while (s < send && isDIGIT(*s))  /* optional digits after the radix */
648         s++;
649     }
650   }
651   else if (GROK_NUMERIC_RADIX(&s, send)) {
652     numtype |= IS_NUMBER_NOT_INT | IS_NUMBER_IN_UV; /* valuep assigned below */
653     /* no digits before the radix means we need digits after it */
654     if (s < send && isDIGIT(*s)) {
655       do {
656         s++;
657       } while (s < send && isDIGIT(*s));
658       if (valuep) {
659         /* integer approximation is valid - it's 0.  */
660         *valuep = 0;
661       }
662     }
663     else
664       return 0;
665   } else if (*s == 'I' || *s == 'i') {
666     s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
667     s++; if (s == send || (*s != 'F' && *s != 'f')) return 0;
668     s++; if (s < send && (*s == 'I' || *s == 'i')) {
669       s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
670       s++; if (s == send || (*s != 'I' && *s != 'i')) return 0;
671       s++; if (s == send || (*s != 'T' && *s != 't')) return 0;
672       s++; if (s == send || (*s != 'Y' && *s != 'y')) return 0;
673       s++;
674     }
675     sawinf = 1;
676   } else if (*s == 'N' || *s == 'n') {
677     /* XXX TODO: There are signaling NaNs and quiet NaNs. */
678     s++; if (s == send || (*s != 'A' && *s != 'a')) return 0;
679     s++; if (s == send || (*s != 'N' && *s != 'n')) return 0;
680     s++;
681     sawnan = 1;
682   } else
683     return 0;
684 
685   if (sawinf) {
686     numtype &= IS_NUMBER_NEG; /* Keep track of sign  */
687     numtype |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT;
688   } else if (sawnan) {
689     numtype &= IS_NUMBER_NEG; /* Keep track of sign  */
690     numtype |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT;
691   } else if (s < send) {
692     /* we can have an optional exponent part */
693     if (*s == 'e' || *s == 'E') {
694       /* The only flag we keep is sign.  Blow away any "it's UV"  */
695       numtype &= IS_NUMBER_NEG;
696       numtype |= IS_NUMBER_NOT_INT;
697       s++;
698       if (s < send && (*s == '-' || *s == '+'))
699         s++;
700       if (s < send && isDIGIT(*s)) {
701         do {
702           s++;
703         } while (s < send && isDIGIT(*s));
704       }
705       else
706       return 0;
707     }
708   }
709   while (s < send && isSPACE(*s))
710     s++;
711   if (s >= send)
712     return numtype;
713   if (len == 10 && memEQ(pv, "0 but true", 10)) {
714     if (valuep)
715       *valuep = 0;
716     return IS_NUMBER_IN_UV;
717   }
718   return 0;
719 }
720 
721 STATIC NV
S_mulexp10(NV value,I32 exponent)722 S_mulexp10(NV value, I32 exponent)
723 {
724     NV result = 1.0;
725     NV power = 10.0;
726     bool negative = 0;
727     I32 bit;
728 
729     if (exponent == 0)
730 	return value;
731     if (value == 0)
732 	return 0;
733 
734     /* On OpenVMS VAX we by default use the D_FLOAT double format,
735      * and that format does not have *easy* capabilities [1] for
736      * overflowing doubles 'silently' as IEEE fp does.  We also need
737      * to support G_FLOAT on both VAX and Alpha, and though the exponent
738      * range is much larger than D_FLOAT it still doesn't do silent
739      * overflow.  Therefore we need to detect early whether we would
740      * overflow (this is the behaviour of the native string-to-float
741      * conversion routines, and therefore of native applications, too).
742      *
743      * [1] Trying to establish a condition handler to trap floating point
744      *     exceptions is not a good idea. */
745 
746     /* In UNICOS and in certain Cray models (such as T90) there is no
747      * IEEE fp, and no way at all from C to catch fp overflows gracefully.
748      * There is something you can do if you are willing to use some
749      * inline assembler: the instruction is called DFI-- but that will
750      * disable *all* floating point interrupts, a little bit too large
751      * a hammer.  Therefore we need to catch potential overflows before
752      * it's too late. */
753 
754 #if ((defined(VMS) && !defined(__IEEE_FP)) || defined(_UNICOS)) && defined(NV_MAX_10_EXP)
755     STMT_START {
756 	NV exp_v = log10(value);
757 	if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP)
758 	    return NV_MAX;
759 	if (exponent < 0) {
760 	    if (-(exponent + exp_v) >= NV_MAX_10_EXP)
761 		return 0.0;
762 	    while (-exponent >= NV_MAX_10_EXP) {
763 		/* combination does not overflow, but 10^(-exponent) does */
764 		value /= 10;
765 		++exponent;
766 	    }
767 	}
768     } STMT_END;
769 #endif
770 
771     if (exponent < 0) {
772 	negative = 1;
773 	exponent = -exponent;
774     }
775     for (bit = 1; exponent; bit <<= 1) {
776 	if (exponent & bit) {
777 	    exponent ^= bit;
778 	    result *= power;
779 	    /* Floating point exceptions are supposed to be turned off,
780 	     *  but if we're obviously done, don't risk another iteration.
781 	     */
782 	     if (exponent == 0) break;
783 	}
784 	power *= power;
785     }
786     return negative ? value / result : value * result;
787 }
788 
789 NV
Perl_my_atof(pTHX_ const char * s)790 Perl_my_atof(pTHX_ const char* s)
791 {
792     NV x = 0.0;
793 #ifdef USE_LOCALE_NUMERIC
794     if (PL_numeric_local && IN_LOCALE) {
795 	NV y;
796 
797 	/* Scan the number twice; once using locale and once without;
798 	 * choose the larger result (in absolute value). */
799 	Perl_atof2(s, x);
800 	SET_NUMERIC_STANDARD();
801 	Perl_atof2(s, y);
802 	SET_NUMERIC_LOCAL();
803 	if ((y < 0.0 && y < x) || (y > 0.0 && y > x))
804 	    return y;
805     }
806     else
807 	Perl_atof2(s, x);
808 #else
809     Perl_atof2(s, x);
810 #endif
811     return x;
812 }
813 
814 char*
Perl_my_atof2(pTHX_ const char * orig,NV * value)815 Perl_my_atof2(pTHX_ const char* orig, NV* value)
816 {
817     NV result[3] = {0.0, 0.0, 0.0};
818     char* s = (char*)orig;
819 #ifdef USE_PERL_ATOF
820     UV accumulator[2] = {0,0};	/* before/after dp */
821     bool negative = 0;
822     char* send = s + strlen(orig) - 1;
823     bool seen_digit = 0;
824     I32 exp_adjust[2] = {0,0};
825     I32 exp_acc[2] = {-1, -1};
826     /* the current exponent adjust for the accumulators */
827     I32 exponent = 0;
828     I32	seen_dp  = 0;
829     I32 digit = 0;
830     I32 old_digit = 0;
831     I32 sig_digits = 0; /* noof significant digits seen so far */
832 
833 /* There is no point in processing more significant digits
834  * than the NV can hold. Note that NV_DIG is a lower-bound value,
835  * while we need an upper-bound value. We add 2 to account for this;
836  * since it will have been conservative on both the first and last digit.
837  * For example a 32-bit mantissa with an exponent of 4 would have
838  * exact values in the set
839  *               4
840  *               8
841  *              ..
842  *     17179869172
843  *     17179869176
844  *     17179869180
845  *
846  * where for the purposes of calculating NV_DIG we would have to discount
847  * both the first and last digit, since neither can hold all values from
848  * 0..9; but for calculating the value we must examine those two digits.
849  */
850 #define MAX_SIG_DIGITS (NV_DIG+2)
851 
852 /* the max number we can accumulate in a UV, and still safely do 10*N+9 */
853 #define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10))
854 
855     /* leading whitespace */
856     while (isSPACE(*s))
857 	++s;
858 
859     /* sign */
860     switch (*s) {
861 	case '-':
862 	    negative = 1;
863 	    /* fall through */
864 	case '+':
865 	    ++s;
866     }
867 
868     /* we accumulate digits into an integer; when this becomes too
869      * large, we add the total to NV and start again */
870 
871     while (1) {
872 	if (isDIGIT(*s)) {
873 	    seen_digit = 1;
874 	    old_digit = digit;
875 	    digit = *s++ - '0';
876 	    if (seen_dp)
877 		exp_adjust[1]++;
878 
879 	    /* don't start counting until we see the first significant
880 	     * digit, eg the 5 in 0.00005... */
881 	    if (!sig_digits && digit == 0)
882 		continue;
883 
884 	    if (++sig_digits > MAX_SIG_DIGITS) {
885 		/* limits of precision reached */
886 	        if (digit > 5) {
887 		    ++accumulator[seen_dp];
888 		} else if (digit == 5) {
889 		    if (old_digit % 2) { /* round to even - Allen */
890 			++accumulator[seen_dp];
891 		    }
892 		}
893 		if (seen_dp) {
894 		    exp_adjust[1]--;
895 		} else {
896 		    exp_adjust[0]++;
897 		}
898 		/* skip remaining digits */
899 		while (isDIGIT(*s)) {
900 		    ++s;
901 		    if (! seen_dp) {
902 			exp_adjust[0]++;
903 		    }
904 		}
905 		/* warn of loss of precision? */
906 	    }
907 	    else {
908 		if (accumulator[seen_dp] > MAX_ACCUMULATE) {
909 		    /* add accumulator to result and start again */
910 		    result[seen_dp] = S_mulexp10(result[seen_dp],
911 						 exp_acc[seen_dp])
912 			+ (NV)accumulator[seen_dp];
913 		    accumulator[seen_dp] = 0;
914 		    exp_acc[seen_dp] = 0;
915 		}
916 		accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit;
917 		++exp_acc[seen_dp];
918 	    }
919 	}
920 	else if (!seen_dp && GROK_NUMERIC_RADIX((const char **)&s, send)) {
921 	    seen_dp = 1;
922 	    if (sig_digits > MAX_SIG_DIGITS) {
923 		++s;
924 		while (isDIGIT(*s)) {
925 		    ++s;
926 		}
927 		break;
928 	    }
929 	}
930 	else {
931 	    break;
932 	}
933     }
934 
935     result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0];
936     if (seen_dp) {
937 	result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1];
938     }
939 
940     if (seen_digit && (*s == 'e' || *s == 'E')) {
941 	bool expnegative = 0;
942 
943 	++s;
944 	switch (*s) {
945 	    case '-':
946 		expnegative = 1;
947 		/* fall through */
948 	    case '+':
949 		++s;
950 	}
951 	while (isDIGIT(*s))
952 	    exponent = exponent * 10 + (*s++ - '0');
953 	if (expnegative)
954 	    exponent = -exponent;
955     }
956 
957 
958 
959     /* now apply the exponent */
960 
961     if (seen_dp) {
962 	result[2] = S_mulexp10(result[0],exponent+exp_adjust[0])
963 		+ S_mulexp10(result[1],exponent-exp_adjust[1]);
964     } else {
965 	result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]);
966     }
967 
968     /* now apply the sign */
969     if (negative)
970 	result[2] = -result[2];
971 #endif /* USE_PERL_ATOF */
972     *value = result[2];
973     return s;
974 }
975 
976 #if ! defined(HAS_MODFL) && defined(HAS_AINTL) && defined(HAS_COPYSIGNL)
977 long double
Perl_my_modfl(long double x,long double * ip)978 Perl_my_modfl(long double x, long double *ip)
979 {
980 	*ip = aintl(x);
981 	return (x == *ip ? copysignl(0.0L, x) : x - *ip);
982 }
983 #endif
984 
985 #if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL)
986 long double
Perl_my_frexpl(long double x,int * e)987 Perl_my_frexpl(long double x, int *e) {
988 	*e = x == 0.0L ? 0 : ilogbl(x) + 1;
989 	return (scalbnl(x, -*e));
990 }
991 #endif
992