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; | |
071db91b | 42 | return ((U32) f) | (1 + (U32_MAX >> 1)); |
98994639 HS |
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; | |
071db91b | 60 | return (I32)(((U32) f) | (1 + (U32_MAX >> 1))); |
98994639 HS |
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; | |
071db91b | 79 | return (IV)(((UV) f) | (1 + (UV_MAX >> 1))); |
98994639 HS |
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; | |
071db91b | 97 | return ((UV) f) | (1 + (UV_MAX >> 1)); |
98994639 HS |
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 | ||
2d7f6611 | 110 | On entry C<start> and C<*len> give the string to scan, C<*flags> gives |
796b6530 | 111 | conversion flags, and C<result> should be C<NULL> or a pointer to an NV. |
53305cf1 | 112 | The scan stops at the end of the string, or the first invalid character. |
2d7f6611 | 113 | Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in C<*flags>, encountering an |
7b667b5f | 114 | invalid character will also trigger a warning. |
2d7f6611 KW |
115 | On return C<*len> is set to the length of the scanned string, |
116 | and C<*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, |
796b6530 KW |
119 | and nothing is written to C<*result>. If the value is > C<UV_MAX>, C<grok_bin> |
120 | returns C<UV_MAX>, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags, | |
2d7f6611 | 121 | and writes the value to C<*result> (or the value is discarded if C<result> |
53305cf1 NC |
122 | is NULL). |
123 | ||
796b6530 | 124 | The binary number may optionally be prefixed with C<"0b"> or C<"b"> unless |
2d7f6611 KW |
125 | C<PERL_SCAN_DISALLOW_PREFIX> is set in C<*flags> on entry. If |
126 | C<PERL_SCAN_ALLOW_UNDERSCORES> is set in C<*flags> then the binary | |
796b6530 | 127 | number may use C<"_"> characters to separate digits. |
53305cf1 NC |
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 | } |
19c1206d | 207 | |
53305cf1 | 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 | ||
2d7f6611 | 233 | On entry C<start> and C<*len_p> give the string to scan, C<*flags> gives |
796b6530 | 234 | conversion flags, and C<result> should be C<NULL> or a pointer to an NV. |
7b667b5f | 235 | The scan stops at the end of the string, or the first invalid character. |
2d7f6611 | 236 | Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in C<*flags>, encountering an |
7b667b5f | 237 | invalid character will also trigger a warning. |
2d7f6611 KW |
238 | On return C<*len> is set to the length of the scanned string, |
239 | and C<*flags> gives output flags. | |
53305cf1 | 240 | |
796b6530 KW |
241 | If the value is <= C<UV_MAX> it is returned as a UV, the output flags are clear, |
242 | and nothing is written to C<*result>. If the value is > C<UV_MAX>, C<grok_hex> | |
243 | returns C<UV_MAX>, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags, | |
2d7f6611 | 244 | and writes the value to C<*result> (or the value is discarded if C<result> |
796b6530 | 245 | is C<NULL>). |
53305cf1 | 246 | |
796b6530 | 247 | The hex number may optionally be prefixed with C<"0x"> or C<"x"> unless |
2d7f6611 KW |
248 | C<PERL_SCAN_DISALLOW_PREFIX> is set in C<*flags> on entry. If |
249 | C<PERL_SCAN_ALLOW_UNDERSCORES> is set in C<*flags> then the hex | |
796b6530 | 250 | number may use C<"_"> characters to separate digits. |
53305cf1 NC |
251 | |
252 | =cut | |
02470786 KW |
253 | |
254 | Not documented yet because experimental is C<PERL_SCAN_SILENT_NON_PORTABLE | |
baf48926 | 255 | which suppresses any message for non-portable numbers, but which are valid |
02470786 | 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 | } | |
19c1206d | 328 | |
53305cf1 NC |
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 | ||
2d7f6611 | 354 | On entry C<start> and C<*len> give the string to scan, C<*flags> gives |
796b6530 | 355 | conversion flags, and C<result> should be C<NULL> or a pointer to an NV. |
7b667b5f | 356 | The scan stops at the end of the string, or the first invalid character. |
2d7f6611 | 357 | Unless C<PERL_SCAN_SILENT_ILLDIGIT> is set in C<*flags>, encountering an |
154bd527 | 358 | 8 or 9 will also trigger a warning. |
2d7f6611 KW |
359 | On return C<*len> is set to the length of the scanned string, |
360 | and C<*flags> gives output flags. | |
7b667b5f | 361 | |
796b6530 KW |
362 | If the value is <= C<UV_MAX> it is returned as a UV, the output flags are clear, |
363 | and nothing is written to C<*result>. If the value is > C<UV_MAX>, C<grok_oct> | |
364 | returns C<UV_MAX>, sets C<PERL_SCAN_GREATER_THAN_UV_MAX> in the output flags, | |
2d7f6611 | 365 | and writes the value to C<*result> (or the value is discarded if C<result> |
796b6530 | 366 | is C<NULL>). |
7b667b5f | 367 | |
2d7f6611 | 368 | If C<PERL_SCAN_ALLOW_UNDERSCORES> is set in C<*flags> then the octal |
796b6530 | 369 | number may use C<"_"> 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 | } |
19c1206d | 435 | |
53305cf1 | 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 | { | |
7918f24d NC |
521 | PERL_ARGS_ASSERT_GROK_NUMERIC_RADIX; |
522 | ||
7ea85fa8 KW |
523 | #ifdef USE_LOCALE_NUMERIC |
524 | ||
d6ded950 | 525 | if (IN_LC(LC_NUMERIC)) { |
f0dafd73 KW |
526 | STRLEN len; |
527 | char * radix; | |
528 | bool matches_radix = FALSE; | |
67d796ae | 529 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; |
f0dafd73 | 530 | |
a1395eaf | 531 | STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); |
f0dafd73 KW |
532 | |
533 | radix = SvPV(PL_numeric_radix_sv, len); | |
534 | radix = savepvn(radix, len); | |
535 | ||
21431899 | 536 | RESTORE_LC_NUMERIC(); |
f0dafd73 KW |
537 | |
538 | if (*sp + len <= send) { | |
539 | matches_radix = memEQ(*sp, radix, len); | |
540 | } | |
541 | ||
542 | Safefree(radix); | |
543 | ||
544 | if (matches_radix) { | |
545 | *sp += len; | |
546 | return TRUE; | |
547 | } | |
98994639 | 548 | } |
f0dafd73 | 549 | |
98994639 | 550 | #endif |
7918f24d | 551 | |
f0dafd73 KW |
552 | /* always try "." if numeric radix didn't match because |
553 | * we may have data from different locales mixed */ | |
98994639 HS |
554 | if (*sp < send && **sp == '.') { |
555 | ++*sp; | |
556 | return TRUE; | |
557 | } | |
f0dafd73 | 558 | |
98994639 HS |
559 | return FALSE; |
560 | } | |
561 | ||
569f27e5 | 562 | /* |
ff4eb398 JH |
563 | =for apidoc grok_infnan |
564 | ||
796b6530 | 565 | Helper for C<grok_number()>, accepts various ways of spelling "infinity" |
ff4eb398 JH |
566 | or "not a number", and returns one of the following flag combinations: |
567 | ||
5962c2f6 | 568 | IS_NUMBER_INFINITY |
ff4eb398 | 569 | IS_NUMBER_NAN |
5962c2f6 | 570 | IS_NUMBER_INFINITY | IS_NUMBER_NEG |
ff4eb398 JH |
571 | IS_NUMBER_NAN | IS_NUMBER_NEG |
572 | 0 | |
573 | ||
796b6530 | 574 | possibly |-ed with C<IS_NUMBER_TRAILING>. |
b489e20f | 575 | |
796b6530 | 576 | If an infinity or a not-a-number is recognized, C<*sp> will point to |
62bdc035 | 577 | one byte past the end of the recognized string. If the recognition fails, |
796b6530 | 578 | zero is returned, and C<*sp> will not move. |
ff4eb398 JH |
579 | |
580 | =cut | |
581 | */ | |
582 | ||
583 | int | |
3823048b | 584 | Perl_grok_infnan(pTHX_ const char** sp, const char* send) |
ff4eb398 JH |
585 | { |
586 | const char* s = *sp; | |
587 | int flags = 0; | |
a5dc2484 | 588 | #if defined(NV_INF) || defined(NV_NAN) |
62bdc035 | 589 | bool odh = FALSE; /* one-dot-hash: 1.#INF */ |
ff4eb398 JH |
590 | |
591 | PERL_ARGS_ASSERT_GROK_INFNAN; | |
592 | ||
8c12dc63 JH |
593 | if (*s == '+') { |
594 | s++; if (s == send) return 0; | |
595 | } | |
596 | else if (*s == '-') { | |
ff4eb398 JH |
597 | flags |= IS_NUMBER_NEG; /* Yes, -NaN happens. Incorrect but happens. */ |
598 | s++; if (s == send) return 0; | |
599 | } | |
600 | ||
601 | if (*s == '1') { | |
62bdc035 JH |
602 | /* Visual C: 1.#SNAN, -1.#QNAN, 1#INF, 1.#IND (maybe also 1.#NAN) |
603 | * Let's keep the dot optional. */ | |
ff4eb398 JH |
604 | s++; if (s == send) return 0; |
605 | if (*s == '.') { | |
606 | s++; if (s == send) return 0; | |
607 | } | |
608 | if (*s == '#') { | |
609 | s++; if (s == send) return 0; | |
610 | } else | |
611 | return 0; | |
e855f543 | 612 | odh = TRUE; |
ff4eb398 JH |
613 | } |
614 | ||
305b8651 | 615 | if (isALPHA_FOLD_EQ(*s, 'I')) { |
62bdc035 JH |
616 | /* INF or IND (1.#IND is "indeterminate", a certain type of NAN) */ |
617 | ||
305b8651 | 618 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0; |
ff4eb398 | 619 | s++; if (s == send) return 0; |
305b8651 | 620 | if (isALPHA_FOLD_EQ(*s, 'F')) { |
ff4eb398 | 621 | s++; |
b8974fcb JH |
622 | if (s < send && (isALPHA_FOLD_EQ(*s, 'I'))) { |
623 | int fail = | |
624 | flags | IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT | IS_NUMBER_TRAILING; | |
625 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return fail; | |
626 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'I')) return fail; | |
627 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'T')) return fail; | |
628 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'Y')) return fail; | |
3396ed30 | 629 | s++; |
b8974fcb JH |
630 | } else if (odh) { |
631 | while (*s == '0') { /* 1.#INF00 */ | |
632 | s++; | |
633 | } | |
3396ed30 | 634 | } |
b489e20f JH |
635 | while (s < send && isSPACE(*s)) |
636 | s++; | |
637 | if (s < send && *s) { | |
3396ed30 | 638 | flags |= IS_NUMBER_TRAILING; |
fae4db12 | 639 | } |
ff4eb398 JH |
640 | flags |= IS_NUMBER_INFINITY | IS_NUMBER_NOT_INT; |
641 | } | |
e855f543 | 642 | else if (isALPHA_FOLD_EQ(*s, 'D') && odh) { /* 1.#IND */ |
ff4eb398 JH |
643 | s++; |
644 | flags |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT; | |
fae4db12 JH |
645 | while (*s == '0') { /* 1.#IND00 */ |
646 | s++; | |
647 | } | |
1e9aa12f JH |
648 | if (*s) { |
649 | flags |= IS_NUMBER_TRAILING; | |
650 | } | |
ff4eb398 JH |
651 | } else |
652 | return 0; | |
ff4eb398 JH |
653 | } |
654 | else { | |
62bdc035 | 655 | /* Maybe NAN of some sort */ |
3823048b JH |
656 | |
657 | if (isALPHA_FOLD_EQ(*s, 'S') || isALPHA_FOLD_EQ(*s, 'Q')) { | |
658 | /* snan, qNaN */ | |
659 | /* XXX do something with the snan/qnan difference */ | |
660 | s++; if (s == send) return 0; | |
661 | } | |
662 | ||
663 | if (isALPHA_FOLD_EQ(*s, 'N')) { | |
664 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'A')) return 0; | |
665 | s++; if (s == send || isALPHA_FOLD_NE(*s, 'N')) return 0; | |
666 | s++; | |
667 | ||
668 | flags |= IS_NUMBER_NAN | IS_NUMBER_NOT_INT; | |
669 | ||
670 | /* NaN can be followed by various stuff (NaNQ, NaNS), but | |
671 | * there are also multiple different NaN values, and some | |
672 | * implementations output the "payload" values, | |
673 | * e.g. NaN123, NAN(abc), while some legacy implementations | |
674 | * have weird stuff like NaN%. */ | |
675 | if (isALPHA_FOLD_EQ(*s, 'q') || | |
676 | isALPHA_FOLD_EQ(*s, 's')) { | |
677 | /* "nanq" or "nans" are ok, though generating | |
678 | * these portably is tricky. */ | |
679 | s++; | |
680 | } | |
681 | if (*s == '(') { | |
682 | /* C99 style "nan(123)" or Perlish equivalent "nan($uv)". */ | |
683 | const char *t; | |
684 | s++; | |
685 | if (s == send) { | |
686 | return flags | IS_NUMBER_TRAILING; | |
687 | } | |
688 | t = s + 1; | |
689 | while (t < send && *t && *t != ')') { | |
690 | t++; | |
691 | } | |
692 | if (t == send) { | |
693 | return flags | IS_NUMBER_TRAILING; | |
694 | } | |
695 | if (*t == ')') { | |
696 | int nantype; | |
697 | UV nanval; | |
698 | if (s[0] == '0' && s + 2 < t && | |
699 | isALPHA_FOLD_EQ(s[1], 'x') && | |
700 | isXDIGIT(s[2])) { | |
701 | STRLEN len = t - s; | |
702 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES; | |
703 | nanval = grok_hex(s, &len, &flags, NULL); | |
704 | if ((flags & PERL_SCAN_GREATER_THAN_UV_MAX)) { | |
705 | nantype = 0; | |
706 | } else { | |
707 | nantype = IS_NUMBER_IN_UV; | |
708 | } | |
709 | s += len; | |
710 | } else if (s[0] == '0' && s + 2 < t && | |
711 | isALPHA_FOLD_EQ(s[1], 'b') && | |
712 | (s[2] == '0' || s[2] == '1')) { | |
713 | STRLEN len = t - s; | |
714 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES; | |
715 | nanval = grok_bin(s, &len, &flags, NULL); | |
716 | if ((flags & PERL_SCAN_GREATER_THAN_UV_MAX)) { | |
717 | nantype = 0; | |
718 | } else { | |
719 | nantype = IS_NUMBER_IN_UV; | |
720 | } | |
721 | s += len; | |
722 | } else { | |
723 | const char *u; | |
724 | nantype = | |
725 | grok_number_flags(s, t - s, &nanval, | |
726 | PERL_SCAN_TRAILING | | |
727 | PERL_SCAN_ALLOW_UNDERSCORES); | |
728 | /* Unfortunately grok_number_flags() doesn't | |
729 | * tell how far we got and the ')' will always | |
730 | * be "trailing", so we need to double-check | |
731 | * whether we had something dubious. */ | |
732 | for (u = s; u < t; u++) { | |
733 | if (!isDIGIT(*u)) { | |
734 | flags |= IS_NUMBER_TRAILING; | |
735 | break; | |
736 | } | |
737 | } | |
738 | s = u; | |
739 | } | |
740 | ||
741 | /* XXX Doesn't do octal: nan("0123"). | |
742 | * Probably not a big loss. */ | |
743 | ||
744 | if ((nantype & IS_NUMBER_NOT_INT) || | |
745 | !(nantype && IS_NUMBER_IN_UV)) { | |
746 | /* XXX the nanval is currently unused, that is, | |
747 | * not inserted as the NaN payload of the NV. | |
748 | * But the above code already parses the C99 | |
749 | * nan(...) format. See below, and see also | |
750 | * the nan() in POSIX.xs. | |
751 | * | |
752 | * Certain configuration combinations where | |
753 | * NVSIZE is greater than UVSIZE mean that | |
754 | * a single UV cannot contain all the possible | |
755 | * NaN payload bits. There would need to be | |
756 | * some more generic syntax than "nan($uv)". | |
757 | * | |
758 | * Issues to keep in mind: | |
759 | * | |
760 | * (1) In most common cases there would | |
761 | * not be an integral number of bytes that | |
762 | * could be set, only a certain number of bits. | |
763 | * For example for the common case of | |
764 | * NVSIZE == UVSIZE == 8 there is room for 52 | |
765 | * bits in the payload, but the most significant | |
766 | * bit is commonly reserved for the | |
767 | * signaling/quiet bit, leaving 51 bits. | |
768 | * Furthermore, the C99 nan() is supposed | |
769 | * to generate quiet NaNs, so it is doubtful | |
770 | * whether it should be able to generate | |
771 | * signaling NaNs. For the x86 80-bit doubles | |
772 | * (if building a long double Perl) there would | |
773 | * be 62 bits (s/q bit being the 63rd). | |
774 | * | |
775 | * (2) Endianness of the payload bits. If the | |
776 | * payload is specified as an UV, the low-order | |
777 | * bits of the UV are naturally little-endianed | |
778 | * (rightmost) bits of the payload. The endianness | |
779 | * of UVs and NVs can be different. */ | |
780 | return 0; | |
781 | } | |
782 | if (s < t) { | |
783 | flags |= IS_NUMBER_TRAILING; | |
784 | } | |
785 | } else { | |
786 | /* Looked like nan(...), but no close paren. */ | |
787 | flags |= IS_NUMBER_TRAILING; | |
788 | } | |
789 | } else { | |
790 | while (s < send && isSPACE(*s)) | |
791 | s++; | |
792 | if (s < send && *s) { | |
793 | /* Note that we here implicitly accept (parse as | |
794 | * "nan", but with warnings) also any other weird | |
795 | * trailing stuff for "nan". In the above we just | |
796 | * check that if we got the C99-style "nan(...)", | |
797 | * the "..." looks sane. | |
798 | * If in future we accept more ways of specifying | |
799 | * the nan payload, the accepting would happen around | |
800 | * here. */ | |
801 | flags |= IS_NUMBER_TRAILING; | |
802 | } | |
803 | } | |
804 | s = send; | |
805 | } | |
806 | else | |
807 | return 0; | |
ff4eb398 JH |
808 | } |
809 | ||
b489e20f JH |
810 | while (s < send && isSPACE(*s)) |
811 | s++; | |
812 | ||
a5dc2484 JH |
813 | #else |
814 | PERL_UNUSED_ARG(send); | |
815 | #endif /* #if defined(NV_INF) || defined(NV_NAN) */ | |
a1fe7cea JH |
816 | *sp = s; |
817 | return flags; | |
ff4eb398 JH |
818 | } |
819 | ||
13393a5e | 820 | /* |
3823048b | 821 | =for apidoc grok_number_flags |
13393a5e JH |
822 | |
823 | Recognise (or not) a number. The type of the number is returned | |
824 | (0 if unrecognised), otherwise it is a bit-ORed combination of | |
796b6530 KW |
825 | C<IS_NUMBER_IN_UV>, C<IS_NUMBER_GREATER_THAN_UV_MAX>, C<IS_NUMBER_NOT_INT>, |
826 | C<IS_NUMBER_NEG>, C<IS_NUMBER_INFINITY>, C<IS_NUMBER_NAN> (defined in perl.h). | |
827 | ||
828 | If the value of the number can fit in a UV, it is returned in C<*valuep>. | |
829 | C<IS_NUMBER_IN_UV> will be set to indicate that C<*valuep> is valid, C<IS_NUMBER_IN_UV> | |
830 | will never be set unless C<*valuep> is valid, but C<*valuep> may have been assigned | |
831 | to during processing even though C<IS_NUMBER_IN_UV> is not set on return. | |
832 | If C<valuep> is C<NULL>, C<IS_NUMBER_IN_UV> will be set for the same cases as when | |
833 | C<valuep> is non-C<NULL>, but no actual assignment (or SEGV) will occur. | |
834 | ||
835 | C<IS_NUMBER_NOT_INT> will be set with C<IS_NUMBER_IN_UV> if trailing decimals were | |
836 | seen (in which case C<*valuep> gives the true value truncated to an integer), and | |
837 | C<IS_NUMBER_NEG> if the number is negative (in which case C<*valuep> holds the | |
838 | absolute value). C<IS_NUMBER_IN_UV> is not set if e notation was used or the | |
13393a5e JH |
839 | number is larger than a UV. |
840 | ||
841 | C<flags> allows only C<PERL_SCAN_TRAILING>, which allows for trailing | |
842 | non-numeric text on an otherwise successful I<grok>, setting | |
843 | C<IS_NUMBER_TRAILING> on the result. | |
844 | ||
845 | =for apidoc grok_number | |
846 | ||
796b6530 | 847 | Identical to C<grok_number_flags()> with C<flags> set to zero. |
13393a5e JH |
848 | |
849 | =cut | |
850 | */ | |
851 | int | |
852 | Perl_grok_number(pTHX_ const char *pv, STRLEN len, UV *valuep) | |
853 | { | |
854 | PERL_ARGS_ASSERT_GROK_NUMBER; | |
855 | ||
856 | return grok_number_flags(pv, len, valuep, 0); | |
857 | } | |
858 | ||
945b524a JH |
859 | static const UV uv_max_div_10 = UV_MAX / 10; |
860 | static const U8 uv_max_mod_10 = UV_MAX % 10; | |
861 | ||
3f7602fa | 862 | int |
3823048b | 863 | Perl_grok_number_flags(pTHX_ const char *pv, STRLEN len, UV *valuep, U32 flags) |
3f7602fa | 864 | { |
60939fb8 | 865 | const char *s = pv; |
c4420975 | 866 | const char * const send = pv + len; |
ae776a2c | 867 | const char *d; |
60939fb8 | 868 | int numtype = 0; |
60939fb8 | 869 | |
3823048b | 870 | PERL_ARGS_ASSERT_GROK_NUMBER_FLAGS; |
7918f24d | 871 | |
60939fb8 NC |
872 | while (s < send && isSPACE(*s)) |
873 | s++; | |
874 | if (s == send) { | |
875 | return 0; | |
876 | } else if (*s == '-') { | |
877 | s++; | |
878 | numtype = IS_NUMBER_NEG; | |
879 | } | |
880 | else if (*s == '+') | |
aa42a541 | 881 | s++; |
60939fb8 NC |
882 | |
883 | if (s == send) | |
884 | return 0; | |
885 | ||
ae776a2c | 886 | /* The first digit (after optional sign): note that might |
8c12dc63 | 887 | * also point to "infinity" or "nan", or "1.#INF". */ |
ae776a2c JH |
888 | d = s; |
889 | ||
8c12dc63 | 890 | /* next must be digit or the radix separator or beginning of infinity/nan */ |
60939fb8 NC |
891 | if (isDIGIT(*s)) { |
892 | /* UVs are at least 32 bits, so the first 9 decimal digits cannot | |
893 | overflow. */ | |
894 | UV value = *s - '0'; | |
895 | /* This construction seems to be more optimiser friendly. | |
896 | (without it gcc does the isDIGIT test and the *s - '0' separately) | |
897 | With it gcc on arm is managing 6 instructions (6 cycles) per digit. | |
898 | In theory the optimiser could deduce how far to unroll the loop | |
899 | before checking for overflow. */ | |
58bb9ec3 NC |
900 | if (++s < send) { |
901 | int digit = *s - '0'; | |
60939fb8 NC |
902 | if (digit >= 0 && digit <= 9) { |
903 | value = value * 10 + digit; | |
58bb9ec3 NC |
904 | if (++s < send) { |
905 | digit = *s - '0'; | |
60939fb8 NC |
906 | if (digit >= 0 && digit <= 9) { |
907 | value = value * 10 + digit; | |
58bb9ec3 NC |
908 | if (++s < send) { |
909 | digit = *s - '0'; | |
60939fb8 NC |
910 | if (digit >= 0 && digit <= 9) { |
911 | value = value * 10 + digit; | |
58bb9ec3 NC |
912 | if (++s < send) { |
913 | digit = *s - '0'; | |
60939fb8 NC |
914 | if (digit >= 0 && digit <= 9) { |
915 | value = value * 10 + digit; | |
58bb9ec3 NC |
916 | if (++s < send) { |
917 | digit = *s - '0'; | |
60939fb8 NC |
918 | if (digit >= 0 && digit <= 9) { |
919 | value = value * 10 + digit; | |
58bb9ec3 NC |
920 | if (++s < send) { |
921 | digit = *s - '0'; | |
60939fb8 NC |
922 | if (digit >= 0 && digit <= 9) { |
923 | value = value * 10 + digit; | |
58bb9ec3 NC |
924 | if (++s < send) { |
925 | digit = *s - '0'; | |
60939fb8 NC |
926 | if (digit >= 0 && digit <= 9) { |
927 | value = value * 10 + digit; | |
58bb9ec3 NC |
928 | if (++s < send) { |
929 | digit = *s - '0'; | |
60939fb8 NC |
930 | if (digit >= 0 && digit <= 9) { |
931 | value = value * 10 + digit; | |
58bb9ec3 | 932 | if (++s < send) { |
60939fb8 NC |
933 | /* Now got 9 digits, so need to check |
934 | each time for overflow. */ | |
58bb9ec3 | 935 | digit = *s - '0'; |
60939fb8 | 936 | while (digit >= 0 && digit <= 9 |
945b524a JH |
937 | && (value < uv_max_div_10 |
938 | || (value == uv_max_div_10 | |
939 | && digit <= uv_max_mod_10))) { | |
60939fb8 | 940 | value = value * 10 + digit; |
58bb9ec3 NC |
941 | if (++s < send) |
942 | digit = *s - '0'; | |
60939fb8 NC |
943 | else |
944 | break; | |
945 | } | |
946 | if (digit >= 0 && digit <= 9 | |
51bd16da | 947 | && (s < send)) { |
60939fb8 NC |
948 | /* value overflowed. |
949 | skip the remaining digits, don't | |
950 | worry about setting *valuep. */ | |
951 | do { | |
952 | s++; | |
953 | } while (s < send && isDIGIT(*s)); | |
954 | numtype |= | |
955 | IS_NUMBER_GREATER_THAN_UV_MAX; | |
956 | goto skip_value; | |
957 | } | |
958 | } | |
959 | } | |
98994639 | 960 | } |
60939fb8 NC |
961 | } |
962 | } | |
963 | } | |
964 | } | |
965 | } | |
966 | } | |
967 | } | |
968 | } | |
969 | } | |
970 | } | |
971 | } | |
98994639 | 972 | } |
60939fb8 | 973 | } |
98994639 | 974 | } |
60939fb8 NC |
975 | numtype |= IS_NUMBER_IN_UV; |
976 | if (valuep) | |
977 | *valuep = value; | |
978 | ||
979 | skip_value: | |
980 | if (GROK_NUMERIC_RADIX(&s, send)) { | |
981 | numtype |= IS_NUMBER_NOT_INT; | |
982 | while (s < send && isDIGIT(*s)) /* optional digits after the radix */ | |
983 | s++; | |
98994639 | 984 | } |
60939fb8 NC |
985 | } |
986 | else if (GROK_NUMERIC_RADIX(&s, send)) { | |
987 | numtype |= IS_NUMBER_NOT_INT | IS_NUMBER_IN_UV; /* valuep assigned below */ | |
988 | /* no digits before the radix means we need digits after it */ | |
989 | if (s < send && isDIGIT(*s)) { | |
990 | do { | |
991 | s++; | |
992 | } while (s < send && isDIGIT(*s)); | |
993 | if (valuep) { | |
994 | /* integer approximation is valid - it's 0. */ | |
995 | *valuep = 0; | |
996 | } | |
98994639 | 997 | } |
60939fb8 | 998 | else |
ae776a2c | 999 | return 0; |
ff4eb398 | 1000 | } |
60939fb8 | 1001 | |
926f5fc6 | 1002 | if (s > d && s < send) { |
60939fb8 | 1003 | /* we can have an optional exponent part */ |
305b8651 | 1004 | if (isALPHA_FOLD_EQ(*s, 'e')) { |
60939fb8 NC |
1005 | s++; |
1006 | if (s < send && (*s == '-' || *s == '+')) | |
1007 | s++; | |
1008 | if (s < send && isDIGIT(*s)) { | |
1009 | do { | |
1010 | s++; | |
1011 | } while (s < send && isDIGIT(*s)); | |
1012 | } | |
3f7602fa TC |
1013 | else if (flags & PERL_SCAN_TRAILING) |
1014 | return numtype | IS_NUMBER_TRAILING; | |
60939fb8 | 1015 | else |
3f7602fa TC |
1016 | return 0; |
1017 | ||
1018 | /* The only flag we keep is sign. Blow away any "it's UV" */ | |
1019 | numtype &= IS_NUMBER_NEG; | |
1020 | numtype |= IS_NUMBER_NOT_INT; | |
60939fb8 NC |
1021 | } |
1022 | } | |
1023 | while (s < send && isSPACE(*s)) | |
1024 | s++; | |
1025 | if (s >= send) | |
aa8b85de | 1026 | return numtype; |
b59bf0b2 | 1027 | if (memEQs(pv, len, "0 but true")) { |
60939fb8 NC |
1028 | if (valuep) |
1029 | *valuep = 0; | |
1030 | return IS_NUMBER_IN_UV; | |
1031 | } | |
8c12dc63 JH |
1032 | /* We could be e.g. at "Inf" or "NaN", or at the "#" of "1.#INF". */ |
1033 | if ((s + 2 < send) && strchr("inqs#", toFOLD(*s))) { | |
1034 | /* Really detect inf/nan. Start at d, not s, since the above | |
1035 | * code might have already consumed the "1." or "1". */ | |
7eff3d39 | 1036 | const int infnan = Perl_grok_infnan(aTHX_ &d, send); |
8c12dc63 JH |
1037 | if ((infnan & IS_NUMBER_INFINITY)) { |
1038 | return (numtype | infnan); /* Keep sign for infinity. */ | |
1039 | } | |
1040 | else if ((infnan & IS_NUMBER_NAN)) { | |
1041 | return (numtype | infnan) & ~IS_NUMBER_NEG; /* Clear sign for nan. */ | |
1042 | } | |
1043 | } | |
3f7602fa TC |
1044 | else if (flags & PERL_SCAN_TRAILING) { |
1045 | return numtype | IS_NUMBER_TRAILING; | |
1046 | } | |
1047 | ||
60939fb8 | 1048 | return 0; |
98994639 HS |
1049 | } |
1050 | ||
6313e544 | 1051 | /* |
5d4a52b5 | 1052 | =for apidoc grok_atoUV |
6313e544 | 1053 | |
5d4a52b5 | 1054 | parse a string, looking for a decimal unsigned integer. |
338aa8b0 | 1055 | |
5d4a52b5 KW |
1056 | On entry, C<pv> points to the beginning of the string; |
1057 | C<valptr> points to a UV that will receive the converted value, if found; | |
1058 | C<endptr> is either NULL or points to a variable that points to one byte | |
1059 | beyond the point in C<pv> that this routine should examine. | |
1060 | If C<endptr> is NULL, C<pv> is assumed to be NUL-terminated. | |
f4379102 | 1061 | |
5d4a52b5 KW |
1062 | Returns FALSE if C<pv> doesn't represent a valid unsigned integer value (with |
1063 | no leading zeros). Otherwise it returns TRUE, and sets C<*valptr> to that | |
1064 | value. | |
6313e544 | 1065 | |
5d4a52b5 KW |
1066 | If you constrain the portion of C<pv> that is looked at by this function (by |
1067 | passing a non-NULL C<endptr>), and if the intial bytes of that portion form a | |
1068 | valid value, it will return TRUE, setting C<*endptr> to the byte following the | |
1069 | final digit of the value. But if there is no constraint at what's looked at, | |
1070 | all of C<pv> must be valid in order for TRUE to be returned. | |
6313e544 | 1071 | |
5d4a52b5 | 1072 | The only characters this accepts are the decimal digits '0'..'9'. |
338aa8b0 | 1073 | |
5d4a52b5 KW |
1074 | As opposed to L<atoi(3)> or L<strtol(3)>, C<grok_atoUV> does NOT allow optional |
1075 | leading whitespace, nor negative inputs. If such features are required, the | |
1076 | calling code needs to explicitly implement those. | |
6313e544 | 1077 | |
5d4a52b5 KW |
1078 | Note that this function returns FALSE for inputs that would overflow a UV, |
1079 | or have leading zeros. Thus a single C<0> is accepted, but not C<00> nor | |
1080 | C<01>, C<002>, I<etc>. | |
1081 | ||
1082 | Background: C<atoi> has severe problems with illegal inputs, it cannot be | |
d62b8c6a | 1083 | used for incremental parsing, and therefore should be avoided |
5d4a52b5 | 1084 | C<atoi> and C<strtol> are also affected by locale settings, which can also be |
d62b8c6a JH |
1085 | seen as a bug (global state controlled by user environment). |
1086 | ||
238217e5 JK |
1087 | =cut |
1088 | ||
6313e544 JH |
1089 | */ |
1090 | ||
22ff3130 HS |
1091 | bool |
1092 | Perl_grok_atoUV(const char *pv, UV *valptr, const char** endptr) | |
6313e544 JH |
1093 | { |
1094 | const char* s = pv; | |
1095 | const char** eptr; | |
1096 | const char* end2; /* Used in case endptr is NULL. */ | |
22ff3130 | 1097 | UV val = 0; /* The parsed value. */ |
6313e544 | 1098 | |
22ff3130 | 1099 | PERL_ARGS_ASSERT_GROK_ATOUV; |
6313e544 | 1100 | |
5d4a52b5 KW |
1101 | if (endptr) { |
1102 | eptr = endptr; | |
1103 | } | |
1104 | else { | |
1105 | end2 = s + strlen(s); | |
1106 | eptr = &end2; | |
1107 | } | |
1108 | ||
1109 | if ( *eptr <= s | |
1110 | || ! isDIGIT(*s)) | |
1111 | { | |
1112 | return FALSE; | |
1113 | } | |
1114 | ||
97d95d46 KW |
1115 | /* Single-digit inputs are quite common. */ |
1116 | val = *s++ - '0'; | |
1117 | if (s < *eptr && isDIGIT(*s)) { | |
1118 | /* Fail on extra leading zeros. */ | |
1119 | if (val == 0) | |
1120 | return FALSE; | |
1121 | while (s < *eptr && isDIGIT(*s)) { | |
1122 | /* This could be unrolled like in grok_number(), but | |
1123 | * the expected uses of this are not speed-needy, and | |
1124 | * unlikely to need full 64-bitness. */ | |
1125 | const U8 digit = *s++ - '0'; | |
1126 | if (val < uv_max_div_10 || | |
1127 | (val == uv_max_div_10 && digit <= uv_max_mod_10)) { | |
1128 | val = val * 10 + digit; | |
1129 | } else { | |
22ff3130 | 1130 | return FALSE; |
6313e544 JH |
1131 | } |
1132 | } | |
97d95d46 KW |
1133 | } |
1134 | ||
5d4a52b5 KW |
1135 | if (endptr == NULL) { |
1136 | if (*s) { | |
1137 | return FALSE; /* If endptr is NULL, no trailing non-digits allowed. */ | |
1138 | } | |
1139 | } | |
1140 | else { | |
1141 | *endptr = s; | |
75feedba | 1142 | } |
97d95d46 | 1143 | |
22ff3130 HS |
1144 | *valptr = val; |
1145 | return TRUE; | |
6313e544 JH |
1146 | } |
1147 | ||
ce6f496d | 1148 | #ifndef Perl_strtod |
4801ca72 | 1149 | STATIC NV |
98994639 HS |
1150 | S_mulexp10(NV value, I32 exponent) |
1151 | { | |
1152 | NV result = 1.0; | |
1153 | NV power = 10.0; | |
1154 | bool negative = 0; | |
1155 | I32 bit; | |
1156 | ||
1157 | if (exponent == 0) | |
1158 | return value; | |
659c4b96 DM |
1159 | if (value == 0) |
1160 | return (NV)0; | |
87032ba1 | 1161 | |
24866caa | 1162 | /* On OpenVMS VAX we by default use the D_FLOAT double format, |
67597c89 | 1163 | * and that format does not have *easy* capabilities [1] for |
19c1206d KW |
1164 | * overflowing doubles 'silently' as IEEE fp does. We also need |
1165 | * to support G_FLOAT on both VAX and Alpha, and though the exponent | |
1166 | * range is much larger than D_FLOAT it still doesn't do silent | |
1167 | * overflow. Therefore we need to detect early whether we would | |
1168 | * overflow (this is the behaviour of the native string-to-float | |
24866caa | 1169 | * conversion routines, and therefore of native applications, too). |
67597c89 | 1170 | * |
24866caa CB |
1171 | * [1] Trying to establish a condition handler to trap floating point |
1172 | * exceptions is not a good idea. */ | |
87032ba1 JH |
1173 | |
1174 | /* In UNICOS and in certain Cray models (such as T90) there is no | |
1175 | * IEEE fp, and no way at all from C to catch fp overflows gracefully. | |
1176 | * There is something you can do if you are willing to use some | |
1177 | * inline assembler: the instruction is called DFI-- but that will | |
1178 | * disable *all* floating point interrupts, a little bit too large | |
1179 | * a hammer. Therefore we need to catch potential overflows before | |
1180 | * it's too late. */ | |
353813d9 | 1181 | |
a7157111 | 1182 | #if ((defined(VMS) && !defined(_IEEE_FP)) || defined(_UNICOS) || defined(DOUBLE_IS_VAX_FLOAT)) && defined(NV_MAX_10_EXP) |
353813d9 | 1183 | STMT_START { |
c4420975 | 1184 | const NV exp_v = log10(value); |
353813d9 HS |
1185 | if (exponent >= NV_MAX_10_EXP || exponent + exp_v >= NV_MAX_10_EXP) |
1186 | return NV_MAX; | |
1187 | if (exponent < 0) { | |
1188 | if (-(exponent + exp_v) >= NV_MAX_10_EXP) | |
1189 | return 0.0; | |
1190 | while (-exponent >= NV_MAX_10_EXP) { | |
1191 | /* combination does not overflow, but 10^(-exponent) does */ | |
1192 | value /= 10; | |
1193 | ++exponent; | |
1194 | } | |
1195 | } | |
1196 | } STMT_END; | |
87032ba1 JH |
1197 | #endif |
1198 | ||
353813d9 HS |
1199 | if (exponent < 0) { |
1200 | negative = 1; | |
1201 | exponent = -exponent; | |
b27804d8 DM |
1202 | #ifdef NV_MAX_10_EXP |
1203 | /* for something like 1234 x 10^-309, the action of calculating | |
1204 | * the intermediate value 10^309 then returning 1234 / (10^309) | |
1205 | * will fail, since 10^309 becomes infinity. In this case try to | |
1206 | * refactor it as 123 / (10^308) etc. | |
1207 | */ | |
1208 | while (value && exponent > NV_MAX_10_EXP) { | |
1209 | exponent--; | |
1210 | value /= 10; | |
1211 | } | |
48853916 JH |
1212 | if (value == 0.0) |
1213 | return value; | |
b27804d8 | 1214 | #endif |
353813d9 | 1215 | } |
c62e754c JH |
1216 | #if defined(__osf__) |
1217 | /* Even with cc -ieee + ieee_set_fp_control(IEEE_TRAP_ENABLE_INV) | |
1218 | * Tru64 fp behavior on inf/nan is somewhat broken. Another way | |
1219 | * to do this would be ieee_set_fp_control(IEEE_TRAP_ENABLE_OVF) | |
1220 | * but that breaks another set of infnan.t tests. */ | |
1221 | # define FP_OVERFLOWS_TO_ZERO | |
1222 | #endif | |
98994639 HS |
1223 | for (bit = 1; exponent; bit <<= 1) { |
1224 | if (exponent & bit) { | |
1225 | exponent ^= bit; | |
1226 | result *= power; | |
c62e754c JH |
1227 | #ifdef FP_OVERFLOWS_TO_ZERO |
1228 | if (result == 0) | |
a7157111 | 1229 | # ifdef NV_INF |
c62e754c | 1230 | return value < 0 ? -NV_INF : NV_INF; |
a7157111 JH |
1231 | # else |
1232 | return value < 0 ? -FLT_MAX : FLT_MAX; | |
1233 | # endif | |
c62e754c | 1234 | #endif |
236f0012 | 1235 | /* Floating point exceptions are supposed to be turned off, |
19c1206d | 1236 | * but if we're obviously done, don't risk another iteration. |
236f0012 CB |
1237 | */ |
1238 | if (exponent == 0) break; | |
98994639 HS |
1239 | } |
1240 | power *= power; | |
1241 | } | |
1242 | return negative ? value / result : value * result; | |
1243 | } | |
ce6f496d | 1244 | #endif /* #ifndef Perl_strtod */ |
98994639 | 1245 | |
ce6f496d | 1246 | #ifdef Perl_strtod |
b93d1309 | 1247 | # define ATOF(s, x) my_atof2(s, &x) |
f7b64c80 | 1248 | #else |
b93d1309 | 1249 | # define ATOF(s, x) Perl_atof2(s, x) |
f7b64c80 | 1250 | #endif |
b93d1309 | 1251 | |
98994639 HS |
1252 | NV |
1253 | Perl_my_atof(pTHX_ const char* s) | |
1254 | { | |
f720c878 KW |
1255 | /* 's' must be NUL terminated */ |
1256 | ||
98994639 | 1257 | NV x = 0.0; |
9eda1ea6 KW |
1258 | |
1259 | PERL_ARGS_ASSERT_MY_ATOF; | |
1260 | ||
b93d1309 | 1261 | #if ! defined(USE_LOCALE_NUMERIC) |
9eda1ea6 | 1262 | |
b93d1309 | 1263 | ATOF(s, x); |
9eda1ea6 KW |
1264 | |
1265 | #else | |
7918f24d | 1266 | |
a2287a13 | 1267 | { |
67d796ae KW |
1268 | DECLARATION_FOR_LC_NUMERIC_MANIPULATION; |
1269 | STORE_LC_NUMERIC_SET_TO_NEEDED(); | |
fdf55d20 KW |
1270 | if (! (PL_numeric_radix_sv && IN_LC(LC_NUMERIC))) { |
1271 | ATOF(s,x); | |
1272 | } | |
1273 | else { | |
19c1206d | 1274 | |
e4850248 KW |
1275 | /* Look through the string for the first thing that looks like a |
1276 | * decimal point: either the value in the current locale or the | |
1277 | * standard fallback of '.'. The one which appears earliest in the | |
1278 | * input string is the one that we should have atof look for. Note | |
1279 | * that we have to determine this beforehand because on some | |
1280 | * systems, Perl_atof2 is just a wrapper around the system's atof. | |
1281 | * */ | |
1ae85f6c KW |
1282 | const char * const standard_pos = strchr(s, '.'); |
1283 | const char * const local_pos | |
1284 | = strstr(s, SvPV_nolen(PL_numeric_radix_sv)); | |
1285 | const bool use_standard_radix | |
1286 | = standard_pos && (!local_pos || standard_pos < local_pos); | |
78787052 | 1287 | |
665873e9 | 1288 | if (use_standard_radix) { |
e4850248 | 1289 | SET_NUMERIC_STANDARD(); |
665873e9 KW |
1290 | LOCK_LC_NUMERIC_STANDARD(); |
1291 | } | |
78787052 | 1292 | |
b93d1309 | 1293 | ATOF(s,x); |
78787052 | 1294 | |
665873e9 KW |
1295 | if (use_standard_radix) { |
1296 | UNLOCK_LC_NUMERIC_STANDARD(); | |
67d796ae | 1297 | SET_NUMERIC_UNDERLYING(); |
665873e9 | 1298 | } |
e4850248 | 1299 | } |
a2287a13 KW |
1300 | RESTORE_LC_NUMERIC(); |
1301 | } | |
9eda1ea6 | 1302 | |
98994639 | 1303 | #endif |
9eda1ea6 | 1304 | |
98994639 HS |
1305 | return x; |
1306 | } | |
1307 | ||
a7157111 | 1308 | #if defined(NV_INF) || defined(NV_NAN) |
3c81f0b3 DD |
1309 | |
1310 | #ifdef USING_MSVC6 | |
1311 | # pragma warning(push) | |
1312 | # pragma warning(disable:4756;disable:4056) | |
1313 | #endif | |
829757a4 | 1314 | static char* |
5563f457 | 1315 | S_my_atof_infnan(pTHX_ const char* s, bool negative, const char* send, NV* value) |
829757a4 JH |
1316 | { |
1317 | const char *p0 = negative ? s - 1 : s; | |
1318 | const char *p = p0; | |
7eff3d39 | 1319 | const int infnan = grok_infnan(&p, send); |
829757a4 JH |
1320 | if (infnan && p != p0) { |
1321 | /* If we can generate inf/nan directly, let's do so. */ | |
1322 | #ifdef NV_INF | |
1323 | if ((infnan & IS_NUMBER_INFINITY)) { | |
3823048b | 1324 | *value = (infnan & IS_NUMBER_NEG) ? -NV_INF: NV_INF; |
829757a4 JH |
1325 | return (char*)p; |
1326 | } | |
1327 | #endif | |
1328 | #ifdef NV_NAN | |
1329 | if ((infnan & IS_NUMBER_NAN)) { | |
3823048b | 1330 | *value = NV_NAN; |
829757a4 JH |
1331 | return (char*)p; |
1332 | } | |
1333 | #endif | |
1334 | #ifdef Perl_strtod | |
68611e6f | 1335 | /* If still here, we didn't have either NV_INF or NV_NAN, |
829757a4 JH |
1336 | * and can try falling back to native strtod/strtold. |
1337 | * | |
1338 | * The native interface might not recognize all the possible | |
1339 | * inf/nan strings Perl recognizes. What we can try | |
1340 | * is to try faking the input. We will try inf/-inf/nan | |
1341 | * as the most promising/portable input. */ | |
1342 | { | |
6d37e916 | 1343 | const char* fake = "silence compiler warning"; |
829757a4 JH |
1344 | char* endp; |
1345 | NV nv; | |
a7157111 | 1346 | #ifdef NV_INF |
829757a4 JH |
1347 | if ((infnan & IS_NUMBER_INFINITY)) { |
1348 | fake = ((infnan & IS_NUMBER_NEG)) ? "-inf" : "inf"; | |
1349 | } | |
a7157111 JH |
1350 | #endif |
1351 | #ifdef NV_NAN | |
1352 | if ((infnan & IS_NUMBER_NAN)) { | |
829757a4 JH |
1353 | fake = "nan"; |
1354 | } | |
a7157111 | 1355 | #endif |
6d37e916 | 1356 | assert(strNE(fake, "silence compiler warning")); |
829757a4 JH |
1357 | nv = Perl_strtod(fake, &endp); |
1358 | if (fake != endp) { | |
a7157111 | 1359 | #ifdef NV_INF |
829757a4 | 1360 | if ((infnan & IS_NUMBER_INFINITY)) { |
a7157111 | 1361 | # ifdef Perl_isinf |
829757a4 JH |
1362 | if (Perl_isinf(nv)) |
1363 | *value = nv; | |
a7157111 | 1364 | # else |
829757a4 JH |
1365 | /* last resort, may generate SIGFPE */ |
1366 | *value = Perl_exp((NV)1e9); | |
1367 | if ((infnan & IS_NUMBER_NEG)) | |
1368 | *value = -*value; | |
a7157111 | 1369 | # endif |
829757a4 JH |
1370 | return (char*)p; /* p, not endp */ |
1371 | } | |
a7157111 JH |
1372 | #endif |
1373 | #ifdef NV_NAN | |
1374 | if ((infnan & IS_NUMBER_NAN)) { | |
1375 | # ifdef Perl_isnan | |
829757a4 JH |
1376 | if (Perl_isnan(nv)) |
1377 | *value = nv; | |
a7157111 | 1378 | # else |
829757a4 JH |
1379 | /* last resort, may generate SIGFPE */ |
1380 | *value = Perl_log((NV)-1.0); | |
a7157111 | 1381 | # endif |
829757a4 | 1382 | return (char*)p; /* p, not endp */ |
a7157111 | 1383 | #endif |
829757a4 JH |
1384 | } |
1385 | } | |
1386 | } | |
1387 | #endif /* #ifdef Perl_strtod */ | |
1388 | } | |
1389 | return NULL; | |
1390 | } | |
3c81f0b3 DD |
1391 | #ifdef USING_MSVC6 |
1392 | # pragma warning(pop) | |
1393 | #endif | |
829757a4 | 1394 | |
a7157111 JH |
1395 | #endif /* if defined(NV_INF) || defined(NV_NAN) */ |
1396 | ||
98994639 HS |
1397 | char* |
1398 | Perl_my_atof2(pTHX_ const char* orig, NV* value) | |
1399 | { | |
6928bedc KW |
1400 | PERL_ARGS_ASSERT_MY_ATOF2; |
1401 | return my_atof3(orig, value, 0); | |
1402 | } | |
1403 | ||
1404 | char* | |
1405 | Perl_my_atof3(pTHX_ const char* orig, NV* value, STRLEN len) | |
1406 | { | |
e1ec3a88 | 1407 | const char* s = orig; |
a4eca1d4 | 1408 | NV result[3] = {0.0, 0.0, 0.0}; |
ce6f496d | 1409 | #if defined(USE_PERL_ATOF) || defined(Perl_strtod) |
6928bedc KW |
1410 | const char* send = s + ((len != 0) |
1411 | ? len | |
1412 | : strlen(orig)); /* one past the last */ | |
a4eca1d4 JH |
1413 | bool negative = 0; |
1414 | #endif | |
ce6f496d | 1415 | #if defined(USE_PERL_ATOF) && !defined(Perl_strtod) |
a4eca1d4 | 1416 | UV accumulator[2] = {0,0}; /* before/after dp */ |
8194bf88 | 1417 | bool seen_digit = 0; |
20f6aaab AS |
1418 | I32 exp_adjust[2] = {0,0}; |
1419 | I32 exp_acc[2] = {-1, -1}; | |
1420 | /* the current exponent adjust for the accumulators */ | |
98994639 | 1421 | I32 exponent = 0; |
8194bf88 | 1422 | I32 seen_dp = 0; |
20f6aaab AS |
1423 | I32 digit = 0; |
1424 | I32 old_digit = 0; | |
8194bf88 | 1425 | I32 sig_digits = 0; /* noof significant digits seen so far */ |
a4eca1d4 | 1426 | #endif |
8194bf88 | 1427 | |
ce6f496d | 1428 | #if defined(USE_PERL_ATOF) || defined(Perl_strtod) |
6928bedc | 1429 | PERL_ARGS_ASSERT_MY_ATOF3; |
7918f24d | 1430 | |
a4eca1d4 | 1431 | /* leading whitespace */ |
6928bedc | 1432 | while (s < send && isSPACE(*s)) |
a4eca1d4 JH |
1433 | ++s; |
1434 | ||
1435 | /* sign */ | |
1436 | switch (*s) { | |
1437 | case '-': | |
1438 | negative = 1; | |
1439 | /* FALLTHROUGH */ | |
1440 | case '+': | |
1441 | ++s; | |
1442 | } | |
1443 | #endif | |
1444 | ||
ce6f496d | 1445 | #ifdef Perl_strtod |
a4eca1d4 JH |
1446 | { |
1447 | char* endp; | |
d94e901a KW |
1448 | char* copy = NULL; |
1449 | ||
adc55e02 | 1450 | if ((endp = S_my_atof_infnan(aTHX_ s, negative, send, value))) |
a4eca1d4 | 1451 | return endp; |
d94e901a KW |
1452 | |
1453 | /* If the length is passed in, the input string isn't NUL-terminated, | |
1454 | * and in it turns out the function below assumes it is; therefore we | |
1455 | * create a copy and NUL-terminate that */ | |
1456 | if (len) { | |
1457 | Newx(copy, len + 1, char); | |
1458 | Copy(orig, copy, len, char); | |
1459 | copy[len] = '\0'; | |
1460 | s = copy + (s - orig); | |
1461 | } | |
1462 | ||
ce6f496d | 1463 | result[2] = Perl_strtod(s, &endp); |
d94e901a KW |
1464 | |
1465 | /* If we created a copy, 'endp' is in terms of that. Convert back to | |
1466 | * the original */ | |
1467 | if (copy) { | |
1468 | endp = (endp - copy) + (char *) orig; | |
1469 | Safefree(copy); | |
1470 | } | |
1471 | ||
a4eca1d4 JH |
1472 | if (s != endp) { |
1473 | *value = negative ? -result[2] : result[2]; | |
1474 | return endp; | |
1475 | } | |
1476 | return NULL; | |
1477 | } | |
1478 | #elif defined(USE_PERL_ATOF) | |
1479 | ||
8194bf88 DM |
1480 | /* There is no point in processing more significant digits |
1481 | * than the NV can hold. Note that NV_DIG is a lower-bound value, | |
1482 | * while we need an upper-bound value. We add 2 to account for this; | |
1483 | * since it will have been conservative on both the first and last digit. | |
1484 | * For example a 32-bit mantissa with an exponent of 4 would have | |
1485 | * exact values in the set | |
1486 | * 4 | |
1487 | * 8 | |
1488 | * .. | |
1489 | * 17179869172 | |
1490 | * 17179869176 | |
1491 | * 17179869180 | |
1492 | * | |
1493 | * where for the purposes of calculating NV_DIG we would have to discount | |
1494 | * both the first and last digit, since neither can hold all values from | |
1495 | * 0..9; but for calculating the value we must examine those two digits. | |
1496 | */ | |
ffa277e5 AS |
1497 | #ifdef MAX_SIG_DIG_PLUS |
1498 | /* It is not necessarily the case that adding 2 to NV_DIG gets all the | |
1499 | possible digits in a NV, especially if NVs are not IEEE compliant | |
1500 | (e.g., long doubles on IRIX) - Allen <allens@cpan.org> */ | |
1501 | # define MAX_SIG_DIGITS (NV_DIG+MAX_SIG_DIG_PLUS) | |
1502 | #else | |
1503 | # define MAX_SIG_DIGITS (NV_DIG+2) | |
1504 | #endif | |
8194bf88 DM |
1505 | |
1506 | /* the max number we can accumulate in a UV, and still safely do 10*N+9 */ | |
1507 | #define MAX_ACCUMULATE ( (UV) ((UV_MAX - 9)/10)) | |
98994639 | 1508 | |
a5dc2484 | 1509 | #if defined(NV_INF) || defined(NV_NAN) |
ae776a2c | 1510 | { |
7eff3d39 | 1511 | char* endp; |
5563f457 | 1512 | if ((endp = S_my_atof_infnan(aTHX_ s, negative, send, value))) |
7eff3d39 | 1513 | return endp; |
ae776a2c | 1514 | } |
a5dc2484 | 1515 | #endif |
2b54f59f | 1516 | |
8194bf88 DM |
1517 | /* we accumulate digits into an integer; when this becomes too |
1518 | * large, we add the total to NV and start again */ | |
98994639 | 1519 | |
6928bedc | 1520 | while (s < send) { |
8194bf88 DM |
1521 | if (isDIGIT(*s)) { |
1522 | seen_digit = 1; | |
20f6aaab | 1523 | old_digit = digit; |
8194bf88 | 1524 | digit = *s++ - '0'; |
20f6aaab AS |
1525 | if (seen_dp) |
1526 | exp_adjust[1]++; | |
98994639 | 1527 | |
8194bf88 DM |
1528 | /* don't start counting until we see the first significant |
1529 | * digit, eg the 5 in 0.00005... */ | |
1530 | if (!sig_digits && digit == 0) | |
1531 | continue; | |
1532 | ||
1533 | if (++sig_digits > MAX_SIG_DIGITS) { | |
98994639 | 1534 | /* limits of precision reached */ |
20f6aaab AS |
1535 | if (digit > 5) { |
1536 | ++accumulator[seen_dp]; | |
1537 | } else if (digit == 5) { | |
1538 | if (old_digit % 2) { /* round to even - Allen */ | |
1539 | ++accumulator[seen_dp]; | |
1540 | } | |
1541 | } | |
1542 | if (seen_dp) { | |
1543 | exp_adjust[1]--; | |
1544 | } else { | |
1545 | exp_adjust[0]++; | |
1546 | } | |
8194bf88 | 1547 | /* skip remaining digits */ |
6928bedc | 1548 | while (s < send && isDIGIT(*s)) { |
98994639 | 1549 | ++s; |
20f6aaab AS |
1550 | if (! seen_dp) { |
1551 | exp_adjust[0]++; | |
1552 | } | |
98994639 HS |
1553 | } |
1554 | /* warn of loss of precision? */ | |
98994639 | 1555 | } |
8194bf88 | 1556 | else { |
20f6aaab | 1557 | if (accumulator[seen_dp] > MAX_ACCUMULATE) { |
8194bf88 | 1558 | /* add accumulator to result and start again */ |
20f6aaab AS |
1559 | result[seen_dp] = S_mulexp10(result[seen_dp], |
1560 | exp_acc[seen_dp]) | |
1561 | + (NV)accumulator[seen_dp]; | |
1562 | accumulator[seen_dp] = 0; | |
1563 | exp_acc[seen_dp] = 0; | |
98994639 | 1564 | } |
20f6aaab AS |
1565 | accumulator[seen_dp] = accumulator[seen_dp] * 10 + digit; |
1566 | ++exp_acc[seen_dp]; | |
98994639 | 1567 | } |
8194bf88 | 1568 | } |
e1ec3a88 | 1569 | else if (!seen_dp && GROK_NUMERIC_RADIX(&s, send)) { |
8194bf88 | 1570 | seen_dp = 1; |
20f6aaab | 1571 | if (sig_digits > MAX_SIG_DIGITS) { |
6928bedc | 1572 | while (s < send && isDIGIT(*s)) { |
20f6aaab | 1573 | ++s; |
9604fbf0 | 1574 | } |
20f6aaab AS |
1575 | break; |
1576 | } | |
8194bf88 DM |
1577 | } |
1578 | else { | |
1579 | break; | |
98994639 HS |
1580 | } |
1581 | } | |
1582 | ||
20f6aaab AS |
1583 | result[0] = S_mulexp10(result[0], exp_acc[0]) + (NV)accumulator[0]; |
1584 | if (seen_dp) { | |
1585 | result[1] = S_mulexp10(result[1], exp_acc[1]) + (NV)accumulator[1]; | |
1586 | } | |
98994639 | 1587 | |
6928bedc | 1588 | if (s < send && seen_digit && (isALPHA_FOLD_EQ(*s, 'e'))) { |
98994639 HS |
1589 | bool expnegative = 0; |
1590 | ||
1591 | ++s; | |
1592 | switch (*s) { | |
1593 | case '-': | |
1594 | expnegative = 1; | |
924ba076 | 1595 | /* FALLTHROUGH */ |
98994639 HS |
1596 | case '+': |
1597 | ++s; | |
1598 | } | |
6928bedc | 1599 | while (s < send && isDIGIT(*s)) |
98994639 HS |
1600 | exponent = exponent * 10 + (*s++ - '0'); |
1601 | if (expnegative) | |
1602 | exponent = -exponent; | |
1603 | } | |
1604 | ||
1605 | /* now apply the exponent */ | |
20f6aaab AS |
1606 | |
1607 | if (seen_dp) { | |
1608 | result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]) | |
1609 | + S_mulexp10(result[1],exponent-exp_adjust[1]); | |
1610 | } else { | |
1611 | result[2] = S_mulexp10(result[0],exponent+exp_adjust[0]); | |
1612 | } | |
98994639 HS |
1613 | |
1614 | /* now apply the sign */ | |
1615 | if (negative) | |
20f6aaab | 1616 | result[2] = -result[2]; |
a36244b7 | 1617 | #endif /* USE_PERL_ATOF */ |
20f6aaab | 1618 | *value = result[2]; |
73d840c0 | 1619 | return (char *)s; |
98994639 HS |
1620 | } |
1621 | ||
5d34af89 | 1622 | /* |
3d9d9213 | 1623 | =for apidoc isinfnan |
5d34af89 | 1624 | |
796b6530 KW |
1625 | C<Perl_isinfnan()> is utility function that returns true if the NV |
1626 | argument is either an infinity or a C<NaN>, false otherwise. To test | |
1627 | in more detail, use C<Perl_isinf()> and C<Perl_isnan()>. | |
5d34af89 | 1628 | |
68611e6f JH |
1629 | This is also the logical inverse of Perl_isfinite(). |
1630 | ||
5d34af89 JH |
1631 | =cut |
1632 | */ | |
1cd88304 JH |
1633 | bool |
1634 | Perl_isinfnan(NV nv) | |
1635 | { | |
a5dc2484 | 1636 | PERL_UNUSED_ARG(nv); |
1cd88304 JH |
1637 | #ifdef Perl_isinf |
1638 | if (Perl_isinf(nv)) | |
1639 | return TRUE; | |
1640 | #endif | |
1641 | #ifdef Perl_isnan | |
1642 | if (Perl_isnan(nv)) | |
1643 | return TRUE; | |
1644 | #endif | |
1645 | return FALSE; | |
1646 | } | |
1647 | ||
354b74ae FC |
1648 | /* |
1649 | =for apidoc | |
1650 | ||
796b6530 | 1651 | Checks whether the argument would be either an infinity or C<NaN> when used |
354b74ae | 1652 | as a number, but is careful not to trigger non-numeric or uninitialized |
796b6530 | 1653 | warnings. it assumes the caller has done C<SvGETMAGIC(sv)> already. |
354b74ae FC |
1654 | |
1655 | =cut | |
1656 | */ | |
1657 | ||
1658 | bool | |
1659 | Perl_isinfnansv(pTHX_ SV *sv) | |
1660 | { | |
1661 | PERL_ARGS_ASSERT_ISINFNANSV; | |
1662 | if (!SvOK(sv)) | |
1663 | return FALSE; | |
1664 | if (SvNOKp(sv)) | |
1665 | return Perl_isinfnan(SvNVX(sv)); | |
1666 | if (SvIOKp(sv)) | |
1667 | return FALSE; | |
1668 | { | |
1669 | STRLEN len; | |
1670 | const char *s = SvPV_nomg_const(sv, len); | |
3823048b | 1671 | return cBOOL(grok_infnan(&s, s+len)); |
354b74ae FC |
1672 | } |
1673 | } | |
1674 | ||
d67dac15 | 1675 | #ifndef HAS_MODFL |
68611e6f JH |
1676 | /* C99 has truncl, pre-C99 Solaris had aintl. We can use either with |
1677 | * copysignl to emulate modfl, which is in some platforms missing or | |
1678 | * broken. */ | |
d67dac15 JH |
1679 | # if defined(HAS_TRUNCL) && defined(HAS_COPYSIGNL) |
1680 | long double | |
1681 | Perl_my_modfl(long double x, long double *ip) | |
1682 | { | |
68611e6f JH |
1683 | *ip = truncl(x); |
1684 | return (x == *ip ? copysignl(0.0L, x) : x - *ip); | |
d67dac15 JH |
1685 | } |
1686 | # elif defined(HAS_AINTL) && defined(HAS_COPYSIGNL) | |
55954f19 JH |
1687 | long double |
1688 | Perl_my_modfl(long double x, long double *ip) | |
1689 | { | |
68611e6f JH |
1690 | *ip = aintl(x); |
1691 | return (x == *ip ? copysignl(0.0L, x) : x - *ip); | |
55954f19 | 1692 | } |
d67dac15 | 1693 | # endif |
55954f19 JH |
1694 | #endif |
1695 | ||
7b9b7dff | 1696 | /* Similarly, with ilogbl and scalbnl we can emulate frexpl. */ |
55954f19 JH |
1697 | #if ! defined(HAS_FREXPL) && defined(HAS_ILOGBL) && defined(HAS_SCALBNL) |
1698 | long double | |
1699 | Perl_my_frexpl(long double x, int *e) { | |
68611e6f JH |
1700 | *e = x == 0.0L ? 0 : ilogbl(x) + 1; |
1701 | return (scalbnl(x, -*e)); | |
55954f19 JH |
1702 | } |
1703 | #endif | |
66610fdd RGS |
1704 | |
1705 | /* | |
ed140128 AD |
1706 | =for apidoc Perl_signbit |
1707 | ||
1708 | Return a non-zero integer if the sign bit on an NV is set, and 0 if | |
19c1206d | 1709 | it is not. |
ed140128 | 1710 | |
796b6530 KW |
1711 | If F<Configure> detects this system has a C<signbit()> that will work with |
1712 | our NVs, then we just use it via the C<#define> in F<perl.h>. Otherwise, | |
8b7fad81 | 1713 | fall back on this implementation. The main use of this function |
796b6530 | 1714 | is catching C<-0.0>. |
ed140128 | 1715 | |
796b6530 KW |
1716 | C<Configure> notes: This function is called C<'Perl_signbit'> instead of a |
1717 | plain C<'signbit'> because it is easy to imagine a system having a C<signbit()> | |
ed140128 | 1718 | function or macro that doesn't happen to work with our particular choice |
796b6530 | 1719 | of NVs. We shouldn't just re-C<#define> C<signbit> as C<Perl_signbit> and expect |
ed140128 | 1720 | the standard system headers to be happy. Also, this is a no-context |
796b6530 KW |
1721 | function (no C<pTHX_>) because C<Perl_signbit()> is usually re-C<#defined> in |
1722 | F<perl.h> as a simple macro call to the system's C<signbit()>. | |
1723 | Users should just always call C<Perl_signbit()>. | |
ed140128 AD |
1724 | |
1725 | =cut | |
1726 | */ | |
1727 | #if !defined(HAS_SIGNBIT) | |
1728 | int | |
1729 | Perl_signbit(NV x) { | |
8b7fad81 | 1730 | # ifdef Perl_fp_class_nzero |
406d5545 JH |
1731 | return Perl_fp_class_nzero(x); |
1732 | /* Try finding the high byte, and assume it's highest bit | |
1733 | * is the sign. This assumption is probably wrong somewhere. */ | |
572cd850 JH |
1734 | # elif defined(USE_LONG_DOUBLE) && LONG_DOUBLEKIND == LONG_DOUBLE_IS_X86_80_BIT_LITTLE_ENDIAN |
1735 | return (((unsigned char *)&x)[9] & 0x80); | |
1736 | # elif defined(NV_LITTLE_ENDIAN) | |
1737 | /* Note that NVSIZE is sizeof(NV), which would make the below be | |
1738 | * wrong if the end bytes are unused, which happens with the x86 | |
1739 | * 80-bit long doubles, which is why take care of that above. */ | |
1740 | return (((unsigned char *)&x)[NVSIZE - 1] & 0x80); | |
1741 | # elif defined(NV_BIG_ENDIAN) | |
1742 | return (((unsigned char *)&x)[0] & 0x80); | |
1743 | # else | |
406d5545 | 1744 | /* This last resort fallback is wrong for the negative zero. */ |
3585840c | 1745 | return (x < 0.0) ? 1 : 0; |
572cd850 | 1746 | # endif |
ed140128 AD |
1747 | } |
1748 | #endif | |
1749 | ||
1750 | /* | |
14d04a33 | 1751 | * ex: set ts=8 sts=4 sw=4 et: |
37442d52 | 1752 | */ |