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) { | |
305b8651 | 156 | if (isALPHA_FOLD_EQ(s[0], 'b')) { |
a4c04bdc NC |
157 | s++; |
158 | len--; | |
159 | } | |
305b8651 | 160 | else if (len >= 2 && s[0] == '0' && (isALPHA_FOLD_EQ(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) { | |
305b8651 | 277 | if (isALPHA_FOLD_EQ(s[0], 'x')) { |
a4c04bdc NC |
278 | s++; |
279 | len--; | |
280 | } | |
305b8651 | 281 | else if (len >= 2 && s[0] == '0' && (isALPHA_FOLD_EQ(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 | /* | |
3f7602fa | 551 | =for apidoc grok_number_flags |
98994639 HS |
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 | 570 | |
3f7602fa TC |
571 | C<flags> allows only C<PERL_SCAN_TRAILING>, which allows for trailing |
572 | non-numeric text on an otherwise successful I<grok>, setting | |
573 | C<IS_NUMBER_TRAILING> on the result. | |
574 | ||
575 | =for apidoc grok_number | |
576 | ||
577 | Identical to grok_number_flags() with flags set to zero. | |
578 | ||
98994639 HS |
579 | =cut |
580 | */ | |
581 | int | |
582 | Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep) | |
583 | { | |
3f7602fa TC |
584 | PERL_ARGS_ASSERT_GROK_NUMBER; |
585 | ||
586 | return grok_number_flags(pv, len, valuep, 0); | |
587 | } | |
588 | ||
ae776a2c | 589 | /* Peek ahead to see whether this could be Inf/NaN/qNaN/snan/1.#INF */ |
9ff909cf JH |
590 | #define INFNAN_PEEK(s, send) \ |
591 | (s < send && \ | |
592 | ((isALPHA_FOLD_EQ(*s, 'I') || isALPHA_FOLD_EQ(*s, 'N')) || \ | |
593 | ((s + 4) < send && \ | |
594 | (isALPHA_FOLD_EQ(*s, 'Q') || isALPHA_FOLD_EQ(*s, 'S')) && \ | |
595 | isALPHA_FOLD_EQ(s[1], 'N')) || \ | |
596 | ((s + 5) < send && \ | |
597 | (*s == '1' && ((s[1] == '.' && s[2] == '#') || s[1] == '#'))))) | |
ae776a2c | 598 | |
ff4eb398 JH |
599 | /* |
600 | =for apidoc grok_infnan | |
601 | ||
602 | Helper for grok_number(), accepts various ways of spelling "infinity" | |
603 | or "not a number", and returns one of the following flag combinations: | |
604 | ||
605 | IS_NUMBER_INFINITE | |
606 | IS_NUMBER_NAN | |
607 | IS_NUMBER_INFINITE | IS_NUMBER_NEG | |
608 | IS_NUMBER_NAN | IS_NUMBER_NEG | |
609 | 0 | |
610 | ||
611 | If an infinity or not-a-number is recognized, the *sp will point to | |
612 | one past the end of the recognized string. If the recognition fails, | |
613 | zero is returned, and the *sp will not move. | |
614 | ||
615 | =cut | |
616 | */ | |
617 | ||
618 | int | |
619 | Perl_grok_infnan(const char** sp, const char* send) | |
620 | { | |
621 | const char* s = *sp; | |
622 | int flags = 0; | |
623 | ||
624 | PERL_ARGS_ASSERT_GROK_INFNAN; | |
625 | ||
626 | if (*s == '-') { | |
627 | flags |= IS_NUMBER_NEG; /* Yes, -NaN happens. Incorrect but happens. */ | |
628 | s++; if (s == send) return 0; | |
629 | } | |
630 | ||
631 | if (*s == '1') { | |
632 | /* Visual C: 1.#SNAN, -1.#QNAN, 1#INF, 1#.IND (maybe also 1.#NAN) */ | |
633 | s++; if (s == send) return 0; | |
634 | if (*s == '.') { | |
635 | s++; if (s == send) return 0; | |
636 | } | |
637 | if (*s == '#') { | |
638 | s++; if (s == send) return 0; | |
639 | } else | |
640 | return 0; | |
641 | } | |
642 | ||
305b8651 | 643 | if (isALPHA_FOLD_EQ(*s, 'I')) { |
ff4eb398 | 644 | /* INF or IND (1.#IND is indeterminate, a certain type of NAN) */ |
305b8651 | 645 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0; |
ff4eb398 | 646 | s++; if (s == send) return 0; |
305b8651 | 647 | if (isALPHA_FOLD_EQ(*s, 'F')) { |
ff4eb398 | 648 | s++; |
305b8651 KW |
649 | if (s < send && (isALPHA_FOLD_EQ(*s, 'I'))) { |
650 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0; | |
651 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'I')) return 0; | |
652 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'T')) return 0; | |
ff4eb398 | 653 | /* XXX maybe also grok "infinite"? */ |
305b8651 | 654 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'Y')) return 0; |
ff4eb398 JH |
655 | s++; |
656 | } else if (*s) | |
657 | return 0; | |
658 | flags |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT; | |
659 | } | |
305b8651 | 660 | else if (isALPHA_FOLD_EQ(*s, 'D')) { |
ff4eb398 JH |
661 | s++; |
662 | flags |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT; | |
663 | } else | |
664 | return 0; | |
665 | ||
666 | *sp = s; | |
667 | return flags; | |
668 | } | |
669 | else { | |
670 | /* NAN */ | |
305b8651 | 671 | if (isALPHA_FOLD_EQ(*s, 'S') || isALPHA_FOLD_EQ(*s, 'Q')) { |
ff4eb398 JH |
672 | /* snan, qNaN */ |
673 | /* XXX do something with the snan/qnan difference */ | |
ae776a2c | 674 | s++; if (s == send) return 0; |
ff4eb398 JH |
675 | } |
676 | ||
305b8651 KW |
677 | if (isALPHA_FOLD_EQ(*s, 'N')) { |
678 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'A')) return 0; | |
679 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0; | |
ff4eb398 JH |
680 | s++; |
681 | ||
ae776a2c JH |
682 | flags |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT; |
683 | ||
ff4eb398 JH |
684 | /* NaN can be followed by various stuff since there are |
685 | * multiple different NaN values, and some implementations | |
ae776a2c JH |
686 | * output the "payload" values, e.g. NaN123, NAN(abc), |
687 | * some implementation just have weird stuff like NaN%. */ | |
688 | s = send; | |
ff4eb398 | 689 | } |
ae776a2c JH |
690 | else |
691 | return 0; | |
ff4eb398 JH |
692 | |
693 | *sp = s; | |
694 | return flags; | |
695 | } | |
696 | ||
697 | return 0; | |
698 | } | |
699 | ||
945b524a JH |
700 | static const UV uv_max_div_10 = UV_MAX / 10; |
701 | static const U8 uv_max_mod_10 = UV_MAX % 10; | |
702 | ||
3f7602fa TC |
703 | int |
704 | Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) | |
705 | { | |
60939fb8 | 706 | const char *s = pv; |
c4420975 | 707 | const char * const send = pv + len; |
ae776a2c | 708 | const char *d; |
60939fb8 NC |
709 | int numtype = 0; |
710 | int sawinf = 0; | |
aa8b85de | 711 | int sawnan = 0; |
60939fb8 | 712 | |
3f7602fa | 713 | PERL_ARGS_ASSERT_GROK_NUMBER_FLAGS; |
7918f24d | 714 | |
60939fb8 NC |
715 | while (s < send && isSPACE(*s)) |
716 | s++; | |
717 | if (s == send) { | |
718 | return 0; | |
719 | } else if (*s == '-') { | |
720 | s++; | |
721 | numtype = IS_NUMBER_NEG; | |
722 | } | |
723 | else if (*s == '+') | |
aa42a541 | 724 | s++; |
60939fb8 NC |
725 | |
726 | if (s == send) | |
727 | return 0; | |
728 | ||
ae776a2c JH |
729 | /* The first digit (after optional sign): note that might |
730 | * also point to "infinity" or "nan". */ | |
731 | d = s; | |
732 | ||
60939fb8 NC |
733 | /* next must be digit or the radix separator or beginning of infinity */ |
734 | if (isDIGIT(*s)) { | |
735 | /* UVs are at least 32 bits, so the first 9 decimal digits cannot | |
736 | overflow. */ | |
737 | UV value = *s - '0'; | |
738 | /* This construction seems to be more optimiser friendly. | |
739 | (without it gcc does the isDIGIT test and the *s - '0' separately) | |
740 | With it gcc on arm is managing 6 instructions (6 cycles) per digit. | |
741 | In theory the optimiser could deduce how far to unroll the loop | |
742 | before checking for overflow. */ | |
58bb9ec3 NC |
743 | if (++s < send) { |
744 | int digit = *s - '0'; | |
60939fb8 NC |
745 | if (digit >= 0 && digit <= 9) { |
746 | value = value * 10 + digit; | |
58bb9ec3 NC |
747 | if (++s < send) { |
748 | digit = *s - '0'; | |
60939fb8 NC |
749 | if (digit >= 0 && digit <= 9) { |
750 | value = value * 10 + digit; | |
58bb9ec3 NC |
751 | if (++s < send) { |
752 | digit = *s - '0'; | |
60939fb8 NC |
753 | if (digit >= 0 && digit <= 9) { |
754 | value = value * 10 + digit; | |
58bb9ec3 NC |
755 | if (++s < send) { |
756 | digit = *s - '0'; | |
60939fb8 NC |
757 | if (digit >= 0 && digit <= 9) { |
758 | value = value * 10 + digit; | |
58bb9ec3 NC |
759 | if (++s < send) { |
760 | digit = *s - '0'; | |
60939fb8 NC |
761 | if (digit >= 0 && digit <= 9) { |
762 | value = value * 10 + digit; | |
58bb9ec3 NC |
763 | if (++s < send) { |
764 | digit = *s - '0'; | |
60939fb8 NC |
765 | if (digit >= 0 && digit <= 9) { |
766 | value = value * 10 + digit; | |
58bb9ec3 NC |
767 | if (++s < send) { |
768 | digit = *s - '0'; | |
60939fb8 NC |
769 | if (digit >= 0 && digit <= 9) { |
770 | value = value * 10 + digit; | |
58bb9ec3 NC |
771 | if (++s < send) { |
772 | digit = *s - '0'; | |
60939fb8 NC |
773 | if (digit >= 0 && digit <= 9) { |
774 | value = value * 10 + digit; | |
58bb9ec3 | 775 | if (++s < send) { |
60939fb8 NC |
776 | /* Now got 9 digits, so need to check |
777 | each time for overflow. */ | |
58bb9ec3 | 778 | digit = *s - '0'; |
60939fb8 | 779 | while (digit >= 0 && digit <= 9 |
945b524a JH |
780 | && (value < uv_max_div_10 |
781 | || (value == uv_max_div_10 | |
782 | && digit <= uv_max_mod_10))) { | |
60939fb8 | 783 | value = value * 10 + digit; |
58bb9ec3 NC |
784 | if (++s < send) |
785 | digit = *s - '0'; | |
60939fb8 NC |
786 | else |
787 | break; | |
788 | } | |
789 | if (digit >= 0 && digit <= 9 | |
51bd16da | 790 | && (s < send)) { |
60939fb8 NC |
791 | /* value overflowed. |
792 | skip the remaining digits, don't | |
793 | worry about setting *valuep. */ | |
794 | do { | |
795 | s++; | |
796 | } while (s < send && isDIGIT(*s)); | |
797 | numtype |= | |
798 | IS_NUMBER_GREATER_THAN_UV_MAX; | |
799 | goto skip_value; | |
800 | } | |
801 | } | |
802 | } | |
98994639 | 803 | } |
60939fb8 NC |
804 | } |
805 | } | |
806 | } | |
807 | } | |
808 | } | |
809 | } | |
810 | } | |
811 | } | |
812 | } | |
813 | } | |
814 | } | |
98994639 | 815 | } |
60939fb8 | 816 | } |
98994639 | 817 | } |
60939fb8 NC |
818 | numtype |= IS_NUMBER_IN_UV; |
819 | if (valuep) | |
820 | *valuep = value; | |
821 | ||
822 | skip_value: | |
823 | if (GROK_NUMERIC_RADIX(&s, send)) { | |
824 | numtype |= IS_NUMBER_NOT_INT; | |
825 | while (s < send && isDIGIT(*s)) /* optional digits after the radix */ | |
826 | s++; | |
98994639 | 827 | } |
60939fb8 NC |
828 | } |
829 | else if (GROK_NUMERIC_RADIX(&s, send)) { | |
830 | numtype |= IS_NUMBER_NOT_INT | IS_NUMBER_IN_UV; /* valuep assigned below */ | |
831 | /* no digits before the radix means we need digits after it */ | |
832 | if (s < send && isDIGIT(*s)) { | |
833 | do { | |
834 | s++; | |
835 | } while (s < send && isDIGIT(*s)); | |
836 | if (valuep) { | |
837 | /* integer approximation is valid - it's 0. */ | |
838 | *valuep = 0; | |
839 | } | |
98994639 | 840 | } |
60939fb8 | 841 | else |
ae776a2c | 842 | return 0; |
ff4eb398 JH |
843 | } |
844 | else { | |
9ff909cf | 845 | if (INFNAN_PEEK(d, send)) { |
ae776a2c JH |
846 | int infnan = Perl_grok_infnan(&d, send); |
847 | if ((infnan & IS_NUMBER_INFINITY)) { | |
848 | numtype |= infnan; | |
849 | sawinf = 1; | |
850 | } | |
851 | else if ((infnan & IS_NUMBER_NAN)) { | |
852 | numtype |= infnan; | |
853 | sawnan = 1; | |
854 | } | |
855 | else | |
856 | return 0; | |
857 | s = d; | |
ff4eb398 | 858 | } |
ff4eb398 | 859 | } |
60939fb8 NC |
860 | |
861 | if (sawinf) { | |
ff4eb398 | 862 | /* Keep the sign for infinity. */ |
60939fb8 | 863 | numtype |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT; |
aa8b85de | 864 | } else if (sawnan) { |
ff4eb398 | 865 | numtype &= IS_NUMBER_NEG; /* Clear sign for nan. */ |
aa8b85de | 866 | numtype |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT; |
60939fb8 NC |
867 | } else if (s < send) { |
868 | /* we can have an optional exponent part */ | |
305b8651 | 869 | if (isALPHA_FOLD_EQ(*s, 'e')) { |
60939fb8 NC |
870 | s++; |
871 | if (s < send && (*s == '-' || *s == '+')) | |
872 | s++; | |
873 | if (s < send && isDIGIT(*s)) { | |
874 | do { | |
875 | s++; | |
876 | } while (s < send && isDIGIT(*s)); | |
877 | } | |
3f7602fa TC |
878 | else if (flags & PERL_SCAN_TRAILING) |
879 | return numtype | IS_NUMBER_TRAILING; | |
60939fb8 | 880 | else |
3f7602fa TC |
881 | return 0; |
882 | ||
883 | /* The only flag we keep is sign. Blow away any "it's UV" */ | |
884 | numtype &= IS_NUMBER_NEG; | |
885 | numtype |= IS_NUMBER_NOT_INT; | |
60939fb8 NC |
886 | } |
887 | } | |
888 | while (s < send && isSPACE(*s)) | |
889 | s++; | |
890 | if (s >= send) | |
aa8b85de | 891 | return numtype; |
60939fb8 NC |
892 | if (len == 10 && memEQ(pv, "0 but true", 10)) { |
893 | if (valuep) | |
894 | *valuep = 0; | |
895 | return IS_NUMBER_IN_UV; | |
896 | } | |
3f7602fa TC |
897 | else if (flags & PERL_SCAN_TRAILING) { |
898 | return numtype | IS_NUMBER_TRAILING; | |
899 | } | |
900 | ||
60939fb8 | 901 | return 0; |
98994639 HS |
902 | } |
903 | ||
6313e544 | 904 | /* |
d62b8c6a | 905 | =for apidoc grok_atou |
6313e544 | 906 | |
d62b8c6a | 907 | grok_atou is a safer replacement for atoi and strtol. |
6313e544 | 908 | |
d62b8c6a JH |
909 | grok_atou parses a C-style zero-byte terminated string, looking for |
910 | a decimal unsigned integer. | |
338aa8b0 | 911 | |
d62b8c6a JH |
912 | Returns the unsigned integer, if a valid value can be parsed |
913 | from the beginning of the string. | |
f4379102 | 914 | |
d62b8c6a | 915 | Accepts only the decimal digits '0'..'9'. |
6313e544 | 916 | |
d62b8c6a JH |
917 | As opposed to atoi or strtol, grok_atou does NOT allow optional |
918 | leading whitespace, or negative inputs. If such features are | |
919 | required, the calling code needs to explicitly implement those. | |
6313e544 | 920 | |
d62b8c6a | 921 | If a valid value cannot be parsed, returns either zero (if non-digits |
75feedba | 922 | are met before any digits) or UV_MAX (if the value overflows). |
6313e544 | 923 | |
d62b8c6a JH |
924 | Note that extraneous leading zeros also count as an overflow |
925 | (meaning that only "0" is the zero). | |
338aa8b0 | 926 | |
d62b8c6a | 927 | On failure, the *endptr is also set to NULL, unless endptr is NULL. |
338aa8b0 JH |
928 | |
929 | Trailing non-digit bytes are allowed if the endptr is non-NULL. | |
6313e544 JH |
930 | On return the *endptr will contain the pointer to the first non-digit byte. |
931 | ||
6313e544 | 932 | If the endptr is NULL, the first non-digit byte MUST be |
f4379102 | 933 | the zero byte terminating the pv, or zero will be returned. |
6313e544 | 934 | |
d62b8c6a JH |
935 | Background: atoi has severe problems with illegal inputs, it cannot be |
936 | used for incremental parsing, and therefore should be avoided | |
937 | atoi and strtol are also affected by locale settings, which can also be | |
938 | seen as a bug (global state controlled by user environment). | |
939 | ||
6313e544 JH |
940 | =cut |
941 | */ | |
942 | ||
75feedba | 943 | UV |
6313e544 JH |
944 | Perl_grok_atou(const char *pv, const char** endptr) |
945 | { | |
946 | const char* s = pv; | |
947 | const char** eptr; | |
948 | const char* end2; /* Used in case endptr is NULL. */ | |
75feedba | 949 | UV val = 0; /* The return value. */ |
6313e544 JH |
950 | |
951 | PERL_ARGS_ASSERT_GROK_ATOU; | |
952 | ||
953 | eptr = endptr ? endptr : &end2; | |
75feedba JH |
954 | if (isDIGIT(*s)) { |
955 | /* Single-digit inputs are quite common. */ | |
6313e544 | 956 | val = *s++ - '0'; |
75feedba JH |
957 | if (isDIGIT(*s)) { |
958 | /* Extra leading zeros cause overflow. */ | |
959 | if (val == 0) { | |
960 | *eptr = NULL; | |
961 | return UV_MAX; | |
962 | } | |
963 | while (isDIGIT(*s)) { | |
964 | /* This could be unrolled like in grok_number(), but | |
965 | * the expected uses of this are not speed-needy, and | |
966 | * unlikely to need full 64-bitness. */ | |
967 | U8 digit = *s++ - '0'; | |
945b524a JH |
968 | if (val < uv_max_div_10 || |
969 | (val == uv_max_div_10 && digit <= uv_max_mod_10)) { | |
75feedba JH |
970 | val = val * 10 + digit; |
971 | } else { | |
6313e544 | 972 | *eptr = NULL; |
75feedba | 973 | return UV_MAX; |
6313e544 | 974 | } |
6313e544 JH |
975 | } |
976 | } | |
75feedba JH |
977 | } |
978 | if (s == pv) { | |
979 | *eptr = NULL; /* If no progress, failed to parse anything. */ | |
980 | return 0; | |
6313e544 JH |
981 | } |
982 | if (endptr == NULL && *s) { | |
983 | return 0; /* If endptr is NULL, no trailing non-digits allowed. */ | |
984 | } | |
985 | *eptr = s; | |
986 | return val; | |
987 | } | |
988 | ||
4801ca72 | 989 | STATIC NV |
98994639 HS |
990 | S_mulexp10(NV value, I32 exponent) |
991 | { | |
992 | NV result = 1.0; | |
993 | NV power = 10.0; | |
994 | bool negative = 0; | |
995 | I32 bit; | |
996 | ||
997 | if (exponent == 0) | |
998 | return value; | |
659c4b96 DM |
999 | if (value == 0) |
1000 | return (NV)0; | |
87032ba1 | 1001 | |
24866caa | 1002 | /* On OpenVMS VAX we by default use the D_FLOAT double format, |
67597c89 | 1003 | * and that format does not have *easy* capabilities [1] for |
24866caa CB |
1004 | * overflowing doubles 'silently' as IEEE fp does. We also need |
1005 | * to support G_FLOAT on both VAX and Alpha, and though the exponent | |
1006 | * range is much larger than D_FLOAT it still doesn't do silent | |
1007 | * overflow. Therefore we need to detect early whether we would | |
1008 | * overflow (this is the behaviour of the native string-to-float | |
1009 | * conversion routines, and therefore of native applications, too). | |
67597c89 | 1010 | * |
24866caa CB |
1011 | * [1] Trying to establish a condition handler to trap floating point |
1012 | * exceptions is not a good idea. */ | |
87032ba1 JH |
1013 | |
1014 | /* In UNICOS and in certain Cray models (such as T90) there is no | |
1015 | * IEEE fp, and no way at all from C to catch fp overflows gracefully. | |
1016 | * There is something you can do if you are willing to use some | |
1017 | * inline assembler: the instruction is called DFI-- but that will | |
1018 | * disable *all* floating point interrupts, a little bit too large | |
1019 | * a hammer. Therefore we need to catch potential overflows before | |
1020 | * it's too late. */ | |
353813d9 | 1021 | |
85bba25f | 1022 | #if ((defined(VMS) && !defined(_IEEE_FP)) || defined(_UNICOS)) && defined(NV_MAX_10_EXP) |
353813d9 | 1023 | STMT_START { |
c4420975 | 1024 | const NV exp_v = log10(value); |
353813d9 HS |
1025 | if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP) |
1026 | return NV_MAX; | |
1027 | if (exponent < 0) { | |
1028 | if (-(exponent + exp_v) >= NV_MAX_10_EXP) | |
1029 | return 0.0; | |
1030 | while (-exponent >= NV_MAX_10_EXP) { | |
1031 | /* combination does not overflow, but 10^(-exponent) does */ | |
1032 | value /= 10; | |
1033 | ++exponent; | |
1034 | } | |
1035 | } | |
1036 | } STMT_END; | |
87032ba1 JH |
1037 | #endif |
1038 | ||
353813d9 HS |
1039 | if (exponent < 0) { |
1040 | negative = 1; | |
1041 | exponent = -exponent; | |
b27804d8 DM |
1042 | #ifdef NV_MAX_10_EXP |
1043 | /* for something like 1234 x 10^-309, the action of calculating | |
1044 | * the intermediate value 10^309 then returning 1234 / (10^309) | |
1045 | * will fail, since 10^309 becomes infinity. In this case try to | |
1046 | * refactor it as 123 / (10^308) etc. | |
1047 | */ | |
1048 | while (value && exponent > NV_MAX_10_EXP) { | |
1049 | exponent--; | |
1050 | value /= 10; | |
1051 | } | |
1052 | #endif | |
353813d9 | 1053 | } |
98994639 HS |
1054 | for (bit = 1; exponent; bit <<= 1) { |
1055 | if (exponent & bit) { | |
1056 | exponent ^= bit; | |
1057 | result *= power; | |
236f0012 CB |
1058 | /* Floating point exceptions are supposed to be turned off, |
1059 | * but if we're obviously done, don't risk another iteration. | |
1060 | */ | |
1061 | if (exponent == 0) break; | |
98994639 HS |
1062 | } |
1063 | power *= power; | |
1064 | } | |
1065 | return negative ? value / result : value * result; | |
1066 | } | |
1067 | ||
1068 | NV | |
1069 | Perl_my_atof(pTHX_ const char* s) | |
1070 | { | |
1071 | NV x = 0.0; | |
1072 | #ifdef USE_LOCALE_NUMERIC | |
7918f24d NC |
1073 | PERL_ARGS_ASSERT_MY_ATOF; |
1074 | ||
a2287a13 KW |
1075 | { |
1076 | DECLARE_STORE_LC_NUMERIC_SET_TO_NEEDED(); | |
d6ded950 | 1077 | if (PL_numeric_radix_sv && IN_LC(LC_NUMERIC)) { |
e4850248 KW |
1078 | const char *standard = NULL, *local = NULL; |
1079 | bool use_standard_radix; | |
98994639 | 1080 | |
e4850248 KW |
1081 | /* Look through the string for the first thing that looks like a |
1082 | * decimal point: either the value in the current locale or the | |
1083 | * standard fallback of '.'. The one which appears earliest in the | |
1084 | * input string is the one that we should have atof look for. Note | |
1085 | * that we have to determine this beforehand because on some | |
1086 | * systems, Perl_atof2 is just a wrapper around the system's atof. | |
1087 | * */ | |
1088 | standard = strchr(s, '.'); | |
1089 | local = strstr(s, SvPV_nolen(PL_numeric_radix_sv)); | |
78787052 | 1090 | |
e4850248 | 1091 | use_standard_radix = standard && (!local || standard < local); |
78787052 | 1092 | |
e4850248 KW |
1093 | if (use_standard_radix) |
1094 | SET_NUMERIC_STANDARD(); | |
78787052 | 1095 | |
e4850248 | 1096 | Perl_atof2(s, x); |
78787052 | 1097 | |
e4850248 KW |
1098 | if (use_standard_radix) |
1099 | SET_NUMERIC_LOCAL(); | |
1100 | } | |
1101 | else | |
1102 | Perl_atof2(s, x); | |
a2287a13 KW |
1103 | RESTORE_LC_NUMERIC(); |
1104 | } | |
98994639 | 1105 | #else |
a36244b7 | 1106 | Perl_atof2(s, x); |
98994639 HS |
1107 | #endif |
1108 | return x; | |
1109 | } | |
1110 | ||
1111 | char* | |
1112 | Perl_my_atof2(pTHX_ const char* orig, NV* value) | |
1113 | { | |
20f6aaab | 1114 | NV result[3] = {0.0, 0.0, 0.0}; |
e1ec3a88 | 1115 | const char* s = orig; |
a36244b7 | 1116 | #ifdef USE_PERL_ATOF |
20f6aaab | 1117 | UV accumulator[2] = {0,0}; /* before/after dp */ |
a36244b7 | 1118 | bool negative = 0; |
ae776a2c | 1119 | const char* send = s + strlen(orig); /* one past the last */ |
8194bf88 | 1120 | bool seen_digit = 0; |
20f6aaab AS |
1121 | I32 exp_adjust[2] = {0,0}; |
1122 | I32 exp_acc[2] = {-1, -1}; | |
1123 | /* the current exponent adjust for the accumulators */ | |
98994639 | 1124 | I32 exponent = 0; |
8194bf88 | 1125 | I32 seen_dp = 0; |
20f6aaab AS |
1126 | I32 digit = 0; |
1127 | I32 old_digit = 0; | |
8194bf88 DM |
1128 | I32 sig_digits = 0; /* noof significant digits seen so far */ |
1129 | ||
7918f24d NC |
1130 | PERL_ARGS_ASSERT_MY_ATOF2; |
1131 | ||
8194bf88 DM |
1132 | /* There is no point in processing more significant digits |
1133 | * than the NV can hold. Note that NV_DIG is a lower-bound value, | |
1134 | * while we need an upper-bound value. We add 2 to account for this; | |
1135 | * since it will have been conservative on both the first and last digit. | |
1136 | * For example a 32-bit mantissa with an exponent of 4 would have | |
1137 | * exact values in the set | |
1138 | * 4 | |
1139 | * 8 | |
1140 | * .. | |
1141 | * 17179869172 | |
1142 | * 17179869176 | |
1143 | * 17179869180 | |
1144 | * | |
1145 | * where for the purposes of calculating NV_DIG we would have to discount | |
1146 | * both the first and last digit, since neither can hold all values from | |
1147 | * 0..9; but for calculating the value we must examine those two digits. | |
1148 | */ | |
ffa277e5 AS |
1149 | #ifdef MAX_SIG_DIG_PLUS |
1150 | /* It is not necessarily the case that adding 2 to NV_DIG gets all the | |
1151 | possible digits in a NV, especially if NVs are not IEEE compliant | |
1152 | (e.g., long doubles on IRIX) - Allen <allens@cpan.org> */ | |
1153 | # define MAX_SIG_DIGITS (NV_DIG+MAX_SIG_DIG_PLUS) | |
1154 | #else | |
1155 | # define MAX_SIG_DIGITS (NV_DIG+2) | |
1156 | #endif | |
8194bf88 DM |
1157 | |
1158 | /* the max number we can accumulate in a UV, and still safely do 10*N+9 */ | |
1159 | #define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10)) | |
98994639 | 1160 | |
96a05aee HS |
1161 | /* leading whitespace */ |
1162 | while (isSPACE(*s)) | |
1163 | ++s; | |
1164 | ||
98994639 HS |
1165 | /* sign */ |
1166 | switch (*s) { | |
1167 | case '-': | |
1168 | negative = 1; | |
924ba076 | 1169 | /* FALLTHROUGH */ |
98994639 HS |
1170 | case '+': |
1171 | ++s; | |
1172 | } | |
1173 | ||
ae776a2c JH |
1174 | { |
1175 | const char *p0 = negative ? s - 1 : s; | |
1176 | const char *p = p0; | |
1177 | #if defined(NV_INF) && defined(NV_NAN) | |
1178 | int infnan_flags = grok_infnan(&p, send); | |
1179 | if (infnan_flags && p != p0) { | |
1180 | if ((infnan_flags & IS_NUMBER_INFINITY)) { | |
1181 | *value = (infnan_flags & IS_NUMBER_NEG) ? -NV_INF: NV_INF; | |
1182 | return (char*)p; | |
1183 | } | |
1184 | else if ((infnan_flags & IS_NUMBER_NAN)) { | |
1185 | *value = NV_NAN; | |
1186 | return (char*)p; | |
1187 | } | |
1188 | } | |
1189 | #elif defined(HAS_STRTOD) | |
9ff909cf | 1190 | if (INFNAN_PEEK(s, send)) { |
ae776a2c | 1191 | /* The native strtod() may not get all the possible |
9ff909cf | 1192 | * inf/nan strings INFNAN_PEEK() recognizes. */ |
ae776a2c | 1193 | char* endp; |
f1b32d61 | 1194 | NV nv = Perl_strtod(p, &endp); |
ae776a2c JH |
1195 | if (p != endp) { |
1196 | *value = nv; | |
1197 | return endp; | |
1198 | } | |
2b54f59f | 1199 | } |
2b54f59f | 1200 | #endif |
ae776a2c | 1201 | } |
2b54f59f | 1202 | |
8194bf88 DM |
1203 | /* we accumulate digits into an integer; when this becomes too |
1204 | * large, we add the total to NV and start again */ | |
98994639 | 1205 | |
8194bf88 DM |
1206 | while (1) { |
1207 | if (isDIGIT(*s)) { | |
1208 | seen_digit = 1; | |
20f6aaab | 1209 | old_digit = digit; |
8194bf88 | 1210 | digit = *s++ - '0'; |
20f6aaab AS |
1211 | if (seen_dp) |
1212 | exp_adjust[1]++; | |
98994639 | 1213 | |
8194bf88 DM |
1214 | /* don't start counting until we see the first significant |
1215 | * digit, eg the 5 in 0.00005... */ | |
1216 | if (!sig_digits && digit == 0) | |
1217 | continue; | |
1218 | ||
1219 | if (++sig_digits > MAX_SIG_DIGITS) { | |
98994639 | 1220 | /* limits of precision reached */ |
20f6aaab AS |
1221 | if (digit > 5) { |
1222 | ++accumulator[seen_dp]; | |
1223 | } else if (digit == 5) { | |
1224 | if (old_digit % 2) { /* round to even - Allen */ | |
1225 | ++accumulator[seen_dp]; | |
1226 | } | |
1227 | } | |
1228 | if (seen_dp) { | |
1229 | exp_adjust[1]--; | |
1230 | } else { | |
1231 | exp_adjust[0]++; | |
1232 | } | |
8194bf88 | 1233 | /* skip remaining digits */ |
98994639 | 1234 | while (isDIGIT(*s)) { |
98994639 | 1235 | ++s; |
20f6aaab AS |
1236 | if (! seen_dp) { |
1237 | exp_adjust[0]++; | |
1238 | } | |
98994639 HS |
1239 | } |
1240 | /* warn of loss of precision? */ | |
98994639 | 1241 | } |
8194bf88 | 1242 | else { |
20f6aaab | 1243 | if (accumulator[seen_dp] > MAX_ACCUMULATE) { |
8194bf88 | 1244 | /* add accumulator to result and start again */ |
20f6aaab AS |
1245 | result[seen_dp] = S_mulexp10(result[seen_dp], |
1246 | exp_acc[seen_dp]) | |
1247 | + (NV)accumulator[seen_dp]; | |
1248 | accumulator[seen_dp] = 0; | |
1249 | exp_acc[seen_dp] = 0; | |
98994639 | 1250 | } |
20f6aaab AS |
1251 | accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit; |
1252 | ++exp_acc[seen_dp]; | |
98994639 | 1253 | } |
8194bf88 | 1254 | } |
e1ec3a88 | 1255 | else if (!seen_dp && GROK_NUMERIC_RADIX(&s, send)) { |
8194bf88 | 1256 | seen_dp = 1; |
20f6aaab | 1257 | if (sig_digits > MAX_SIG_DIGITS) { |
c86f7df5 | 1258 | do { |
20f6aaab | 1259 | ++s; |
c86f7df5 | 1260 | } while (isDIGIT(*s)); |
20f6aaab AS |
1261 | break; |
1262 | } | |
8194bf88 DM |
1263 | } |
1264 | else { | |
1265 | break; | |
98994639 HS |
1266 | } |
1267 | } | |
1268 | ||
20f6aaab AS |
1269 | result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0]; |
1270 | if (seen_dp) { | |
1271 | result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1]; | |
1272 | } | |
98994639 | 1273 | |
305b8651 | 1274 | if (seen_digit && (isALPHA_FOLD_EQ(*s, 'e'))) { |
98994639 HS |
1275 | bool expnegative = 0; |
1276 | ||
1277 | ++s; | |
1278 | switch (*s) { | |
1279 | case '-': | |
1280 | expnegative = 1; | |
924ba076 | 1281 | /* FALLTHROUGH */ |
98994639 HS |
1282 | case '+': |
1283 | ++s; | |
1284 | } | |
1285 | while (isDIGIT(*s)) | |
1286 | exponent = exponent * 10 + (*s++ - '0'); | |
1287 | if (expnegative) | |
1288 | exponent = -exponent; | |
1289 | } | |
1290 | ||
20f6aaab AS |
1291 | |
1292 | ||
98994639 | 1293 | /* now apply the exponent */ |
20f6aaab AS |
1294 | |
1295 | if (seen_dp) { | |
1296 | result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]) | |
1297 | + S_mulexp10(result[1],exponent-exp_adjust[1]); | |
1298 | } else { | |
1299 | result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]); | |
1300 | } | |
98994639 HS |
1301 | |
1302 | /* now apply the sign */ | |
1303 | if (negative) | |
20f6aaab | 1304 | result[2] = -result[2]; |
a36244b7 | 1305 | #endif /* USE_PERL_ATOF */ |
20f6aaab | 1306 | *value = result[2]; |
73d840c0 | 1307 | return (char *)s; |
98994639 HS |
1308 | } |
1309 | ||
55954f19 JH |
1310 | #if ! defined(HAS_MODFL) && defined(HAS_AINTL) && defined(HAS_COPYSIGNL) |
1311 | long double | |
1312 | Perl_my_modfl(long double x, long double *ip) | |
1313 | { | |
1314 | *ip = aintl(x); | |
1315 | return (x == *ip ? copysignl(0.0L, x) : x - *ip); | |
1316 | } | |
1317 | #endif | |
1318 | ||
1319 | #if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL) | |
1320 | long double | |
1321 | Perl_my_frexpl(long double x, int *e) { | |
1322 | *e = x == 0.0L ? 0 : ilogbl(x) + 1; | |
1323 | return (scalbnl(x, -*e)); | |
1324 | } | |
1325 | #endif | |
66610fdd RGS |
1326 | |
1327 | /* | |
ed140128 AD |
1328 | =for apidoc Perl_signbit |
1329 | ||
1330 | Return a non-zero integer if the sign bit on an NV is set, and 0 if | |
1331 | it is not. | |
1332 | ||
1333 | If Configure detects this system has a signbit() that will work with | |
1334 | our NVs, then we just use it via the #define in perl.h. Otherwise, | |
8b7fad81 JH |
1335 | fall back on this implementation. The main use of this function |
1336 | is catching -0.0. | |
ed140128 AD |
1337 | |
1338 | Configure notes: This function is called 'Perl_signbit' instead of a | |
1339 | plain 'signbit' because it is easy to imagine a system having a signbit() | |
1340 | function or macro that doesn't happen to work with our particular choice | |
1341 | of NVs. We shouldn't just re-#define signbit as Perl_signbit and expect | |
1342 | the standard system headers to be happy. Also, this is a no-context | |
1343 | function (no pTHX_) because Perl_signbit() is usually re-#defined in | |
1344 | perl.h as a simple macro call to the system's signbit(). | |
1345 | Users should just always call Perl_signbit(). | |
1346 | ||
1347 | =cut | |
1348 | */ | |
1349 | #if !defined(HAS_SIGNBIT) | |
1350 | int | |
1351 | Perl_signbit(NV x) { | |
8b7fad81 JH |
1352 | # ifdef Perl_fp_class_nzero |
1353 | if (x == 0) | |
1354 | return Perl_fp_class_nzero(x); | |
8b7fad81 | 1355 | # endif |
3585840c | 1356 | return (x < 0.0) ? 1 : 0; |
ed140128 AD |
1357 | } |
1358 | #endif | |
1359 | ||
1360 | /* | |
66610fdd RGS |
1361 | * Local variables: |
1362 | * c-indentation-style: bsd | |
1363 | * c-basic-offset: 4 | |
14d04a33 | 1364 | * indent-tabs-mode: nil |
66610fdd RGS |
1365 | * End: |
1366 | * | |
14d04a33 | 1367 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 1368 | */ |