Commit | Line | Data |
---|---|---|
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 |
23 | This file contains all the stuff needed by perl for manipulating numeric |
24 | values, 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 | ||
32 | U32 | |
ddeaf645 | 33 | Perl_cast_ulong(NV f) |
98994639 HS |
34 | { |
35 | if (f < 0.0) | |
36 | return f < I32_MIN ? (U32) I32_MIN : (U32)(I32) f; | |
37 | if (f < U32_MAX_P1) { | |
38 | #if CASTFLAGS & 2 | |
39 | if (f < U32_MAX_P1_HALF) | |
40 | return (U32) f; | |
41 | f -= U32_MAX_P1_HALF; | |
42 | return ((U32) f) | (1 + U32_MAX >> 1); | |
43 | #else | |
44 | return (U32) f; | |
45 | #endif | |
46 | } | |
47 | return f > 0 ? U32_MAX : 0 /* NaN */; | |
48 | } | |
49 | ||
50 | I32 | |
ddeaf645 | 51 | Perl_cast_i32(NV f) |
98994639 HS |
52 | { |
53 | if (f < I32_MAX_P1) | |
54 | return f < I32_MIN ? I32_MIN : (I32) f; | |
55 | if (f < U32_MAX_P1) { | |
56 | #if CASTFLAGS & 2 | |
57 | if (f < U32_MAX_P1_HALF) | |
58 | return (I32)(U32) f; | |
59 | f -= U32_MAX_P1_HALF; | |
60 | return (I32)(((U32) f) | (1 + U32_MAX >> 1)); | |
61 | #else | |
62 | return (I32)(U32) f; | |
63 | #endif | |
64 | } | |
65 | return f > 0 ? (I32)U32_MAX : 0 /* NaN */; | |
66 | } | |
67 | ||
68 | IV | |
ddeaf645 | 69 | Perl_cast_iv(NV f) |
98994639 HS |
70 | { |
71 | if (f < IV_MAX_P1) | |
72 | return f < IV_MIN ? IV_MIN : (IV) f; | |
73 | if (f < UV_MAX_P1) { | |
74 | #if CASTFLAGS & 2 | |
75 | /* For future flexibility allowing for sizeof(UV) >= sizeof(IV) */ | |
76 | if (f < UV_MAX_P1_HALF) | |
77 | return (IV)(UV) f; | |
78 | f -= UV_MAX_P1_HALF; | |
79 | return (IV)(((UV) f) | (1 + UV_MAX >> 1)); | |
80 | #else | |
81 | return (IV)(UV) f; | |
82 | #endif | |
83 | } | |
84 | return f > 0 ? (IV)UV_MAX : 0 /* NaN */; | |
85 | } | |
86 | ||
87 | UV | |
ddeaf645 | 88 | Perl_cast_uv(NV f) |
98994639 HS |
89 | { |
90 | if (f < 0.0) | |
91 | return f < IV_MIN ? (UV) IV_MIN : (UV)(IV) f; | |
92 | if (f < UV_MAX_P1) { | |
93 | #if CASTFLAGS & 2 | |
94 | if (f < UV_MAX_P1_HALF) | |
95 | return (UV) f; | |
96 | f -= UV_MAX_P1_HALF; | |
97 | return ((UV) f) | (1 + UV_MAX >> 1); | |
98 | #else | |
99 | return (UV) f; | |
100 | #endif | |
101 | } | |
102 | return f > 0 ? UV_MAX : 0 /* NaN */; | |
103 | } | |
104 | ||
53305cf1 NC |
105 | /* |
106 | =for apidoc grok_bin | |
98994639 | 107 | |
53305cf1 NC |
108 | converts a string representing a binary number to numeric form. |
109 | ||
110 | On entry I<start> and I<*len> give the string to scan, I<*flags> gives | |
111 | conversion flags, and I<result> should be NULL or a pointer to an NV. | |
112 | The scan stops at the end of the string, or the first invalid character. | |
7b667b5f MHM |
113 | Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*flags>, encountering an |
114 | invalid character will also trigger a warning. | |
115 | On return I<*len> is set to the length of the scanned string, | |
116 | and I<*flags> gives output flags. | |
53305cf1 | 117 | |
7fc63493 | 118 | If the value is <= C<UV_MAX> it is returned as a UV, the output flags are clear, |
72d33970 | 119 | and nothing is written to I<*result>. If the value is > UV_MAX C<grok_bin> |
53305cf1 NC |
120 | returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags, |
121 | and writes the value to I<*result> (or the value is discarded if I<result> | |
122 | is NULL). | |
123 | ||
7b667b5f | 124 | The binary number may optionally be prefixed with "0b" or "b" unless |
72d33970 | 125 | C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If |
a4c04bdc | 126 | C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the binary |
53305cf1 NC |
127 | number may use '_' characters to separate digits. |
128 | ||
129 | =cut | |
02470786 KW |
130 | |
131 | Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE | |
132 | which suppresses any message for non-portable numbers that are still valid | |
133 | on this platform. | |
53305cf1 NC |
134 | */ |
135 | ||
136 | UV | |
7918f24d NC |
137 | Perl_grok_bin(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) |
138 | { | |
53305cf1 NC |
139 | const char *s = start; |
140 | STRLEN len = *len_p; | |
141 | UV value = 0; | |
142 | NV value_nv = 0; | |
143 | ||
144 | const UV max_div_2 = UV_MAX / 2; | |
f2338a2e | 145 | const bool allow_underscores = cBOOL(*flags & PERL_SCAN_ALLOW_UNDERSCORES); |
53305cf1 | 146 | bool overflowed = FALSE; |
7fc63493 | 147 | char bit; |
53305cf1 | 148 | |
7918f24d NC |
149 | PERL_ARGS_ASSERT_GROK_BIN; |
150 | ||
a4c04bdc NC |
151 | if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) { |
152 | /* strip off leading b or 0b. | |
153 | for compatibility silently suffer "b" and "0b" as valid binary | |
154 | numbers. */ | |
155 | if (len >= 1) { | |
a674e8db | 156 | if (s[0] == 'b' || s[0] == 'B') { |
a4c04bdc NC |
157 | s++; |
158 | len--; | |
159 | } | |
a674e8db | 160 | else if (len >= 2 && s[0] == '0' && (s[1] == 'b' || s[1] == 'B')) { |
a4c04bdc NC |
161 | s+=2; |
162 | len-=2; | |
163 | } | |
164 | } | |
53305cf1 NC |
165 | } |
166 | ||
7fc63493 | 167 | for (; len-- && (bit = *s); s++) { |
53305cf1 NC |
168 | if (bit == '0' || bit == '1') { |
169 | /* Write it in this wonky order with a goto to attempt to get the | |
170 | compiler to make the common case integer-only loop pretty tight. | |
171 | With gcc seems to be much straighter code than old scan_bin. */ | |
172 | redo: | |
173 | if (!overflowed) { | |
174 | if (value <= max_div_2) { | |
175 | value = (value << 1) | (bit - '0'); | |
176 | continue; | |
177 | } | |
178 | /* Bah. We're just overflowed. */ | |
dcbac5bb | 179 | /* diag_listed_as: Integer overflow in %s number */ |
9b387841 NC |
180 | Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW), |
181 | "Integer overflow in binary number"); | |
53305cf1 NC |
182 | overflowed = TRUE; |
183 | value_nv = (NV) value; | |
184 | } | |
185 | value_nv *= 2.0; | |
98994639 | 186 | /* If an NV has not enough bits in its mantissa to |
d1be9408 | 187 | * represent a UV this summing of small low-order numbers |
98994639 HS |
188 | * is a waste of time (because the NV cannot preserve |
189 | * the low-order bits anyway): we could just remember when | |
53305cf1 | 190 | * did we overflow and in the end just multiply value_nv by the |
98994639 | 191 | * right amount. */ |
53305cf1 NC |
192 | value_nv += (NV)(bit - '0'); |
193 | continue; | |
194 | } | |
195 | if (bit == '_' && len && allow_underscores && (bit = s[1]) | |
196 | && (bit == '0' || bit == '1')) | |
98994639 HS |
197 | { |
198 | --len; | |
199 | ++s; | |
53305cf1 | 200 | goto redo; |
98994639 | 201 | } |
a2a5de95 NC |
202 | if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT)) |
203 | Perl_ck_warner(aTHX_ packWARN(WARN_DIGIT), | |
204 | "Illegal binary digit '%c' ignored", *s); | |
53305cf1 | 205 | break; |
98994639 | 206 | } |
53305cf1 NC |
207 | |
208 | if ( ( overflowed && value_nv > 4294967295.0) | |
98994639 | 209 | #if UVSIZE > 4 |
02470786 KW |
210 | || (!overflowed && value > 0xffffffff |
211 | && ! (*flags & PERL_SCAN_SILENT_NON_PORTABLE)) | |
98994639 HS |
212 | #endif |
213 | ) { | |
a2a5de95 NC |
214 | Perl_ck_warner(aTHX_ packWARN(WARN_PORTABLE), |
215 | "Binary number > 0b11111111111111111111111111111111 non-portable"); | |
53305cf1 NC |
216 | } |
217 | *len_p = s - start; | |
218 | if (!overflowed) { | |
219 | *flags = 0; | |
220 | return value; | |
98994639 | 221 | } |
53305cf1 NC |
222 | *flags = PERL_SCAN_GREATER_THAN_UV_MAX; |
223 | if (result) | |
224 | *result = value_nv; | |
225 | return UV_MAX; | |
98994639 HS |
226 | } |
227 | ||
53305cf1 NC |
228 | /* |
229 | =for apidoc grok_hex | |
230 | ||
231 | converts a string representing a hex number to numeric form. | |
232 | ||
c2da02fc | 233 | On entry I<start> and I<*len_p> give the string to scan, I<*flags> gives |
53305cf1 | 234 | conversion flags, and I<result> should be NULL or a pointer to an NV. |
7b667b5f MHM |
235 | The scan stops at the end of the string, or the first invalid character. |
236 | Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*flags>, encountering an | |
237 | invalid character will also trigger a warning. | |
238 | On return I<*len> is set to the length of the scanned string, | |
239 | and I<*flags> gives output flags. | |
53305cf1 NC |
240 | |
241 | If the value is <= UV_MAX it is returned as a UV, the output flags are clear, | |
72d33970 | 242 | and nothing is written to I<*result>. If the value is > UV_MAX C<grok_hex> |
53305cf1 NC |
243 | returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags, |
244 | and writes the value to I<*result> (or the value is discarded if I<result> | |
245 | is NULL). | |
246 | ||
d1be9408 | 247 | The hex number may optionally be prefixed with "0x" or "x" unless |
72d33970 | 248 | C<PERL_SCAN_DISALLOW_PREFIX> is set in I<*flags> on entry. If |
a4c04bdc | 249 | C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the hex |
53305cf1 NC |
250 | number may use '_' characters to separate digits. |
251 | ||
252 | =cut | |
02470786 KW |
253 | |
254 | Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE | |
255 | which suppresses any message for non-portable numbers that are still valid | |
256 | on this platform. | |
53305cf1 NC |
257 | */ |
258 | ||
259 | UV | |
7918f24d NC |
260 | Perl_grok_hex(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) |
261 | { | |
53305cf1 NC |
262 | const char *s = start; |
263 | STRLEN len = *len_p; | |
264 | UV value = 0; | |
265 | NV value_nv = 0; | |
53305cf1 | 266 | const UV max_div_16 = UV_MAX / 16; |
f2338a2e | 267 | const bool allow_underscores = cBOOL(*flags & PERL_SCAN_ALLOW_UNDERSCORES); |
53305cf1 | 268 | bool overflowed = FALSE; |
98994639 | 269 | |
7918f24d NC |
270 | PERL_ARGS_ASSERT_GROK_HEX; |
271 | ||
a4c04bdc NC |
272 | if (!(*flags & PERL_SCAN_DISALLOW_PREFIX)) { |
273 | /* strip off leading x or 0x. | |
274 | for compatibility silently suffer "x" and "0x" as valid hex numbers. | |
275 | */ | |
276 | if (len >= 1) { | |
a674e8db | 277 | if (s[0] == 'x' || s[0] == 'X') { |
a4c04bdc NC |
278 | s++; |
279 | len--; | |
280 | } | |
a674e8db | 281 | else if (len >= 2 && s[0] == '0' && (s[1] == 'x' || s[1] == 'X')) { |
a4c04bdc NC |
282 | s+=2; |
283 | len-=2; | |
284 | } | |
285 | } | |
98994639 HS |
286 | } |
287 | ||
288 | for (; len-- && *s; s++) { | |
626ef089 | 289 | if (isXDIGIT(*s)) { |
53305cf1 NC |
290 | /* Write it in this wonky order with a goto to attempt to get the |
291 | compiler to make the common case integer-only loop pretty tight. | |
292 | With gcc seems to be much straighter code than old scan_hex. */ | |
293 | redo: | |
294 | if (!overflowed) { | |
295 | if (value <= max_div_16) { | |
626ef089 | 296 | value = (value << 4) | XDIGIT_VALUE(*s); |
53305cf1 NC |
297 | continue; |
298 | } | |
299 | /* Bah. We're just overflowed. */ | |
dcbac5bb | 300 | /* diag_listed_as: Integer overflow in %s number */ |
9b387841 NC |
301 | Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW), |
302 | "Integer overflow in hexadecimal number"); | |
53305cf1 NC |
303 | overflowed = TRUE; |
304 | value_nv = (NV) value; | |
305 | } | |
306 | value_nv *= 16.0; | |
307 | /* If an NV has not enough bits in its mantissa to | |
d1be9408 | 308 | * represent a UV this summing of small low-order numbers |
53305cf1 NC |
309 | * is a waste of time (because the NV cannot preserve |
310 | * the low-order bits anyway): we could just remember when | |
311 | * did we overflow and in the end just multiply value_nv by the | |
312 | * right amount of 16-tuples. */ | |
626ef089 | 313 | value_nv += (NV) XDIGIT_VALUE(*s); |
53305cf1 NC |
314 | continue; |
315 | } | |
316 | if (*s == '_' && len && allow_underscores && s[1] | |
626ef089 | 317 | && isXDIGIT(s[1])) |
98994639 HS |
318 | { |
319 | --len; | |
320 | ++s; | |
53305cf1 | 321 | goto redo; |
98994639 | 322 | } |
a2a5de95 NC |
323 | if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT)) |
324 | Perl_ck_warner(aTHX_ packWARN(WARN_DIGIT), | |
53305cf1 NC |
325 | "Illegal hexadecimal digit '%c' ignored", *s); |
326 | break; | |
327 | } | |
328 | ||
329 | if ( ( overflowed && value_nv > 4294967295.0) | |
330 | #if UVSIZE > 4 | |
02470786 KW |
331 | || (!overflowed && value > 0xffffffff |
332 | && ! (*flags & PERL_SCAN_SILENT_NON_PORTABLE)) | |
53305cf1 NC |
333 | #endif |
334 | ) { | |
a2a5de95 NC |
335 | Perl_ck_warner(aTHX_ packWARN(WARN_PORTABLE), |
336 | "Hexadecimal number > 0xffffffff non-portable"); | |
53305cf1 NC |
337 | } |
338 | *len_p = s - start; | |
339 | if (!overflowed) { | |
340 | *flags = 0; | |
341 | return value; | |
342 | } | |
343 | *flags = PERL_SCAN_GREATER_THAN_UV_MAX; | |
344 | if (result) | |
345 | *result = value_nv; | |
346 | return UV_MAX; | |
347 | } | |
348 | ||
349 | /* | |
350 | =for apidoc grok_oct | |
351 | ||
7b667b5f MHM |
352 | converts a string representing an octal number to numeric form. |
353 | ||
354 | On entry I<start> and I<*len> give the string to scan, I<*flags> gives | |
355 | conversion flags, and I<result> should be NULL or a pointer to an NV. | |
356 | The scan stops at the end of the string, or the first invalid character. | |
357 | Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in I<*flags>, encountering an | |
154bd527 | 358 | 8 or 9 will also trigger a warning. |
7b667b5f MHM |
359 | On return I<*len> is set to the length of the scanned string, |
360 | and I<*flags> gives output flags. | |
361 | ||
362 | If the value is <= UV_MAX it is returned as a UV, the output flags are clear, | |
72d33970 | 363 | and nothing is written to I<*result>. If the value is > UV_MAX C<grok_oct> |
7b667b5f MHM |
364 | returns UV_MAX, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags, |
365 | and writes the value to I<*result> (or the value is discarded if I<result> | |
366 | is NULL). | |
367 | ||
368 | If C<PERL_SCAN_ALLOW_UNDERSCORES> is set in I<*flags> then the octal | |
369 | number may use '_' characters to separate digits. | |
53305cf1 NC |
370 | |
371 | =cut | |
02470786 | 372 | |
333ae27c KW |
373 | Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE> |
374 | which suppresses any message for non-portable numbers, but which are valid | |
02470786 | 375 | on this platform. |
53305cf1 NC |
376 | */ |
377 | ||
378 | UV | |
7918f24d NC |
379 | Perl_grok_oct(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *result) |
380 | { | |
53305cf1 NC |
381 | const char *s = start; |
382 | STRLEN len = *len_p; | |
383 | UV value = 0; | |
384 | NV value_nv = 0; | |
53305cf1 | 385 | const UV max_div_8 = UV_MAX / 8; |
f2338a2e | 386 | const bool allow_underscores = cBOOL(*flags & PERL_SCAN_ALLOW_UNDERSCORES); |
53305cf1 NC |
387 | bool overflowed = FALSE; |
388 | ||
7918f24d NC |
389 | PERL_ARGS_ASSERT_GROK_OCT; |
390 | ||
53305cf1 | 391 | for (; len-- && *s; s++) { |
626ef089 | 392 | if (isOCTAL(*s)) { |
53305cf1 NC |
393 | /* Write it in this wonky order with a goto to attempt to get the |
394 | compiler to make the common case integer-only loop pretty tight. | |
395 | */ | |
396 | redo: | |
397 | if (!overflowed) { | |
398 | if (value <= max_div_8) { | |
626ef089 | 399 | value = (value << 3) | OCTAL_VALUE(*s); |
53305cf1 NC |
400 | continue; |
401 | } | |
402 | /* Bah. We're just overflowed. */ | |
dcbac5bb | 403 | /* diag_listed_as: Integer overflow in %s number */ |
9b387841 NC |
404 | Perl_ck_warner_d(aTHX_ packWARN(WARN_OVERFLOW), |
405 | "Integer overflow in octal number"); | |
53305cf1 NC |
406 | overflowed = TRUE; |
407 | value_nv = (NV) value; | |
408 | } | |
409 | value_nv *= 8.0; | |
98994639 | 410 | /* If an NV has not enough bits in its mantissa to |
d1be9408 | 411 | * represent a UV this summing of small low-order numbers |
98994639 HS |
412 | * is a waste of time (because the NV cannot preserve |
413 | * the low-order bits anyway): we could just remember when | |
53305cf1 NC |
414 | * did we overflow and in the end just multiply value_nv by the |
415 | * right amount of 8-tuples. */ | |
626ef089 | 416 | value_nv += (NV) OCTAL_VALUE(*s); |
53305cf1 NC |
417 | continue; |
418 | } | |
626ef089 KW |
419 | if (*s == '_' && len && allow_underscores && isOCTAL(s[1])) { |
420 | --len; | |
421 | ++s; | |
422 | goto redo; | |
423 | } | |
53305cf1 | 424 | /* Allow \octal to work the DWIM way (that is, stop scanning |
7b667b5f | 425 | * as soon as non-octal characters are seen, complain only if |
626ef089 KW |
426 | * someone seems to want to use the digits eight and nine. Since we |
427 | * know it is not octal, then if isDIGIT, must be an 8 or 9). */ | |
428 | if (isDIGIT(*s)) { | |
a2a5de95 NC |
429 | if (!(*flags & PERL_SCAN_SILENT_ILLDIGIT)) |
430 | Perl_ck_warner(aTHX_ packWARN(WARN_DIGIT), | |
431 | "Illegal octal digit '%c' ignored", *s); | |
53305cf1 NC |
432 | } |
433 | break; | |
98994639 | 434 | } |
53305cf1 NC |
435 | |
436 | if ( ( overflowed && value_nv > 4294967295.0) | |
98994639 | 437 | #if UVSIZE > 4 |
02470786 KW |
438 | || (!overflowed && value > 0xffffffff |
439 | && ! (*flags & PERL_SCAN_SILENT_NON_PORTABLE)) | |
98994639 HS |
440 | #endif |
441 | ) { | |
a2a5de95 NC |
442 | Perl_ck_warner(aTHX_ packWARN(WARN_PORTABLE), |
443 | "Octal number > 037777777777 non-portable"); | |
53305cf1 NC |
444 | } |
445 | *len_p = s - start; | |
446 | if (!overflowed) { | |
447 | *flags = 0; | |
448 | return value; | |
98994639 | 449 | } |
53305cf1 NC |
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 scan_bin | |
458 | ||
72d33970 | 459 | For backwards compatibility. Use C<grok_bin> instead. |
53305cf1 NC |
460 | |
461 | =for apidoc scan_hex | |
462 | ||
72d33970 | 463 | For backwards compatibility. Use C<grok_hex> instead. |
53305cf1 NC |
464 | |
465 | =for apidoc scan_oct | |
466 | ||
72d33970 | 467 | For backwards compatibility. Use C<grok_oct> instead. |
53305cf1 NC |
468 | |
469 | =cut | |
470 | */ | |
471 | ||
472 | NV | |
73d840c0 | 473 | Perl_scan_bin(pTHX_ const char *start, STRLEN len, STRLEN *retlen) |
53305cf1 NC |
474 | { |
475 | NV rnv; | |
476 | I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0; | |
73d840c0 | 477 | const UV ruv = grok_bin (start, &len, &flags, &rnv); |
53305cf1 | 478 | |
7918f24d NC |
479 | PERL_ARGS_ASSERT_SCAN_BIN; |
480 | ||
53305cf1 NC |
481 | *retlen = len; |
482 | return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv; | |
483 | } | |
484 | ||
485 | NV | |
73d840c0 | 486 | Perl_scan_oct(pTHX_ const char *start, STRLEN len, STRLEN *retlen) |
53305cf1 NC |
487 | { |
488 | NV rnv; | |
489 | I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0; | |
73d840c0 | 490 | const UV ruv = grok_oct (start, &len, &flags, &rnv); |
53305cf1 | 491 | |
7918f24d NC |
492 | PERL_ARGS_ASSERT_SCAN_OCT; |
493 | ||
53305cf1 NC |
494 | *retlen = len; |
495 | return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv; | |
496 | } | |
497 | ||
498 | NV | |
73d840c0 | 499 | Perl_scan_hex(pTHX_ const char *start, STRLEN len, STRLEN *retlen) |
53305cf1 NC |
500 | { |
501 | NV rnv; | |
502 | I32 flags = *retlen ? PERL_SCAN_ALLOW_UNDERSCORES : 0; | |
73d840c0 | 503 | const UV ruv = grok_hex (start, &len, &flags, &rnv); |
53305cf1 | 504 | |
7918f24d NC |
505 | PERL_ARGS_ASSERT_SCAN_HEX; |
506 | ||
53305cf1 NC |
507 | *retlen = len; |
508 | return (flags & PERL_SCAN_GREATER_THAN_UV_MAX) ? rnv : (NV)ruv; | |
98994639 HS |
509 | } |
510 | ||
511 | /* | |
512 | =for apidoc grok_numeric_radix | |
513 | ||
514 | Scan and skip for a numeric decimal separator (radix). | |
515 | ||
516 | =cut | |
517 | */ | |
518 | bool | |
519 | Perl_grok_numeric_radix(pTHX_ const char **sp, const char *send) | |
520 | { | |
521 | #ifdef USE_LOCALE_NUMERIC | |
7918f24d NC |
522 | PERL_ARGS_ASSERT_GROK_NUMERIC_RADIX; |
523 | ||
d6ded950 | 524 | if (IN_LC(LC_NUMERIC)) { |
21431899 KW |
525 | DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED(); |
526 | if (PL_numeric_radix_sv) { | |
c5a7e38e KW |
527 | STRLEN len; |
528 | const char * const radix = SvPV(PL_numeric_radix_sv, len); | |
529 | if (*sp + len <= send && memEQ(*sp, radix, len)) { | |
530 | *sp += len; | |
531 | RESTORE_LC_NUMERIC(); | |
532 | return TRUE; | |
533 | } | |
21431899 KW |
534 | } |
535 | RESTORE_LC_NUMERIC(); | |
98994639 HS |
536 | } |
537 | /* always try "." if numeric radix didn't match because | |
538 | * we may have data from different locales mixed */ | |
539 | #endif | |
7918f24d NC |
540 | |
541 | PERL_ARGS_ASSERT_GROK_NUMERIC_RADIX; | |
542 | ||
98994639 HS |
543 | if (*sp < send && **sp == '.') { |
544 | ++*sp; | |
545 | return TRUE; | |
546 | } | |
547 | return FALSE; | |
548 | } | |
549 | ||
550 | /* | |
551 | =for apidoc grok_number | |
552 | ||
553 | Recognise (or not) a number. The type of the number is returned | |
554 | (0 if unrecognised), otherwise it is a bit-ORed combination of | |
555 | IS_NUMBER_IN_UV, IS_NUMBER_GREATER_THAN_UV_MAX, IS_NUMBER_NOT_INT, | |
aa8b85de | 556 | IS_NUMBER_NEG, IS_NUMBER_INFINITY, IS_NUMBER_NAN (defined in perl.h). |
60939fb8 | 557 | |
cd164854 | 558 | If the value of the number can fit in a UV, it is returned in the *valuep |
60939fb8 NC |
559 | IS_NUMBER_IN_UV will be set to indicate that *valuep is valid, IS_NUMBER_IN_UV |
560 | will never be set unless *valuep is valid, but *valuep may have been assigned | |
561 | to during processing even though IS_NUMBER_IN_UV is not set on return. | |
562 | If valuep is NULL, IS_NUMBER_IN_UV will be set for the same cases as when | |
563 | valuep is non-NULL, but no actual assignment (or SEGV) will occur. | |
564 | ||
565 | IS_NUMBER_NOT_INT will be set with IS_NUMBER_IN_UV if trailing decimals were | |
566 | seen (in which case *valuep gives the true value truncated to an integer), and | |
567 | IS_NUMBER_NEG if the number is negative (in which case *valuep holds the | |
568 | absolute value). IS_NUMBER_IN_UV is not set if e notation was used or the | |
569 | number is larger than a UV. | |
98994639 HS |
570 | |
571 | =cut | |
572 | */ | |
573 | int | |
574 | Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep) | |
575 | { | |
60939fb8 | 576 | const char *s = pv; |
c4420975 | 577 | const char * const send = pv + len; |
60939fb8 NC |
578 | const UV max_div_10 = UV_MAX / 10; |
579 | const char max_mod_10 = UV_MAX % 10; | |
580 | int numtype = 0; | |
581 | int sawinf = 0; | |
aa8b85de | 582 | int sawnan = 0; |
60939fb8 | 583 | |
7918f24d NC |
584 | PERL_ARGS_ASSERT_GROK_NUMBER; |
585 | ||
60939fb8 NC |
586 | while (s < send && isSPACE(*s)) |
587 | s++; | |
588 | if (s == send) { | |
589 | return 0; | |
590 | } else if (*s == '-') { | |
591 | s++; | |
592 | numtype = IS_NUMBER_NEG; | |
593 | } | |
594 | else if (*s == '+') | |
aa42a541 | 595 | s++; |
60939fb8 NC |
596 | |
597 | if (s == send) | |
598 | return 0; | |
599 | ||
600 | /* next must be digit or the radix separator or beginning of infinity */ | |
601 | if (isDIGIT(*s)) { | |
602 | /* UVs are at least 32 bits, so the first 9 decimal digits cannot | |
603 | overflow. */ | |
604 | UV value = *s - '0'; | |
605 | /* This construction seems to be more optimiser friendly. | |
606 | (without it gcc does the isDIGIT test and the *s - '0' separately) | |
607 | With it gcc on arm is managing 6 instructions (6 cycles) per digit. | |
608 | In theory the optimiser could deduce how far to unroll the loop | |
609 | before checking for overflow. */ | |
58bb9ec3 NC |
610 | if (++s < send) { |
611 | int digit = *s - '0'; | |
60939fb8 NC |
612 | if (digit >= 0 && digit <= 9) { |
613 | value = value * 10 + digit; | |
58bb9ec3 NC |
614 | if (++s < send) { |
615 | digit = *s - '0'; | |
60939fb8 NC |
616 | if (digit >= 0 && digit <= 9) { |
617 | value = value * 10 + digit; | |
58bb9ec3 NC |
618 | if (++s < send) { |
619 | digit = *s - '0'; | |
60939fb8 NC |
620 | if (digit >= 0 && digit <= 9) { |
621 | value = value * 10 + digit; | |
58bb9ec3 NC |
622 | if (++s < send) { |
623 | digit = *s - '0'; | |
60939fb8 NC |
624 | if (digit >= 0 && digit <= 9) { |
625 | value = value * 10 + digit; | |
58bb9ec3 NC |
626 | if (++s < send) { |
627 | digit = *s - '0'; | |
60939fb8 NC |
628 | if (digit >= 0 && digit <= 9) { |
629 | value = value * 10 + digit; | |
58bb9ec3 NC |
630 | if (++s < send) { |
631 | digit = *s - '0'; | |
60939fb8 NC |
632 | if (digit >= 0 && digit <= 9) { |
633 | value = value * 10 + digit; | |
58bb9ec3 NC |
634 | if (++s < send) { |
635 | digit = *s - '0'; | |
60939fb8 NC |
636 | if (digit >= 0 && digit <= 9) { |
637 | value = value * 10 + digit; | |
58bb9ec3 NC |
638 | if (++s < send) { |
639 | digit = *s - '0'; | |
60939fb8 NC |
640 | if (digit >= 0 && digit <= 9) { |
641 | value = value * 10 + digit; | |
58bb9ec3 | 642 | if (++s < send) { |
60939fb8 NC |
643 | /* Now got 9 digits, so need to check |
644 | each time for overflow. */ | |
58bb9ec3 | 645 | digit = *s - '0'; |
60939fb8 NC |
646 | while (digit >= 0 && digit <= 9 |
647 | && (value < max_div_10 | |
648 | || (value == max_div_10 | |
649 | && digit <= max_mod_10))) { | |
650 | value = value * 10 + digit; | |
58bb9ec3 NC |
651 | if (++s < send) |
652 | digit = *s - '0'; | |
60939fb8 NC |
653 | else |
654 | break; | |
655 | } | |
656 | if (digit >= 0 && digit <= 9 | |
51bd16da | 657 | && (s < send)) { |
60939fb8 NC |
658 | /* value overflowed. |
659 | skip the remaining digits, don't | |
660 | worry about setting *valuep. */ | |
661 | do { | |
662 | s++; | |
663 | } while (s < send && isDIGIT(*s)); | |
664 | numtype |= | |
665 | IS_NUMBER_GREATER_THAN_UV_MAX; | |
666 | goto skip_value; | |
667 | } | |
668 | } | |
669 | } | |
98994639 | 670 | } |
60939fb8 NC |
671 | } |
672 | } | |
673 | } | |
674 | } | |
675 | } | |
676 | } | |
677 | } | |
678 | } | |
679 | } | |
680 | } | |
681 | } | |
98994639 | 682 | } |
60939fb8 | 683 | } |
98994639 | 684 | } |
60939fb8 NC |
685 | numtype |= IS_NUMBER_IN_UV; |
686 | if (valuep) | |
687 | *valuep = value; | |
688 | ||
689 | skip_value: | |
690 | if (GROK_NUMERIC_RADIX(&s, send)) { | |
691 | numtype |= IS_NUMBER_NOT_INT; | |
692 | while (s < send && isDIGIT(*s)) /* optional digits after the radix */ | |
693 | s++; | |
98994639 | 694 | } |
60939fb8 NC |
695 | } |
696 | else if (GROK_NUMERIC_RADIX(&s, send)) { | |
697 | numtype |= IS_NUMBER_NOT_INT | IS_NUMBER_IN_UV; /* valuep assigned below */ | |
698 | /* no digits before the radix means we need digits after it */ | |
699 | if (s < send && isDIGIT(*s)) { | |
700 | do { | |
701 | s++; | |
702 | } while (s < send && isDIGIT(*s)); | |
703 | if (valuep) { | |
704 | /* integer approximation is valid - it's 0. */ | |
705 | *valuep = 0; | |
706 | } | |
98994639 | 707 | } |
60939fb8 NC |
708 | else |
709 | return 0; | |
710 | } else if (*s == 'I' || *s == 'i') { | |
711 | s++; if (s == send || (*s != 'N' && *s != 'n')) return 0; | |
712 | s++; if (s == send || (*s != 'F' && *s != 'f')) return 0; | |
713 | s++; if (s < send && (*s == 'I' || *s == 'i')) { | |
714 | s++; if (s == send || (*s != 'N' && *s != 'n')) return 0; | |
715 | s++; if (s == send || (*s != 'I' && *s != 'i')) return 0; | |
716 | s++; if (s == send || (*s != 'T' && *s != 't')) return 0; | |
717 | s++; if (s == send || (*s != 'Y' && *s != 'y')) return 0; | |
718 | s++; | |
98994639 | 719 | } |
60939fb8 | 720 | sawinf = 1; |
aa8b85de JH |
721 | } else if (*s == 'N' || *s == 'n') { |
722 | /* XXX TODO: There are signaling NaNs and quiet NaNs. */ | |
723 | s++; if (s == send || (*s != 'A' && *s != 'a')) return 0; | |
724 | s++; if (s == send || (*s != 'N' && *s != 'n')) return 0; | |
725 | s++; | |
726 | sawnan = 1; | |
727 | } else | |
98994639 | 728 | return 0; |
60939fb8 NC |
729 | |
730 | if (sawinf) { | |
731 | numtype &= IS_NUMBER_NEG; /* Keep track of sign */ | |
732 | numtype |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT; | |
aa8b85de JH |
733 | } else if (sawnan) { |
734 | numtype &= IS_NUMBER_NEG; /* Keep track of sign */ | |
735 | numtype |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT; | |
60939fb8 NC |
736 | } else if (s < send) { |
737 | /* we can have an optional exponent part */ | |
738 | if (*s == 'e' || *s == 'E') { | |
739 | /* The only flag we keep is sign. Blow away any "it's UV" */ | |
740 | numtype &= IS_NUMBER_NEG; | |
741 | numtype |= IS_NUMBER_NOT_INT; | |
742 | s++; | |
743 | if (s < send && (*s == '-' || *s == '+')) | |
744 | s++; | |
745 | if (s < send && isDIGIT(*s)) { | |
746 | do { | |
747 | s++; | |
748 | } while (s < send && isDIGIT(*s)); | |
749 | } | |
750 | else | |
751 | return 0; | |
752 | } | |
753 | } | |
754 | while (s < send && isSPACE(*s)) | |
755 | s++; | |
756 | if (s >= send) | |
aa8b85de | 757 | return numtype; |
60939fb8 NC |
758 | if (len == 10 && memEQ(pv, "0 but true", 10)) { |
759 | if (valuep) | |
760 | *valuep = 0; | |
761 | return IS_NUMBER_IN_UV; | |
762 | } | |
763 | return 0; | |
98994639 HS |
764 | } |
765 | ||
4801ca72 | 766 | STATIC NV |
98994639 HS |
767 | S_mulexp10(NV value, I32 exponent) |
768 | { | |
769 | NV result = 1.0; | |
770 | NV power = 10.0; | |
771 | bool negative = 0; | |
772 | I32 bit; | |
773 | ||
774 | if (exponent == 0) | |
775 | return value; | |
659c4b96 DM |
776 | if (value == 0) |
777 | return (NV)0; | |
87032ba1 | 778 | |
24866caa | 779 | /* On OpenVMS VAX we by default use the D_FLOAT double format, |
67597c89 | 780 | * and that format does not have *easy* capabilities [1] for |
24866caa CB |
781 | * overflowing doubles 'silently' as IEEE fp does. We also need |
782 | * to support G_FLOAT on both VAX and Alpha, and though the exponent | |
783 | * range is much larger than D_FLOAT it still doesn't do silent | |
784 | * overflow. Therefore we need to detect early whether we would | |
785 | * overflow (this is the behaviour of the native string-to-float | |
786 | * conversion routines, and therefore of native applications, too). | |
67597c89 | 787 | * |
24866caa CB |
788 | * [1] Trying to establish a condition handler to trap floating point |
789 | * exceptions is not a good idea. */ | |
87032ba1 JH |
790 | |
791 | /* In UNICOS and in certain Cray models (such as T90) there is no | |
792 | * IEEE fp, and no way at all from C to catch fp overflows gracefully. | |
793 | * There is something you can do if you are willing to use some | |
794 | * inline assembler: the instruction is called DFI-- but that will | |
795 | * disable *all* floating point interrupts, a little bit too large | |
796 | * a hammer. Therefore we need to catch potential overflows before | |
797 | * it's too late. */ | |
353813d9 | 798 | |
85bba25f | 799 | #if ((defined(VMS) && !defined(_IEEE_FP)) || defined(_UNICOS)) && defined(NV_MAX_10_EXP) |
353813d9 | 800 | STMT_START { |
c4420975 | 801 | const NV exp_v = log10(value); |
353813d9 HS |
802 | if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP) |
803 | return NV_MAX; | |
804 | if (exponent < 0) { | |
805 | if (-(exponent + exp_v) >= NV_MAX_10_EXP) | |
806 | return 0.0; | |
807 | while (-exponent >= NV_MAX_10_EXP) { | |
808 | /* combination does not overflow, but 10^(-exponent) does */ | |
809 | value /= 10; | |
810 | ++exponent; | |
811 | } | |
812 | } | |
813 | } STMT_END; | |
87032ba1 JH |
814 | #endif |
815 | ||
353813d9 HS |
816 | if (exponent < 0) { |
817 | negative = 1; | |
818 | exponent = -exponent; | |
b27804d8 DM |
819 | #ifdef NV_MAX_10_EXP |
820 | /* for something like 1234 x 10^-309, the action of calculating | |
821 | * the intermediate value 10^309 then returning 1234 / (10^309) | |
822 | * will fail, since 10^309 becomes infinity. In this case try to | |
823 | * refactor it as 123 / (10^308) etc. | |
824 | */ | |
825 | while (value && exponent > NV_MAX_10_EXP) { | |
826 | exponent--; | |
827 | value /= 10; | |
828 | } | |
829 | #endif | |
353813d9 | 830 | } |
98994639 HS |
831 | for (bit = 1; exponent; bit <<= 1) { |
832 | if (exponent & bit) { | |
833 | exponent ^= bit; | |
834 | result *= power; | |
236f0012 CB |
835 | /* Floating point exceptions are supposed to be turned off, |
836 | * but if we're obviously done, don't risk another iteration. | |
837 | */ | |
838 | if (exponent == 0) break; | |
98994639 HS |
839 | } |
840 | power *= power; | |
841 | } | |
842 | return negative ? value / result : value * result; | |
843 | } | |
844 | ||
845 | NV | |
846 | Perl_my_atof(pTHX_ const char* s) | |
847 | { | |
848 | NV x = 0.0; | |
849 | #ifdef USE_LOCALE_NUMERIC | |
7918f24d NC |
850 | PERL_ARGS_ASSERT_MY_ATOF; |
851 | ||
a2287a13 KW |
852 | { |
853 | DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED(); | |
d6ded950 | 854 | if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) { |
e4850248 KW |
855 | const char *standard = NULL, *local = NULL; |
856 | bool use_standard_radix; | |
98994639 | 857 | |
e4850248 KW |
858 | /* Look through the string for the first thing that looks like a |
859 | * decimal point: either the value in the current locale or the | |
860 | * standard fallback of '.'. The one which appears earliest in the | |
861 | * input string is the one that we should have atof look for. Note | |
862 | * that we have to determine this beforehand because on some | |
863 | * systems, Perl_atof2 is just a wrapper around the system's atof. | |
864 | * */ | |
865 | standard = strchr(s, '.'); | |
866 | local = strstr(s, SvPV_nolen(PL_numeric_radix_sv)); | |
78787052 | 867 | |
e4850248 | 868 | use_standard_radix = standard && (!local || standard < local); |
78787052 | 869 | |
e4850248 KW |
870 | if (use_standard_radix) |
871 | SET_NUMERIC_STANDARD(); | |
78787052 | 872 | |
e4850248 | 873 | Perl_atof2(s, x); |
78787052 | 874 | |
e4850248 KW |
875 | if (use_standard_radix) |
876 | SET_NUMERIC_LOCAL(); | |
877 | } | |
878 | else | |
879 | Perl_atof2(s, x); | |
a2287a13 KW |
880 | RESTORE_LC_NUMERIC(); |
881 | } | |
98994639 | 882 | #else |
a36244b7 | 883 | Perl_atof2(s, x); |
98994639 HS |
884 | #endif |
885 | return x; | |
886 | } | |
887 | ||
888 | char* | |
889 | Perl_my_atof2(pTHX_ const char* orig, NV* value) | |
890 | { | |
20f6aaab | 891 | NV result[3] = {0.0, 0.0, 0.0}; |
e1ec3a88 | 892 | const char* s = orig; |
a36244b7 | 893 | #ifdef USE_PERL_ATOF |
20f6aaab | 894 | UV accumulator[2] = {0,0}; /* before/after dp */ |
a36244b7 | 895 | bool negative = 0; |
e1ec3a88 | 896 | const char* send = s + strlen(orig) - 1; |
8194bf88 | 897 | bool seen_digit = 0; |
20f6aaab AS |
898 | I32 exp_adjust[2] = {0,0}; |
899 | I32 exp_acc[2] = {-1, -1}; | |
900 | /* the current exponent adjust for the accumulators */ | |
98994639 | 901 | I32 exponent = 0; |
8194bf88 | 902 | I32 seen_dp = 0; |
20f6aaab AS |
903 | I32 digit = 0; |
904 | I32 old_digit = 0; | |
8194bf88 DM |
905 | I32 sig_digits = 0; /* noof significant digits seen so far */ |
906 | ||
7918f24d NC |
907 | PERL_ARGS_ASSERT_MY_ATOF2; |
908 | ||
8194bf88 DM |
909 | /* There is no point in processing more significant digits |
910 | * than the NV can hold. Note that NV_DIG is a lower-bound value, | |
911 | * while we need an upper-bound value. We add 2 to account for this; | |
912 | * since it will have been conservative on both the first and last digit. | |
913 | * For example a 32-bit mantissa with an exponent of 4 would have | |
914 | * exact values in the set | |
915 | * 4 | |
916 | * 8 | |
917 | * .. | |
918 | * 17179869172 | |
919 | * 17179869176 | |
920 | * 17179869180 | |
921 | * | |
922 | * where for the purposes of calculating NV_DIG we would have to discount | |
923 | * both the first and last digit, since neither can hold all values from | |
924 | * 0..9; but for calculating the value we must examine those two digits. | |
925 | */ | |
ffa277e5 AS |
926 | #ifdef MAX_SIG_DIG_PLUS |
927 | /* It is not necessarily the case that adding 2 to NV_DIG gets all the | |
928 | possible digits in a NV, especially if NVs are not IEEE compliant | |
929 | (e.g., long doubles on IRIX) - Allen <allens@cpan.org> */ | |
930 | # define MAX_SIG_DIGITS (NV_DIG+MAX_SIG_DIG_PLUS) | |
931 | #else | |
932 | # define MAX_SIG_DIGITS (NV_DIG+2) | |
933 | #endif | |
8194bf88 DM |
934 | |
935 | /* the max number we can accumulate in a UV, and still safely do 10*N+9 */ | |
936 | #define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10)) | |
98994639 | 937 | |
96a05aee HS |
938 | /* leading whitespace */ |
939 | while (isSPACE(*s)) | |
940 | ++s; | |
941 | ||
98994639 HS |
942 | /* sign */ |
943 | switch (*s) { | |
944 | case '-': | |
945 | negative = 1; | |
924ba076 | 946 | /* FALLTHROUGH */ |
98994639 HS |
947 | case '+': |
948 | ++s; | |
949 | } | |
950 | ||
2b54f59f YST |
951 | /* punt to strtod for NaN/Inf; if no support for it there, tough luck */ |
952 | ||
953 | #ifdef HAS_STRTOD | |
954 | if (*s == 'n' || *s == 'N' || *s == 'i' || *s == 'I') { | |
c042ae3a | 955 | const char *p = negative ? s - 1 : s; |
2b54f59f YST |
956 | char *endp; |
957 | NV rslt; | |
958 | rslt = strtod(p, &endp); | |
959 | if (endp != p) { | |
960 | *value = rslt; | |
961 | return (char *)endp; | |
962 | } | |
963 | } | |
964 | #endif | |
965 | ||
8194bf88 DM |
966 | /* we accumulate digits into an integer; when this becomes too |
967 | * large, we add the total to NV and start again */ | |
98994639 | 968 | |
8194bf88 DM |
969 | while (1) { |
970 | if (isDIGIT(*s)) { | |
971 | seen_digit = 1; | |
20f6aaab | 972 | old_digit = digit; |
8194bf88 | 973 | digit = *s++ - '0'; |
20f6aaab AS |
974 | if (seen_dp) |
975 | exp_adjust[1]++; | |
98994639 | 976 | |
8194bf88 DM |
977 | /* don't start counting until we see the first significant |
978 | * digit, eg the 5 in 0.00005... */ | |
979 | if (!sig_digits && digit == 0) | |
980 | continue; | |
981 | ||
982 | if (++sig_digits > MAX_SIG_DIGITS) { | |
98994639 | 983 | /* limits of precision reached */ |
20f6aaab AS |
984 | if (digit > 5) { |
985 | ++accumulator[seen_dp]; | |
986 | } else if (digit == 5) { | |
987 | if (old_digit % 2) { /* round to even - Allen */ | |
988 | ++accumulator[seen_dp]; | |
989 | } | |
990 | } | |
991 | if (seen_dp) { | |
992 | exp_adjust[1]--; | |
993 | } else { | |
994 | exp_adjust[0]++; | |
995 | } | |
8194bf88 | 996 | /* skip remaining digits */ |
98994639 | 997 | while (isDIGIT(*s)) { |
98994639 | 998 | ++s; |
20f6aaab AS |
999 | if (! seen_dp) { |
1000 | exp_adjust[0]++; | |
1001 | } | |
98994639 HS |
1002 | } |
1003 | /* warn of loss of precision? */ | |
98994639 | 1004 | } |
8194bf88 | 1005 | else { |
20f6aaab | 1006 | if (accumulator[seen_dp] > MAX_ACCUMULATE) { |
8194bf88 | 1007 | /* add accumulator to result and start again */ |
20f6aaab AS |
1008 | result[seen_dp] = S_mulexp10(result[seen_dp], |
1009 | exp_acc[seen_dp]) | |
1010 | + (NV)accumulator[seen_dp]; | |
1011 | accumulator[seen_dp] = 0; | |
1012 | exp_acc[seen_dp] = 0; | |
98994639 | 1013 | } |
20f6aaab AS |
1014 | accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit; |
1015 | ++exp_acc[seen_dp]; | |
98994639 | 1016 | } |
8194bf88 | 1017 | } |
e1ec3a88 | 1018 | else if (!seen_dp && GROK_NUMERIC_RADIX(&s, send)) { |
8194bf88 | 1019 | seen_dp = 1; |
20f6aaab | 1020 | if (sig_digits > MAX_SIG_DIGITS) { |
c86f7df5 | 1021 | do { |
20f6aaab | 1022 | ++s; |
c86f7df5 | 1023 | } while (isDIGIT(*s)); |
20f6aaab AS |
1024 | break; |
1025 | } | |
8194bf88 DM |
1026 | } |
1027 | else { | |
1028 | break; | |
98994639 HS |
1029 | } |
1030 | } | |
1031 | ||
20f6aaab AS |
1032 | result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0]; |
1033 | if (seen_dp) { | |
1034 | result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1]; | |
1035 | } | |
98994639 | 1036 | |
8194bf88 | 1037 | if (seen_digit && (*s == 'e' || *s == 'E')) { |
98994639 HS |
1038 | bool expnegative = 0; |
1039 | ||
1040 | ++s; | |
1041 | switch (*s) { | |
1042 | case '-': | |
1043 | expnegative = 1; | |
924ba076 | 1044 | /* FALLTHROUGH */ |
98994639 HS |
1045 | case '+': |
1046 | ++s; | |
1047 | } | |
1048 | while (isDIGIT(*s)) | |
1049 | exponent = exponent * 10 + (*s++ - '0'); | |
1050 | if (expnegative) | |
1051 | exponent = -exponent; | |
1052 | } | |
1053 | ||
20f6aaab AS |
1054 | |
1055 | ||
98994639 | 1056 | /* now apply the exponent */ |
20f6aaab AS |
1057 | |
1058 | if (seen_dp) { | |
1059 | result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]) | |
1060 | + S_mulexp10(result[1],exponent-exp_adjust[1]); | |
1061 | } else { | |
1062 | result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]); | |
1063 | } | |
98994639 HS |
1064 | |
1065 | /* now apply the sign */ | |
1066 | if (negative) | |
20f6aaab | 1067 | result[2] = -result[2]; |
a36244b7 | 1068 | #endif /* USE_PERL_ATOF */ |
20f6aaab | 1069 | *value = result[2]; |
73d840c0 | 1070 | return (char *)s; |
98994639 HS |
1071 | } |
1072 | ||
55954f19 JH |
1073 | #if ! defined(HAS_MODFL) && defined(HAS_AINTL) && defined(HAS_COPYSIGNL) |
1074 | long double | |
1075 | Perl_my_modfl(long double x, long double *ip) | |
1076 | { | |
1077 | *ip = aintl(x); | |
1078 | return (x == *ip ? copysignl(0.0L, x) : x - *ip); | |
1079 | } | |
1080 | #endif | |
1081 | ||
1082 | #if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL) | |
1083 | long double | |
1084 | Perl_my_frexpl(long double x, int *e) { | |
1085 | *e = x == 0.0L ? 0 : ilogbl(x) + 1; | |
1086 | return (scalbnl(x, -*e)); | |
1087 | } | |
1088 | #endif | |
66610fdd RGS |
1089 | |
1090 | /* | |
ed140128 AD |
1091 | =for apidoc Perl_signbit |
1092 | ||
1093 | Return a non-zero integer if the sign bit on an NV is set, and 0 if | |
1094 | it is not. | |
1095 | ||
1096 | If Configure detects this system has a signbit() that will work with | |
1097 | our NVs, then we just use it via the #define in perl.h. Otherwise, | |
1098 | fall back on this implementation. As a first pass, this gets everything | |
1099 | right except -0.0. Alas, catching -0.0 is the main use for this function, | |
1100 | so this is not too helpful yet. Still, at least we have the scaffolding | |
1101 | in place to support other systems, should that prove useful. | |
1102 | ||
1103 | ||
1104 | Configure notes: This function is called 'Perl_signbit' instead of a | |
1105 | plain 'signbit' because it is easy to imagine a system having a signbit() | |
1106 | function or macro that doesn't happen to work with our particular choice | |
1107 | of NVs. We shouldn't just re-#define signbit as Perl_signbit and expect | |
1108 | the standard system headers to be happy. Also, this is a no-context | |
1109 | function (no pTHX_) because Perl_signbit() is usually re-#defined in | |
1110 | perl.h as a simple macro call to the system's signbit(). | |
1111 | Users should just always call Perl_signbit(). | |
1112 | ||
1113 | =cut | |
1114 | */ | |
1115 | #if !defined(HAS_SIGNBIT) | |
1116 | int | |
1117 | Perl_signbit(NV x) { | |
1118 | return (x < 0.0) ? 1 : 0; | |
1119 | } | |
1120 | #endif | |
1121 | ||
1122 | /* | |
66610fdd RGS |
1123 | * Local variables: |
1124 | * c-indentation-style: bsd | |
1125 | * c-basic-offset: 4 | |
14d04a33 | 1126 | * indent-tabs-mode: nil |
66610fdd RGS |
1127 | * End: |
1128 | * | |
14d04a33 | 1129 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 1130 | */ |