Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 1 | /* -*- Mode: C; c-file-style: "python" -*- */ |
| 2 | |
| 3 | #include <Python.h> |
| 4 | #include <locale.h> |
| 5 | |
Mark Dickinson | 3b38df2 | 2009-10-26 14:36:29 +0000 | [diff] [blame] | 6 | /* Case-insensitive string match used for nan and inf detection; t should be |
| 7 | lower-case. Returns 1 for a successful match, 0 otherwise. */ |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 8 | |
| 9 | static int |
| 10 | case_insensitive_match(const char *s, const char *t) |
| 11 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 12 | while(*t && Py_TOLOWER(*s) == *t) { |
| 13 | s++; |
| 14 | t++; |
| 15 | } |
| 16 | return *t ? 0 : 1; |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 17 | } |
| 18 | |
Mark Dickinson | 3b38df2 | 2009-10-26 14:36:29 +0000 | [diff] [blame] | 19 | /* _Py_parse_inf_or_nan: Attempt to parse a string of the form "nan", "inf" or |
| 20 | "infinity", with an optional leading sign of "+" or "-". On success, |
| 21 | return the NaN or Infinity as a double and set *endptr to point just beyond |
| 22 | the successfully parsed portion of the string. On failure, return -1.0 and |
| 23 | set *endptr to point to the start of the string. */ |
| 24 | |
Mark Dickinson | e383e82 | 2012-04-29 15:31:56 +0100 | [diff] [blame] | 25 | #ifndef PY_NO_SHORT_FLOAT_REPR |
| 26 | |
| 27 | double |
| 28 | _Py_parse_inf_or_nan(const char *p, char **endptr) |
| 29 | { |
| 30 | double retval; |
| 31 | const char *s; |
| 32 | int negate = 0; |
| 33 | |
| 34 | s = p; |
| 35 | if (*s == '-') { |
| 36 | negate = 1; |
| 37 | s++; |
| 38 | } |
| 39 | else if (*s == '+') { |
| 40 | s++; |
| 41 | } |
| 42 | if (case_insensitive_match(s, "inf")) { |
| 43 | s += 3; |
| 44 | if (case_insensitive_match(s, "inity")) |
| 45 | s += 5; |
| 46 | retval = _Py_dg_infinity(negate); |
| 47 | } |
| 48 | else if (case_insensitive_match(s, "nan")) { |
| 49 | s += 3; |
| 50 | retval = _Py_dg_stdnan(negate); |
| 51 | } |
| 52 | else { |
| 53 | s = p; |
| 54 | retval = -1.0; |
| 55 | } |
| 56 | *endptr = (char *)s; |
| 57 | return retval; |
| 58 | } |
| 59 | |
| 60 | #else |
| 61 | |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 62 | double |
| 63 | _Py_parse_inf_or_nan(const char *p, char **endptr) |
| 64 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 65 | double retval; |
| 66 | const char *s; |
| 67 | int negate = 0; |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 68 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 69 | s = p; |
| 70 | if (*s == '-') { |
| 71 | negate = 1; |
| 72 | s++; |
| 73 | } |
| 74 | else if (*s == '+') { |
| 75 | s++; |
| 76 | } |
| 77 | if (case_insensitive_match(s, "inf")) { |
| 78 | s += 3; |
| 79 | if (case_insensitive_match(s, "inity")) |
| 80 | s += 5; |
| 81 | retval = negate ? -Py_HUGE_VAL : Py_HUGE_VAL; |
| 82 | } |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 83 | #ifdef Py_NAN |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 84 | else if (case_insensitive_match(s, "nan")) { |
| 85 | s += 3; |
| 86 | retval = negate ? -Py_NAN : Py_NAN; |
| 87 | } |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 88 | #endif |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 89 | else { |
| 90 | s = p; |
| 91 | retval = -1.0; |
| 92 | } |
| 93 | *endptr = (char *)s; |
| 94 | return retval; |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 95 | } |
| 96 | |
Mark Dickinson | e383e82 | 2012-04-29 15:31:56 +0100 | [diff] [blame] | 97 | #endif |
| 98 | |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 99 | /** |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 100 | * _PyOS_ascii_strtod: |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 101 | * @nptr: the string to convert to a numeric value. |
| 102 | * @endptr: if non-%NULL, it returns the character after |
| 103 | * the last character used in the conversion. |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 104 | * |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 105 | * Converts a string to a #gdouble value. |
| 106 | * This function behaves like the standard strtod() function |
| 107 | * does in the C locale. It does this without actually |
| 108 | * changing the current locale, since that would not be |
| 109 | * thread-safe. |
| 110 | * |
| 111 | * This function is typically used when reading configuration |
| 112 | * files or other non-user input that should be locale independent. |
| 113 | * To handle input from the user you should normally use the |
| 114 | * locale-sensitive system strtod() function. |
| 115 | * |
| 116 | * If the correct value would cause overflow, plus or minus %HUGE_VAL |
| 117 | * is returned (according to the sign of the value), and %ERANGE is |
| 118 | * stored in %errno. If the correct value would cause underflow, |
| 119 | * zero is returned and %ERANGE is stored in %errno. |
Thomas Wouters | 4d70c3d | 2006-06-08 14:42:34 +0000 | [diff] [blame] | 120 | * If memory allocation fails, %ENOMEM is stored in %errno. |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 121 | * |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 122 | * This function resets %errno before calling strtod() so that |
| 123 | * you can reliably detect overflow and underflow. |
| 124 | * |
| 125 | * Return value: the #gdouble value. |
| 126 | **/ |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 127 | |
| 128 | #ifndef PY_NO_SHORT_FLOAT_REPR |
| 129 | |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 130 | static double |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 131 | _PyOS_ascii_strtod(const char *nptr, char **endptr) |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 132 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 133 | double result; |
| 134 | _Py_SET_53BIT_PRECISION_HEADER; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 135 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 136 | assert(nptr != NULL); |
| 137 | /* Set errno to zero, so that we can distinguish zero results |
| 138 | and underflows */ |
| 139 | errno = 0; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 140 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 141 | _Py_SET_53BIT_PRECISION_START; |
| 142 | result = _Py_dg_strtod(nptr, endptr); |
| 143 | _Py_SET_53BIT_PRECISION_END; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 144 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 145 | if (*endptr == nptr) |
| 146 | /* string might represent an inf or nan */ |
| 147 | result = _Py_parse_inf_or_nan(nptr, endptr); |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 148 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 149 | return result; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 150 | |
| 151 | } |
| 152 | |
| 153 | #else |
| 154 | |
| 155 | /* |
| 156 | Use system strtod; since strtod is locale aware, we may |
| 157 | have to first fix the decimal separator. |
| 158 | |
| 159 | Note that unlike _Py_dg_strtod, the system strtod may not always give |
| 160 | correctly rounded results. |
| 161 | */ |
| 162 | |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 163 | static double |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 164 | _PyOS_ascii_strtod(const char *nptr, char **endptr) |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 165 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 166 | char *fail_pos; |
Georg Brandl | 6083a4b | 2013-10-14 06:51:46 +0200 | [diff] [blame] | 167 | double val; |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 168 | struct lconv *locale_data; |
| 169 | const char *decimal_point; |
| 170 | size_t decimal_point_len; |
| 171 | const char *p, *decimal_point_pos; |
| 172 | const char *end = NULL; /* Silence gcc */ |
| 173 | const char *digits_pos = NULL; |
| 174 | int negate = 0; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 175 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 176 | assert(nptr != NULL); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 177 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 178 | fail_pos = NULL; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 179 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 180 | locale_data = localeconv(); |
| 181 | decimal_point = locale_data->decimal_point; |
| 182 | decimal_point_len = strlen(decimal_point); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 183 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 184 | assert(decimal_point_len != 0); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 185 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 186 | decimal_point_pos = NULL; |
Christian Heimes | faf2f63 | 2008-01-06 16:59:19 +0000 | [diff] [blame] | 187 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 188 | /* Parse infinities and nans */ |
| 189 | val = _Py_parse_inf_or_nan(nptr, endptr); |
| 190 | if (*endptr != nptr) |
| 191 | return val; |
Mark Dickinson | bd16edd | 2009-05-20 22:05:25 +0000 | [diff] [blame] | 192 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 193 | /* Set errno to zero, so that we can distinguish zero results |
| 194 | and underflows */ |
| 195 | errno = 0; |
Mark Dickinson | 6d65df1 | 2009-04-26 15:30:47 +0000 | [diff] [blame] | 196 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 197 | /* We process the optional sign manually, then pass the remainder to |
| 198 | the system strtod. This ensures that the result of an underflow |
| 199 | has the correct sign. (bug #1725) */ |
| 200 | p = nptr; |
| 201 | /* Process leading sign, if present */ |
| 202 | if (*p == '-') { |
| 203 | negate = 1; |
| 204 | p++; |
| 205 | } |
| 206 | else if (*p == '+') { |
| 207 | p++; |
| 208 | } |
Christian Heimes | faf2f63 | 2008-01-06 16:59:19 +0000 | [diff] [blame] | 209 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 210 | /* Some platform strtods accept hex floats; Python shouldn't (at the |
| 211 | moment), so we check explicitly for strings starting with '0x'. */ |
| 212 | if (*p == '0' && (*(p+1) == 'x' || *(p+1) == 'X')) |
| 213 | goto invalid_string; |
Mark Dickinson | 6d65df1 | 2009-04-26 15:30:47 +0000 | [diff] [blame] | 214 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 215 | /* Check that what's left begins with a digit or decimal point */ |
| 216 | if (!Py_ISDIGIT(*p) && *p != '.') |
| 217 | goto invalid_string; |
Mark Dickinson | 6d65df1 | 2009-04-26 15:30:47 +0000 | [diff] [blame] | 218 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 219 | digits_pos = p; |
| 220 | if (decimal_point[0] != '.' || |
| 221 | decimal_point[1] != 0) |
| 222 | { |
| 223 | /* Look for a '.' in the input; if present, it'll need to be |
| 224 | swapped for the current locale's decimal point before we |
| 225 | call strtod. On the other hand, if we find the current |
| 226 | locale's decimal point then the input is invalid. */ |
| 227 | while (Py_ISDIGIT(*p)) |
| 228 | p++; |
Neal Norwitz | e7214a1 | 2005-12-18 05:03:17 +0000 | [diff] [blame] | 229 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 230 | if (*p == '.') |
| 231 | { |
| 232 | decimal_point_pos = p++; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 233 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 234 | /* locate end of number */ |
| 235 | while (Py_ISDIGIT(*p)) |
| 236 | p++; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 237 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 238 | if (*p == 'e' || *p == 'E') |
| 239 | p++; |
| 240 | if (*p == '+' || *p == '-') |
| 241 | p++; |
| 242 | while (Py_ISDIGIT(*p)) |
| 243 | p++; |
| 244 | end = p; |
| 245 | } |
| 246 | else if (strncmp(p, decimal_point, decimal_point_len) == 0) |
| 247 | /* Python bug #1417699 */ |
| 248 | goto invalid_string; |
| 249 | /* For the other cases, we need not convert the decimal |
| 250 | point */ |
| 251 | } |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 252 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 253 | if (decimal_point_pos) { |
| 254 | char *copy, *c; |
| 255 | /* Create a copy of the input, with the '.' converted to the |
| 256 | locale-specific decimal point */ |
| 257 | copy = (char *)PyMem_MALLOC(end - digits_pos + |
| 258 | 1 + decimal_point_len); |
| 259 | if (copy == NULL) { |
| 260 | *endptr = (char *)nptr; |
| 261 | errno = ENOMEM; |
| 262 | return val; |
| 263 | } |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 264 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 265 | c = copy; |
| 266 | memcpy(c, digits_pos, decimal_point_pos - digits_pos); |
| 267 | c += decimal_point_pos - digits_pos; |
| 268 | memcpy(c, decimal_point, decimal_point_len); |
| 269 | c += decimal_point_len; |
| 270 | memcpy(c, decimal_point_pos + 1, |
| 271 | end - (decimal_point_pos + 1)); |
| 272 | c += end - (decimal_point_pos + 1); |
| 273 | *c = 0; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 274 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 275 | val = strtod(copy, &fail_pos); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 276 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 277 | if (fail_pos) |
| 278 | { |
| 279 | if (fail_pos > decimal_point_pos) |
| 280 | fail_pos = (char *)digits_pos + |
| 281 | (fail_pos - copy) - |
| 282 | (decimal_point_len - 1); |
| 283 | else |
| 284 | fail_pos = (char *)digits_pos + |
| 285 | (fail_pos - copy); |
| 286 | } |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 287 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 288 | PyMem_FREE(copy); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 289 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 290 | } |
| 291 | else { |
| 292 | val = strtod(digits_pos, &fail_pos); |
| 293 | } |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 294 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 295 | if (fail_pos == digits_pos) |
| 296 | goto invalid_string; |
Christian Heimes | faf2f63 | 2008-01-06 16:59:19 +0000 | [diff] [blame] | 297 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 298 | if (negate && fail_pos != nptr) |
| 299 | val = -val; |
| 300 | *endptr = fail_pos; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 301 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 302 | return val; |
Mark Dickinson | 6d65df1 | 2009-04-26 15:30:47 +0000 | [diff] [blame] | 303 | |
| 304 | invalid_string: |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 305 | *endptr = (char*)nptr; |
| 306 | errno = EINVAL; |
| 307 | return -1.0; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 308 | } |
| 309 | |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 310 | #endif |
| 311 | |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 312 | /* PyOS_string_to_double converts a null-terminated byte string s (interpreted |
| 313 | as a string of ASCII characters) to a float. The string should not have |
| 314 | leading or trailing whitespace. The conversion is independent of the |
| 315 | current locale. |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 316 | |
| 317 | If endptr is NULL, try to convert the whole string. Raise ValueError and |
| 318 | return -1.0 if the string is not a valid representation of a floating-point |
| 319 | number. |
| 320 | |
| 321 | If endptr is non-NULL, try to convert as much of the string as possible. |
| 322 | If no initial segment of the string is the valid representation of a |
| 323 | floating-point number then *endptr is set to point to the beginning of the |
| 324 | string, -1.0 is returned and again ValueError is raised. |
| 325 | |
| 326 | On overflow (e.g., when trying to convert '1e500' on an IEEE 754 machine), |
| 327 | if overflow_exception is NULL then +-Py_HUGE_VAL is returned, and no Python |
Serhiy Storchaka | 56a6d85 | 2014-12-01 18:28:43 +0200 | [diff] [blame] | 328 | exception is raised. Otherwise, overflow_exception should point to |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 329 | a Python exception, this exception will be raised, -1.0 will be returned, |
| 330 | and *endptr will point just past the end of the converted value. |
| 331 | |
| 332 | If any other failure occurs (for example lack of memory), -1.0 is returned |
| 333 | and the appropriate Python exception will have been set. |
| 334 | */ |
| 335 | |
| 336 | double |
| 337 | PyOS_string_to_double(const char *s, |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 338 | char **endptr, |
| 339 | PyObject *overflow_exception) |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 340 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 341 | double x, result=-1.0; |
| 342 | char *fail_pos; |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 343 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 344 | errno = 0; |
| 345 | PyFPE_START_PROTECT("PyOS_string_to_double", return -1.0) |
| 346 | x = _PyOS_ascii_strtod(s, &fail_pos); |
| 347 | PyFPE_END_PROTECT(x) |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 348 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 349 | if (errno == ENOMEM) { |
| 350 | PyErr_NoMemory(); |
| 351 | fail_pos = (char *)s; |
| 352 | } |
| 353 | else if (!endptr && (fail_pos == s || *fail_pos != '\0')) |
| 354 | PyErr_Format(PyExc_ValueError, |
| 355 | "could not convert string to float: " |
| 356 | "%.200s", s); |
| 357 | else if (fail_pos == s) |
| 358 | PyErr_Format(PyExc_ValueError, |
| 359 | "could not convert string to float: " |
| 360 | "%.200s", s); |
| 361 | else if (errno == ERANGE && fabs(x) >= 1.0 && overflow_exception) |
| 362 | PyErr_Format(overflow_exception, |
| 363 | "value too large to convert to float: " |
| 364 | "%.200s", s); |
| 365 | else |
| 366 | result = x; |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 367 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 368 | if (endptr != NULL) |
| 369 | *endptr = fail_pos; |
| 370 | return result; |
Mark Dickinson | 725bfd8 | 2009-05-03 20:33:40 +0000 | [diff] [blame] | 371 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 372 | |
Brett Cannon | a721aba | 2016-09-09 14:57:09 -0700 | [diff] [blame] | 373 | /* Remove underscores that follow the underscore placement rule from |
| 374 | the string and then call the `innerfunc` function on the result. |
| 375 | It should return a new object or NULL on exception. |
| 376 | |
| 377 | `what` is used for the error message emitted when underscores are detected |
| 378 | that don't follow the rule. `arg` is an opaque pointer passed to the inner |
| 379 | function. |
| 380 | |
| 381 | This is used to implement underscore-agnostic conversion for floats |
| 382 | and complex numbers. |
| 383 | */ |
| 384 | PyObject * |
| 385 | _Py_string_to_number_with_underscores( |
| 386 | const char *s, Py_ssize_t orig_len, const char *what, PyObject *obj, void *arg, |
| 387 | PyObject *(*innerfunc)(const char *, Py_ssize_t, void *)) |
| 388 | { |
| 389 | char prev; |
| 390 | const char *p, *last; |
| 391 | char *dup, *end; |
| 392 | PyObject *result; |
| 393 | |
INADA Naoki | 16dfca4 | 2018-07-14 12:06:43 +0900 | [diff] [blame^] | 394 | assert(s[orig_len] == '\0'); |
| 395 | |
Brett Cannon | a721aba | 2016-09-09 14:57:09 -0700 | [diff] [blame] | 396 | if (strchr(s, '_') == NULL) { |
| 397 | return innerfunc(s, orig_len, arg); |
| 398 | } |
| 399 | |
| 400 | dup = PyMem_Malloc(orig_len + 1); |
| 401 | end = dup; |
| 402 | prev = '\0'; |
| 403 | last = s + orig_len; |
| 404 | for (p = s; *p; p++) { |
| 405 | if (*p == '_') { |
| 406 | /* Underscores are only allowed after digits. */ |
| 407 | if (!(prev >= '0' && prev <= '9')) { |
| 408 | goto error; |
| 409 | } |
| 410 | } |
| 411 | else { |
| 412 | *end++ = *p; |
| 413 | /* Underscores are only allowed before digits. */ |
| 414 | if (prev == '_' && !(*p >= '0' && *p <= '9')) { |
| 415 | goto error; |
| 416 | } |
| 417 | } |
| 418 | prev = *p; |
| 419 | } |
| 420 | /* Underscores are not allowed at the end. */ |
| 421 | if (prev == '_') { |
| 422 | goto error; |
| 423 | } |
| 424 | /* No embedded NULs allowed. */ |
| 425 | if (p != last) { |
| 426 | goto error; |
| 427 | } |
| 428 | *end = '\0'; |
| 429 | result = innerfunc(dup, end - dup, arg); |
| 430 | PyMem_Free(dup); |
| 431 | return result; |
| 432 | |
| 433 | error: |
| 434 | PyMem_Free(dup); |
| 435 | PyErr_Format(PyExc_ValueError, |
Barry Warsaw | b2e5794 | 2017-09-14 18:13:16 -0700 | [diff] [blame] | 436 | "could not convert string to %s: " |
| 437 | "%R", what, obj); |
Brett Cannon | a721aba | 2016-09-09 14:57:09 -0700 | [diff] [blame] | 438 | return NULL; |
| 439 | } |
| 440 | |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 441 | #ifdef PY_NO_SHORT_FLOAT_REPR |
| 442 | |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 443 | /* Given a string that may have a decimal point in the current |
| 444 | locale, change it back to a dot. Since the string cannot get |
| 445 | longer, no need for a maximum buffer size parameter. */ |
| 446 | Py_LOCAL_INLINE(void) |
| 447 | change_decimal_from_locale_to_dot(char* buffer) |
| 448 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 449 | struct lconv *locale_data = localeconv(); |
| 450 | const char *decimal_point = locale_data->decimal_point; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 451 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 452 | if (decimal_point[0] != '.' || decimal_point[1] != 0) { |
| 453 | size_t decimal_point_len = strlen(decimal_point); |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 454 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 455 | if (*buffer == '+' || *buffer == '-') |
| 456 | buffer++; |
| 457 | while (Py_ISDIGIT(*buffer)) |
| 458 | buffer++; |
| 459 | if (strncmp(buffer, decimal_point, decimal_point_len) == 0) { |
| 460 | *buffer = '.'; |
| 461 | buffer++; |
| 462 | if (decimal_point_len > 1) { |
| 463 | /* buffer needs to get smaller */ |
| 464 | size_t rest_len = strlen(buffer + |
| 465 | (decimal_point_len - 1)); |
| 466 | memmove(buffer, |
| 467 | buffer + (decimal_point_len - 1), |
| 468 | rest_len); |
| 469 | buffer[rest_len] = 0; |
| 470 | } |
| 471 | } |
| 472 | } |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 473 | } |
| 474 | |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 475 | |
Christian Heimes | c3f30c4 | 2008-02-22 16:37:40 +0000 | [diff] [blame] | 476 | /* From the C99 standard, section 7.19.6: |
| 477 | The exponent always contains at least two digits, and only as many more digits |
| 478 | as necessary to represent the exponent. |
| 479 | */ |
| 480 | #define MIN_EXPONENT_DIGITS 2 |
| 481 | |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 482 | /* Ensure that any exponent, if present, is at least MIN_EXPONENT_DIGITS |
| 483 | in length. */ |
| 484 | Py_LOCAL_INLINE(void) |
Mark Dickinson | ce95e56 | 2009-04-26 20:02:24 +0000 | [diff] [blame] | 485 | ensure_minimum_exponent_length(char* buffer, size_t buf_size) |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 486 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 487 | char *p = strpbrk(buffer, "eE"); |
| 488 | if (p && (*(p + 1) == '-' || *(p + 1) == '+')) { |
| 489 | char *start = p + 2; |
| 490 | int exponent_digit_cnt = 0; |
| 491 | int leading_zero_cnt = 0; |
| 492 | int in_leading_zeros = 1; |
| 493 | int significant_digit_cnt; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 494 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 495 | /* Skip over the exponent and the sign. */ |
| 496 | p += 2; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 497 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 498 | /* Find the end of the exponent, keeping track of leading |
| 499 | zeros. */ |
| 500 | while (*p && Py_ISDIGIT(*p)) { |
| 501 | if (in_leading_zeros && *p == '0') |
| 502 | ++leading_zero_cnt; |
| 503 | if (*p != '0') |
| 504 | in_leading_zeros = 0; |
| 505 | ++p; |
| 506 | ++exponent_digit_cnt; |
| 507 | } |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 508 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 509 | significant_digit_cnt = exponent_digit_cnt - leading_zero_cnt; |
| 510 | if (exponent_digit_cnt == MIN_EXPONENT_DIGITS) { |
| 511 | /* If there are 2 exactly digits, we're done, |
| 512 | regardless of what they contain */ |
| 513 | } |
| 514 | else if (exponent_digit_cnt > MIN_EXPONENT_DIGITS) { |
| 515 | int extra_zeros_cnt; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 516 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 517 | /* There are more than 2 digits in the exponent. See |
| 518 | if we can delete some of the leading zeros */ |
| 519 | if (significant_digit_cnt < MIN_EXPONENT_DIGITS) |
| 520 | significant_digit_cnt = MIN_EXPONENT_DIGITS; |
| 521 | extra_zeros_cnt = exponent_digit_cnt - |
| 522 | significant_digit_cnt; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 523 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 524 | /* Delete extra_zeros_cnt worth of characters from the |
| 525 | front of the exponent */ |
| 526 | assert(extra_zeros_cnt >= 0); |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 527 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 528 | /* Add one to significant_digit_cnt to copy the |
| 529 | trailing 0 byte, thus setting the length */ |
| 530 | memmove(start, |
| 531 | start + extra_zeros_cnt, |
| 532 | significant_digit_cnt + 1); |
| 533 | } |
| 534 | else { |
| 535 | /* If there are fewer than 2 digits, add zeros |
| 536 | until there are 2, if there's enough room */ |
| 537 | int zeros = MIN_EXPONENT_DIGITS - exponent_digit_cnt; |
| 538 | if (start + zeros + exponent_digit_cnt + 1 |
| 539 | < buffer + buf_size) { |
| 540 | memmove(start + zeros, start, |
| 541 | exponent_digit_cnt + 1); |
| 542 | memset(start, '0', zeros); |
| 543 | } |
| 544 | } |
| 545 | } |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 546 | } |
| 547 | |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 548 | /* Remove trailing zeros after the decimal point from a numeric string; also |
| 549 | remove the decimal point if all digits following it are zero. The numeric |
| 550 | string must end in '\0', and should not have any leading or trailing |
| 551 | whitespace. Assumes that the decimal point is '.'. */ |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 552 | Py_LOCAL_INLINE(void) |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 553 | remove_trailing_zeros(char *buffer) |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 554 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 555 | char *old_fraction_end, *new_fraction_end, *end, *p; |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 556 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 557 | p = buffer; |
| 558 | if (*p == '-' || *p == '+') |
| 559 | /* Skip leading sign, if present */ |
| 560 | ++p; |
| 561 | while (Py_ISDIGIT(*p)) |
| 562 | ++p; |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 563 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 564 | /* if there's no decimal point there's nothing to do */ |
| 565 | if (*p++ != '.') |
| 566 | return; |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 567 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 568 | /* scan any digits after the point */ |
| 569 | while (Py_ISDIGIT(*p)) |
| 570 | ++p; |
| 571 | old_fraction_end = p; |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 572 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 573 | /* scan up to ending '\0' */ |
| 574 | while (*p != '\0') |
| 575 | p++; |
| 576 | /* +1 to make sure that we move the null byte as well */ |
| 577 | end = p+1; |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 578 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 579 | /* scan back from fraction_end, looking for removable zeros */ |
| 580 | p = old_fraction_end; |
| 581 | while (*(p-1) == '0') |
| 582 | --p; |
| 583 | /* and remove point if we've got that far */ |
| 584 | if (*(p-1) == '.') |
| 585 | --p; |
| 586 | new_fraction_end = p; |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 587 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 588 | memmove(new_fraction_end, old_fraction_end, end-old_fraction_end); |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 589 | } |
| 590 | |
| 591 | /* Ensure that buffer has a decimal point in it. The decimal point will not |
| 592 | be in the current locale, it will always be '.'. Don't add a decimal point |
| 593 | if an exponent is present. Also, convert to exponential notation where |
| 594 | adding a '.0' would produce too many significant digits (see issue 5864). |
| 595 | |
| 596 | Returns a pointer to the fixed buffer, or NULL on failure. |
| 597 | */ |
| 598 | Py_LOCAL_INLINE(char *) |
| 599 | ensure_decimal_point(char* buffer, size_t buf_size, int precision) |
| 600 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 601 | int digit_count, insert_count = 0, convert_to_exp = 0; |
Serhiy Storchaka | e2f92de | 2017-11-11 13:06:26 +0200 | [diff] [blame] | 602 | const char *chars_to_insert; |
| 603 | char *digits_start; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 604 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 605 | /* search for the first non-digit character */ |
| 606 | char *p = buffer; |
| 607 | if (*p == '-' || *p == '+') |
| 608 | /* Skip leading sign, if present. I think this could only |
| 609 | ever be '-', but it can't hurt to check for both. */ |
| 610 | ++p; |
| 611 | digits_start = p; |
| 612 | while (*p && Py_ISDIGIT(*p)) |
| 613 | ++p; |
| 614 | digit_count = Py_SAFE_DOWNCAST(p - digits_start, Py_ssize_t, int); |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 615 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 616 | if (*p == '.') { |
| 617 | if (Py_ISDIGIT(*(p+1))) { |
| 618 | /* Nothing to do, we already have a decimal |
| 619 | point and a digit after it */ |
| 620 | } |
| 621 | else { |
| 622 | /* We have a decimal point, but no following |
| 623 | digit. Insert a zero after the decimal. */ |
| 624 | /* can't ever get here via PyOS_double_to_string */ |
| 625 | assert(precision == -1); |
| 626 | ++p; |
| 627 | chars_to_insert = "0"; |
| 628 | insert_count = 1; |
| 629 | } |
| 630 | } |
| 631 | else if (!(*p == 'e' || *p == 'E')) { |
| 632 | /* Don't add ".0" if we have an exponent. */ |
| 633 | if (digit_count == precision) { |
| 634 | /* issue 5864: don't add a trailing .0 in the case |
| 635 | where the '%g'-formatted result already has as many |
| 636 | significant digits as were requested. Switch to |
| 637 | exponential notation instead. */ |
| 638 | convert_to_exp = 1; |
| 639 | /* no exponent, no point, and we shouldn't land here |
| 640 | for infs and nans, so we must be at the end of the |
| 641 | string. */ |
| 642 | assert(*p == '\0'); |
| 643 | } |
| 644 | else { |
| 645 | assert(precision == -1 || digit_count < precision); |
| 646 | chars_to_insert = ".0"; |
| 647 | insert_count = 2; |
| 648 | } |
| 649 | } |
| 650 | if (insert_count) { |
| 651 | size_t buf_len = strlen(buffer); |
| 652 | if (buf_len + insert_count + 1 >= buf_size) { |
| 653 | /* If there is not enough room in the buffer |
| 654 | for the additional text, just skip it. It's |
| 655 | not worth generating an error over. */ |
| 656 | } |
| 657 | else { |
| 658 | memmove(p + insert_count, p, |
| 659 | buffer + strlen(buffer) - p + 1); |
| 660 | memcpy(p, chars_to_insert, insert_count); |
| 661 | } |
| 662 | } |
| 663 | if (convert_to_exp) { |
| 664 | int written; |
| 665 | size_t buf_avail; |
| 666 | p = digits_start; |
| 667 | /* insert decimal point */ |
| 668 | assert(digit_count >= 1); |
| 669 | memmove(p+2, p+1, digit_count); /* safe, but overwrites nul */ |
| 670 | p[1] = '.'; |
| 671 | p += digit_count+1; |
| 672 | assert(p <= buf_size+buffer); |
| 673 | buf_avail = buf_size+buffer-p; |
| 674 | if (buf_avail == 0) |
| 675 | return NULL; |
| 676 | /* Add exponent. It's okay to use lower case 'e': we only |
| 677 | arrive here as a result of using the empty format code or |
| 678 | repr/str builtins and those never want an upper case 'E' */ |
| 679 | written = PyOS_snprintf(p, buf_avail, "e%+.02d", digit_count-1); |
| 680 | if (!(0 <= written && |
| 681 | written < Py_SAFE_DOWNCAST(buf_avail, size_t, int))) |
| 682 | /* output truncated, or something else bad happened */ |
| 683 | return NULL; |
| 684 | remove_trailing_zeros(buffer); |
| 685 | } |
| 686 | return buffer; |
Eric Smith | b2c7af8 | 2008-04-30 02:12:09 +0000 | [diff] [blame] | 687 | } |
| 688 | |
Christian Heimes | c3f30c4 | 2008-02-22 16:37:40 +0000 | [diff] [blame] | 689 | /* see FORMATBUFLEN in unicodeobject.c */ |
| 690 | #define FLOAT_FORMATBUFLEN 120 |
| 691 | |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 692 | /** |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 693 | * _PyOS_ascii_formatd: |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 694 | * @buffer: A buffer to place the resulting string in |
Christian Heimes | b186d00 | 2008-03-18 15:15:01 +0000 | [diff] [blame] | 695 | * @buf_size: The length of the buffer. |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 696 | * @format: The printf()-style format to use for the |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 697 | * code to use for converting. |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 698 | * @d: The #gdouble to convert |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 699 | * @precision: The precision to use when formatting. |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 700 | * |
| 701 | * Converts a #gdouble to a string, using the '.' as |
| 702 | * decimal point. To format the number you pass in |
| 703 | * a printf()-style format string. Allowed conversion |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 704 | * specifiers are 'e', 'E', 'f', 'F', 'g', 'G', and 'Z'. |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 705 | * |
Christian Heimes | b186d00 | 2008-03-18 15:15:01 +0000 | [diff] [blame] | 706 | * 'Z' is the same as 'g', except it always has a decimal and |
| 707 | * at least one digit after the decimal. |
Christian Heimes | c3f30c4 | 2008-02-22 16:37:40 +0000 | [diff] [blame] | 708 | * |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 709 | * Return value: The pointer to the buffer with the converted string. |
Mark Dickinson | d3ca557 | 2009-04-29 18:47:07 +0000 | [diff] [blame] | 710 | * On failure returns NULL but does not set any Python exception. |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 711 | **/ |
Eric Smith | 68af50b | 2010-02-22 14:58:30 +0000 | [diff] [blame] | 712 | static char * |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 713 | _PyOS_ascii_formatd(char *buffer, |
| 714 | size_t buf_size, |
| 715 | const char *format, |
| 716 | double d, |
| 717 | int precision) |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 718 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 719 | char format_char; |
| 720 | size_t format_len = strlen(format); |
Christian Heimes | c3f30c4 | 2008-02-22 16:37:40 +0000 | [diff] [blame] | 721 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 722 | /* Issue 2264: code 'Z' requires copying the format. 'Z' is 'g', but |
| 723 | also with at least one character past the decimal. */ |
| 724 | char tmp_format[FLOAT_FORMATBUFLEN]; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 725 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 726 | /* The last character in the format string must be the format char */ |
| 727 | format_char = format[format_len - 1]; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 728 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 729 | if (format[0] != '%') |
| 730 | return NULL; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 731 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 732 | /* I'm not sure why this test is here. It's ensuring that the format |
| 733 | string after the first character doesn't have a single quote, a |
| 734 | lowercase l, or a percent. This is the reverse of the commented-out |
| 735 | test about 10 lines ago. */ |
| 736 | if (strpbrk(format + 1, "'l%")) |
| 737 | return NULL; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 738 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 739 | /* Also curious about this function is that it accepts format strings |
| 740 | like "%xg", which are invalid for floats. In general, the |
| 741 | interface to this function is not very good, but changing it is |
| 742 | difficult because it's a public API. */ |
Christian Heimes | b186d00 | 2008-03-18 15:15:01 +0000 | [diff] [blame] | 743 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 744 | if (!(format_char == 'e' || format_char == 'E' || |
| 745 | format_char == 'f' || format_char == 'F' || |
| 746 | format_char == 'g' || format_char == 'G' || |
| 747 | format_char == 'Z')) |
| 748 | return NULL; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 749 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 750 | /* Map 'Z' format_char to 'g', by copying the format string and |
| 751 | replacing the final char with a 'g' */ |
| 752 | if (format_char == 'Z') { |
| 753 | if (format_len + 1 >= sizeof(tmp_format)) { |
| 754 | /* The format won't fit in our copy. Error out. In |
| 755 | practice, this will never happen and will be |
| 756 | detected by returning NULL */ |
| 757 | return NULL; |
| 758 | } |
| 759 | strcpy(tmp_format, format); |
| 760 | tmp_format[format_len - 1] = 'g'; |
| 761 | format = tmp_format; |
| 762 | } |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 763 | |
Christian Heimes | b186d00 | 2008-03-18 15:15:01 +0000 | [diff] [blame] | 764 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 765 | /* Have PyOS_snprintf do the hard work */ |
| 766 | PyOS_snprintf(buffer, buf_size, format, d); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 767 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 768 | /* Do various fixups on the return string */ |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 769 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 770 | /* Get the current locale, and find the decimal point string. |
| 771 | Convert that string back to a dot. */ |
| 772 | change_decimal_from_locale_to_dot(buffer); |
Christian Heimes | c3f30c4 | 2008-02-22 16:37:40 +0000 | [diff] [blame] | 773 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 774 | /* If an exponent exists, ensure that the exponent is at least |
| 775 | MIN_EXPONENT_DIGITS digits, providing the buffer is large enough |
| 776 | for the extra zeros. Also, if there are more than |
| 777 | MIN_EXPONENT_DIGITS, remove as many zeros as possible until we get |
| 778 | back to MIN_EXPONENT_DIGITS */ |
| 779 | ensure_minimum_exponent_length(buffer, buf_size); |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 780 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 781 | /* If format_char is 'Z', make sure we have at least one character |
| 782 | after the decimal point (and make sure we have a decimal point); |
| 783 | also switch to exponential notation in some edge cases where the |
| 784 | extra character would produce more significant digits that we |
| 785 | really want. */ |
| 786 | if (format_char == 'Z') |
| 787 | buffer = ensure_decimal_point(buffer, buf_size, precision); |
Christian Heimes | b186d00 | 2008-03-18 15:15:01 +0000 | [diff] [blame] | 788 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 789 | return buffer; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 790 | } |
| 791 | |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 792 | /* The fallback code to use if _Py_dg_dtoa is not available. */ |
| 793 | |
| 794 | PyAPI_FUNC(char *) PyOS_double_to_string(double val, |
| 795 | char format_code, |
| 796 | int precision, |
| 797 | int flags, |
| 798 | int *type) |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 799 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 800 | char format[32]; |
| 801 | Py_ssize_t bufsize; |
| 802 | char *buf; |
| 803 | int t, exp; |
| 804 | int upper = 0; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 805 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 806 | /* Validate format_code, and map upper and lower case */ |
| 807 | switch (format_code) { |
| 808 | case 'e': /* exponent */ |
| 809 | case 'f': /* fixed */ |
| 810 | case 'g': /* general */ |
| 811 | break; |
| 812 | case 'E': |
| 813 | upper = 1; |
| 814 | format_code = 'e'; |
| 815 | break; |
| 816 | case 'F': |
| 817 | upper = 1; |
| 818 | format_code = 'f'; |
| 819 | break; |
| 820 | case 'G': |
| 821 | upper = 1; |
| 822 | format_code = 'g'; |
| 823 | break; |
| 824 | case 'r': /* repr format */ |
| 825 | /* Supplied precision is unused, must be 0. */ |
| 826 | if (precision != 0) { |
| 827 | PyErr_BadInternalCall(); |
| 828 | return NULL; |
| 829 | } |
| 830 | /* The repr() precision (17 significant decimal digits) is the |
| 831 | minimal number that is guaranteed to have enough precision |
| 832 | so that if the number is read back in the exact same binary |
| 833 | value is recreated. This is true for IEEE floating point |
| 834 | by design, and also happens to work for all other modern |
| 835 | hardware. */ |
| 836 | precision = 17; |
| 837 | format_code = 'g'; |
| 838 | break; |
| 839 | default: |
| 840 | PyErr_BadInternalCall(); |
| 841 | return NULL; |
| 842 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 843 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 844 | /* Here's a quick-and-dirty calculation to figure out how big a buffer |
| 845 | we need. In general, for a finite float we need: |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 846 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 847 | 1 byte for each digit of the decimal significand, and |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 848 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 849 | 1 for a possible sign |
| 850 | 1 for a possible decimal point |
| 851 | 2 for a possible [eE][+-] |
| 852 | 1 for each digit of the exponent; if we allow 19 digits |
| 853 | total then we're safe up to exponents of 2**63. |
| 854 | 1 for the trailing nul byte |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 855 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 856 | This gives a total of 24 + the number of digits in the significand, |
| 857 | and the number of digits in the significand is: |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 858 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 859 | for 'g' format: at most precision, except possibly |
| 860 | when precision == 0, when it's 1. |
| 861 | for 'e' format: precision+1 |
| 862 | for 'f' format: precision digits after the point, at least 1 |
| 863 | before. To figure out how many digits appear before the point |
| 864 | we have to examine the size of the number. If fabs(val) < 1.0 |
| 865 | then there will be only one digit before the point. If |
| 866 | fabs(val) >= 1.0, then there are at most |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 867 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 868 | 1+floor(log10(ceiling(fabs(val)))) |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 869 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 870 | digits before the point (where the 'ceiling' allows for the |
| 871 | possibility that the rounding rounds the integer part of val |
| 872 | up). A safe upper bound for the above quantity is |
| 873 | 1+floor(exp/3), where exp is the unique integer such that 0.5 |
| 874 | <= fabs(val)/2**exp < 1.0. This exp can be obtained from |
| 875 | frexp. |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 876 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 877 | So we allow room for precision+1 digits for all formats, plus an |
| 878 | extra floor(exp/3) digits for 'f' format. |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 879 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 880 | */ |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 881 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 882 | if (Py_IS_NAN(val) || Py_IS_INFINITY(val)) |
| 883 | /* 3 for 'inf'/'nan', 1 for sign, 1 for '\0' */ |
| 884 | bufsize = 5; |
| 885 | else { |
| 886 | bufsize = 25 + precision; |
| 887 | if (format_code == 'f' && fabs(val) >= 1.0) { |
| 888 | frexp(val, &exp); |
| 889 | bufsize += exp/3; |
| 890 | } |
| 891 | } |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 892 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 893 | buf = PyMem_Malloc(bufsize); |
| 894 | if (buf == NULL) { |
| 895 | PyErr_NoMemory(); |
| 896 | return NULL; |
| 897 | } |
Mark Dickinson | f489caf | 2009-05-01 11:42:00 +0000 | [diff] [blame] | 898 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 899 | /* Handle nan and inf. */ |
| 900 | if (Py_IS_NAN(val)) { |
| 901 | strcpy(buf, "nan"); |
| 902 | t = Py_DTST_NAN; |
| 903 | } else if (Py_IS_INFINITY(val)) { |
| 904 | if (copysign(1., val) == 1.) |
| 905 | strcpy(buf, "inf"); |
| 906 | else |
| 907 | strcpy(buf, "-inf"); |
| 908 | t = Py_DTST_INFINITE; |
| 909 | } else { |
| 910 | t = Py_DTST_FINITE; |
| 911 | if (flags & Py_DTSF_ADD_DOT_0) |
| 912 | format_code = 'Z'; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 913 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 914 | PyOS_snprintf(format, sizeof(format), "%%%s.%i%c", |
| 915 | (flags & Py_DTSF_ALT ? "#" : ""), precision, |
| 916 | format_code); |
| 917 | _PyOS_ascii_formatd(buf, bufsize, format, val, precision); |
| 918 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 919 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 920 | /* Add sign when requested. It's convenient (esp. when formatting |
| 921 | complex numbers) to include a sign even for inf and nan. */ |
| 922 | if (flags & Py_DTSF_SIGN && buf[0] != '-') { |
| 923 | size_t len = strlen(buf); |
| 924 | /* the bufsize calculations above should ensure that we've got |
| 925 | space to add a sign */ |
| 926 | assert((size_t)bufsize >= len+2); |
| 927 | memmove(buf+1, buf, len+1); |
| 928 | buf[0] = '+'; |
| 929 | } |
| 930 | if (upper) { |
| 931 | /* Convert to upper case. */ |
| 932 | char *p1; |
| 933 | for (p1 = buf; *p1; p1++) |
| 934 | *p1 = Py_TOUPPER(*p1); |
| 935 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 936 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 937 | if (type) |
| 938 | *type = t; |
| 939 | return buf; |
Martin v. Löwis | 737ea82 | 2004-06-08 18:52:54 +0000 | [diff] [blame] | 940 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 941 | |
| 942 | #else |
| 943 | |
| 944 | /* _Py_dg_dtoa is available. */ |
| 945 | |
| 946 | /* I'm using a lookup table here so that I don't have to invent a non-locale |
| 947 | specific way to convert to uppercase */ |
| 948 | #define OFS_INF 0 |
| 949 | #define OFS_NAN 1 |
| 950 | #define OFS_E 2 |
| 951 | |
| 952 | /* The lengths of these are known to the code below, so don't change them */ |
Serhiy Storchaka | 2d06e84 | 2015-12-25 19:53:18 +0200 | [diff] [blame] | 953 | static const char * const lc_float_strings[] = { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 954 | "inf", |
| 955 | "nan", |
| 956 | "e", |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 957 | }; |
Serhiy Storchaka | 2d06e84 | 2015-12-25 19:53:18 +0200 | [diff] [blame] | 958 | static const char * const uc_float_strings[] = { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 959 | "INF", |
| 960 | "NAN", |
| 961 | "E", |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 962 | }; |
| 963 | |
| 964 | |
| 965 | /* Convert a double d to a string, and return a PyMem_Malloc'd block of |
| 966 | memory contain the resulting string. |
| 967 | |
| 968 | Arguments: |
| 969 | d is the double to be converted |
Eric Smith | 6337622 | 2009-05-05 14:04:18 +0000 | [diff] [blame] | 970 | format_code is one of 'e', 'f', 'g', 'r'. 'e', 'f' and 'g' |
| 971 | correspond to '%e', '%f' and '%g'; 'r' corresponds to repr. |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 972 | mode is one of '0', '2' or '3', and is completely determined by |
Eric Smith | 6337622 | 2009-05-05 14:04:18 +0000 | [diff] [blame] | 973 | format_code: 'e' and 'g' use mode 2; 'f' mode 3, 'r' mode 0. |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 974 | precision is the desired precision |
| 975 | always_add_sign is nonzero if a '+' sign should be included for positive |
| 976 | numbers |
| 977 | add_dot_0_if_integer is nonzero if integers in non-exponential form |
Eric Smith | 6337622 | 2009-05-05 14:04:18 +0000 | [diff] [blame] | 978 | should have ".0" added. Only applies to format codes 'r' and 'g'. |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 979 | use_alt_formatting is nonzero if alternative formatting should be |
Eric Smith | 6337622 | 2009-05-05 14:04:18 +0000 | [diff] [blame] | 980 | used. Only applies to format codes 'e', 'f' and 'g'. For code 'g', |
| 981 | at most one of use_alt_formatting and add_dot_0_if_integer should |
| 982 | be nonzero. |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 983 | type, if non-NULL, will be set to one of these constants to identify |
| 984 | the type of the 'd' argument: |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 985 | Py_DTST_FINITE |
| 986 | Py_DTST_INFINITE |
| 987 | Py_DTST_NAN |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 988 | |
| 989 | Returns a PyMem_Malloc'd block of memory containing the resulting string, |
| 990 | or NULL on error. If NULL is returned, the Python error has been set. |
| 991 | */ |
| 992 | |
| 993 | static char * |
| 994 | format_float_short(double d, char format_code, |
Victor Stinner | 7b25135 | 2013-06-24 23:37:40 +0200 | [diff] [blame] | 995 | int mode, int precision, |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 996 | int always_add_sign, int add_dot_0_if_integer, |
Serhiy Storchaka | 2d06e84 | 2015-12-25 19:53:18 +0200 | [diff] [blame] | 997 | int use_alt_formatting, const char * const *float_strings, |
| 998 | int *type) |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 999 | { |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1000 | char *buf = NULL; |
| 1001 | char *p = NULL; |
| 1002 | Py_ssize_t bufsize = 0; |
| 1003 | char *digits, *digits_end; |
| 1004 | int decpt_as_int, sign, exp_len, exp = 0, use_exp = 0; |
| 1005 | Py_ssize_t decpt, digits_len, vdigits_start, vdigits_end; |
| 1006 | _Py_SET_53BIT_PRECISION_HEADER; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1007 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1008 | /* _Py_dg_dtoa returns a digit string (no decimal point or exponent). |
| 1009 | Must be matched by a call to _Py_dg_freedtoa. */ |
| 1010 | _Py_SET_53BIT_PRECISION_START; |
| 1011 | digits = _Py_dg_dtoa(d, mode, precision, &decpt_as_int, &sign, |
| 1012 | &digits_end); |
| 1013 | _Py_SET_53BIT_PRECISION_END; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1014 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1015 | decpt = (Py_ssize_t)decpt_as_int; |
| 1016 | if (digits == NULL) { |
| 1017 | /* The only failure mode is no memory. */ |
| 1018 | PyErr_NoMemory(); |
| 1019 | goto exit; |
| 1020 | } |
| 1021 | assert(digits_end != NULL && digits_end >= digits); |
| 1022 | digits_len = digits_end - digits; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1023 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1024 | if (digits_len && !Py_ISDIGIT(digits[0])) { |
| 1025 | /* Infinities and nans here; adapt Gay's output, |
| 1026 | so convert Infinity to inf and NaN to nan, and |
| 1027 | ignore sign of nan. Then return. */ |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1028 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1029 | /* ignore the actual sign of a nan */ |
| 1030 | if (digits[0] == 'n' || digits[0] == 'N') |
| 1031 | sign = 0; |
Mark Dickinson | ad476da | 2009-04-23 19:14:16 +0000 | [diff] [blame] | 1032 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1033 | /* We only need 5 bytes to hold the result "+inf\0" . */ |
| 1034 | bufsize = 5; /* Used later in an assert. */ |
| 1035 | buf = (char *)PyMem_Malloc(bufsize); |
| 1036 | if (buf == NULL) { |
| 1037 | PyErr_NoMemory(); |
| 1038 | goto exit; |
| 1039 | } |
| 1040 | p = buf; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1041 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1042 | if (sign == 1) { |
| 1043 | *p++ = '-'; |
| 1044 | } |
| 1045 | else if (always_add_sign) { |
| 1046 | *p++ = '+'; |
| 1047 | } |
| 1048 | if (digits[0] == 'i' || digits[0] == 'I') { |
| 1049 | strncpy(p, float_strings[OFS_INF], 3); |
| 1050 | p += 3; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1051 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1052 | if (type) |
| 1053 | *type = Py_DTST_INFINITE; |
| 1054 | } |
| 1055 | else if (digits[0] == 'n' || digits[0] == 'N') { |
| 1056 | strncpy(p, float_strings[OFS_NAN], 3); |
| 1057 | p += 3; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1058 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1059 | if (type) |
| 1060 | *type = Py_DTST_NAN; |
| 1061 | } |
| 1062 | else { |
| 1063 | /* shouldn't get here: Gay's code should always return |
| 1064 | something starting with a digit, an 'I', or 'N' */ |
Barry Warsaw | b2e5794 | 2017-09-14 18:13:16 -0700 | [diff] [blame] | 1065 | Py_UNREACHABLE(); |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1066 | } |
| 1067 | goto exit; |
| 1068 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1069 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1070 | /* The result must be finite (not inf or nan). */ |
| 1071 | if (type) |
| 1072 | *type = Py_DTST_FINITE; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1073 | |
| 1074 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1075 | /* We got digits back, format them. We may need to pad 'digits' |
| 1076 | either on the left or right (or both) with extra zeros, so in |
| 1077 | general the resulting string has the form |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1078 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1079 | [<sign>]<zeros><digits><zeros>[<exponent>] |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1080 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1081 | where either of the <zeros> pieces could be empty, and there's a |
| 1082 | decimal point that could appear either in <digits> or in the |
| 1083 | leading or trailing <zeros>. |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1084 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1085 | Imagine an infinite 'virtual' string vdigits, consisting of the |
| 1086 | string 'digits' (starting at index 0) padded on both the left and |
| 1087 | right with infinite strings of zeros. We want to output a slice |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1088 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1089 | vdigits[vdigits_start : vdigits_end] |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1090 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1091 | of this virtual string. Thus if vdigits_start < 0 then we'll end |
| 1092 | up producing some leading zeros; if vdigits_end > digits_len there |
| 1093 | will be trailing zeros in the output. The next section of code |
| 1094 | determines whether to use an exponent or not, figures out the |
| 1095 | position 'decpt' of the decimal point, and computes 'vdigits_start' |
| 1096 | and 'vdigits_end'. */ |
| 1097 | vdigits_end = digits_len; |
| 1098 | switch (format_code) { |
| 1099 | case 'e': |
| 1100 | use_exp = 1; |
| 1101 | vdigits_end = precision; |
| 1102 | break; |
| 1103 | case 'f': |
| 1104 | vdigits_end = decpt + precision; |
| 1105 | break; |
| 1106 | case 'g': |
| 1107 | if (decpt <= -4 || decpt > |
| 1108 | (add_dot_0_if_integer ? precision-1 : precision)) |
| 1109 | use_exp = 1; |
| 1110 | if (use_alt_formatting) |
| 1111 | vdigits_end = precision; |
| 1112 | break; |
| 1113 | case 'r': |
| 1114 | /* convert to exponential format at 1e16. We used to convert |
| 1115 | at 1e17, but that gives odd-looking results for some values |
| 1116 | when a 16-digit 'shortest' repr is padded with bogus zeros. |
| 1117 | For example, repr(2e16+8) would give 20000000000000010.0; |
| 1118 | the true value is 20000000000000008.0. */ |
| 1119 | if (decpt <= -4 || decpt > 16) |
| 1120 | use_exp = 1; |
| 1121 | break; |
| 1122 | default: |
| 1123 | PyErr_BadInternalCall(); |
| 1124 | goto exit; |
| 1125 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1126 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1127 | /* if using an exponent, reset decimal point position to 1 and adjust |
| 1128 | exponent accordingly.*/ |
| 1129 | if (use_exp) { |
Victor Stinner | 7b25135 | 2013-06-24 23:37:40 +0200 | [diff] [blame] | 1130 | exp = (int)decpt - 1; |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1131 | decpt = 1; |
| 1132 | } |
| 1133 | /* ensure vdigits_start < decpt <= vdigits_end, or vdigits_start < |
| 1134 | decpt < vdigits_end if add_dot_0_if_integer and no exponent */ |
| 1135 | vdigits_start = decpt <= 0 ? decpt-1 : 0; |
| 1136 | if (!use_exp && add_dot_0_if_integer) |
| 1137 | vdigits_end = vdigits_end > decpt ? vdigits_end : decpt + 1; |
| 1138 | else |
| 1139 | vdigits_end = vdigits_end > decpt ? vdigits_end : decpt; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1140 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1141 | /* double check inequalities */ |
| 1142 | assert(vdigits_start <= 0 && |
| 1143 | 0 <= digits_len && |
| 1144 | digits_len <= vdigits_end); |
| 1145 | /* decimal point should be in (vdigits_start, vdigits_end] */ |
| 1146 | assert(vdigits_start < decpt && decpt <= vdigits_end); |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1147 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1148 | /* Compute an upper bound how much memory we need. This might be a few |
| 1149 | chars too long, but no big deal. */ |
| 1150 | bufsize = |
| 1151 | /* sign, decimal point and trailing 0 byte */ |
| 1152 | 3 + |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1153 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1154 | /* total digit count (including zero padding on both sides) */ |
| 1155 | (vdigits_end - vdigits_start) + |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1156 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1157 | /* exponent "e+100", max 3 numerical digits */ |
| 1158 | (use_exp ? 5 : 0); |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1159 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1160 | /* Now allocate the memory and initialize p to point to the start of |
| 1161 | it. */ |
| 1162 | buf = (char *)PyMem_Malloc(bufsize); |
| 1163 | if (buf == NULL) { |
| 1164 | PyErr_NoMemory(); |
| 1165 | goto exit; |
| 1166 | } |
| 1167 | p = buf; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1168 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1169 | /* Add a negative sign if negative, and a plus sign if non-negative |
| 1170 | and always_add_sign is true. */ |
| 1171 | if (sign == 1) |
| 1172 | *p++ = '-'; |
| 1173 | else if (always_add_sign) |
| 1174 | *p++ = '+'; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1175 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1176 | /* note that exactly one of the three 'if' conditions is true, |
| 1177 | so we include exactly one decimal point */ |
| 1178 | /* Zero padding on left of digit string */ |
| 1179 | if (decpt <= 0) { |
| 1180 | memset(p, '0', decpt-vdigits_start); |
| 1181 | p += decpt - vdigits_start; |
| 1182 | *p++ = '.'; |
| 1183 | memset(p, '0', 0-decpt); |
| 1184 | p += 0-decpt; |
| 1185 | } |
| 1186 | else { |
| 1187 | memset(p, '0', 0-vdigits_start); |
| 1188 | p += 0 - vdigits_start; |
| 1189 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1190 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1191 | /* Digits, with included decimal point */ |
| 1192 | if (0 < decpt && decpt <= digits_len) { |
| 1193 | strncpy(p, digits, decpt-0); |
| 1194 | p += decpt-0; |
| 1195 | *p++ = '.'; |
| 1196 | strncpy(p, digits+decpt, digits_len-decpt); |
| 1197 | p += digits_len-decpt; |
| 1198 | } |
| 1199 | else { |
| 1200 | strncpy(p, digits, digits_len); |
| 1201 | p += digits_len; |
| 1202 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1203 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1204 | /* And zeros on the right */ |
| 1205 | if (digits_len < decpt) { |
| 1206 | memset(p, '0', decpt-digits_len); |
| 1207 | p += decpt-digits_len; |
| 1208 | *p++ = '.'; |
| 1209 | memset(p, '0', vdigits_end-decpt); |
| 1210 | p += vdigits_end-decpt; |
| 1211 | } |
| 1212 | else { |
| 1213 | memset(p, '0', vdigits_end-digits_len); |
| 1214 | p += vdigits_end-digits_len; |
| 1215 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1216 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1217 | /* Delete a trailing decimal pt unless using alternative formatting. */ |
| 1218 | if (p[-1] == '.' && !use_alt_formatting) |
| 1219 | p--; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1220 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1221 | /* Now that we've done zero padding, add an exponent if needed. */ |
| 1222 | if (use_exp) { |
| 1223 | *p++ = float_strings[OFS_E][0]; |
| 1224 | exp_len = sprintf(p, "%+.02d", exp); |
| 1225 | p += exp_len; |
| 1226 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1227 | exit: |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1228 | if (buf) { |
| 1229 | *p = '\0'; |
| 1230 | /* It's too late if this fails, as we've already stepped on |
| 1231 | memory that isn't ours. But it's an okay debugging test. */ |
| 1232 | assert(p-buf < bufsize); |
| 1233 | } |
| 1234 | if (digits) |
| 1235 | _Py_dg_freedtoa(digits); |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1236 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1237 | return buf; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1238 | } |
| 1239 | |
| 1240 | |
| 1241 | PyAPI_FUNC(char *) PyOS_double_to_string(double val, |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1242 | char format_code, |
| 1243 | int precision, |
| 1244 | int flags, |
| 1245 | int *type) |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1246 | { |
Serhiy Storchaka | 2d06e84 | 2015-12-25 19:53:18 +0200 | [diff] [blame] | 1247 | const char * const *float_strings = lc_float_strings; |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1248 | int mode; |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1249 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1250 | /* Validate format_code, and map upper and lower case. Compute the |
| 1251 | mode and make any adjustments as needed. */ |
| 1252 | switch (format_code) { |
| 1253 | /* exponent */ |
| 1254 | case 'E': |
| 1255 | float_strings = uc_float_strings; |
| 1256 | format_code = 'e'; |
| 1257 | /* Fall through. */ |
| 1258 | case 'e': |
| 1259 | mode = 2; |
| 1260 | precision++; |
| 1261 | break; |
Eric Smith | 193125a | 2009-04-16 22:08:31 +0000 | [diff] [blame] | 1262 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1263 | /* fixed */ |
| 1264 | case 'F': |
| 1265 | float_strings = uc_float_strings; |
| 1266 | format_code = 'f'; |
| 1267 | /* Fall through. */ |
| 1268 | case 'f': |
| 1269 | mode = 3; |
| 1270 | break; |
Eric Smith | 193125a | 2009-04-16 22:08:31 +0000 | [diff] [blame] | 1271 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1272 | /* general */ |
| 1273 | case 'G': |
| 1274 | float_strings = uc_float_strings; |
| 1275 | format_code = 'g'; |
| 1276 | /* Fall through. */ |
| 1277 | case 'g': |
| 1278 | mode = 2; |
| 1279 | /* precision 0 makes no sense for 'g' format; interpret as 1 */ |
| 1280 | if (precision == 0) |
| 1281 | precision = 1; |
| 1282 | break; |
Eric Smith | 193125a | 2009-04-16 22:08:31 +0000 | [diff] [blame] | 1283 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1284 | /* repr format */ |
| 1285 | case 'r': |
| 1286 | mode = 0; |
| 1287 | /* Supplied precision is unused, must be 0. */ |
| 1288 | if (precision != 0) { |
| 1289 | PyErr_BadInternalCall(); |
| 1290 | return NULL; |
| 1291 | } |
| 1292 | break; |
Eric Smith | 193125a | 2009-04-16 22:08:31 +0000 | [diff] [blame] | 1293 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1294 | default: |
| 1295 | PyErr_BadInternalCall(); |
| 1296 | return NULL; |
| 1297 | } |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1298 | |
Antoine Pitrou | f95a1b3 | 2010-05-09 15:52:27 +0000 | [diff] [blame] | 1299 | return format_float_short(val, format_code, mode, precision, |
| 1300 | flags & Py_DTSF_SIGN, |
| 1301 | flags & Py_DTSF_ADD_DOT_0, |
| 1302 | flags & Py_DTSF_ALT, |
| 1303 | float_strings, type); |
Eric Smith | 0923d1d | 2009-04-16 20:16:10 +0000 | [diff] [blame] | 1304 | } |
| 1305 | #endif /* ifdef PY_NO_SHORT_FLOAT_REPR */ |