Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 1 | /* implements the string, long, and float formatters. that is, |
| 2 | string.__format__, etc. */ |
| 3 | |
| 4 | /* Before including this, you must include either: |
| 5 | stringlib/unicodedefs.h |
| 6 | stringlib/stringdefs.h |
| 7 | |
| 8 | Also, you should define the names: |
| 9 | FORMAT_STRING |
| 10 | FORMAT_LONG |
| 11 | FORMAT_FLOAT |
| 12 | to be whatever you want the public names of these functions to |
| 13 | be. These are the only non-static functions defined here. |
| 14 | */ |
| 15 | |
| 16 | /* |
| 17 | get_integer consumes 0 or more decimal digit characters from an |
| 18 | input string, updates *result with the corresponding positive |
| 19 | integer, and returns the number of digits consumed. |
| 20 | |
| 21 | returns -1 on error. |
| 22 | */ |
| 23 | static int |
| 24 | get_integer(STRINGLIB_CHAR **ptr, STRINGLIB_CHAR *end, |
| 25 | Py_ssize_t *result) |
| 26 | { |
| 27 | Py_ssize_t accumulator, digitval, oldaccumulator; |
| 28 | int numdigits; |
| 29 | accumulator = numdigits = 0; |
| 30 | for (;;(*ptr)++, numdigits++) { |
| 31 | if (*ptr >= end) |
| 32 | break; |
| 33 | digitval = STRINGLIB_TODECIMAL(**ptr); |
| 34 | if (digitval < 0) |
| 35 | break; |
| 36 | /* |
| 37 | This trick was copied from old Unicode format code. It's cute, |
| 38 | but would really suck on an old machine with a slow divide |
| 39 | implementation. Fortunately, in the normal case we do not |
| 40 | expect too many digits. |
| 41 | */ |
| 42 | oldaccumulator = accumulator; |
| 43 | accumulator *= 10; |
| 44 | if ((accumulator+10)/10 != oldaccumulator+1) { |
| 45 | PyErr_Format(PyExc_ValueError, |
| 46 | "Too many decimal digits in format string"); |
| 47 | return -1; |
| 48 | } |
| 49 | accumulator += digitval; |
| 50 | } |
| 51 | *result = accumulator; |
| 52 | return numdigits; |
| 53 | } |
| 54 | |
| 55 | /************************************************************************/ |
| 56 | /*********** standard format specifier parsing **************************/ |
| 57 | /************************************************************************/ |
| 58 | |
| 59 | /* returns true if this character is a specifier alignment token */ |
| 60 | Py_LOCAL_INLINE(int) |
| 61 | is_alignment_token(STRINGLIB_CHAR c) |
| 62 | { |
| 63 | switch (c) { |
| 64 | case '<': case '>': case '=': case '^': |
| 65 | return 1; |
| 66 | default: |
| 67 | return 0; |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | /* returns true if this character is a sign element */ |
| 72 | Py_LOCAL_INLINE(int) |
| 73 | is_sign_element(STRINGLIB_CHAR c) |
| 74 | { |
| 75 | switch (c) { |
| 76 | case ' ': case '+': case '-': case '(': |
| 77 | return 1; |
| 78 | default: |
| 79 | return 0; |
| 80 | } |
| 81 | } |
| 82 | |
| 83 | |
| 84 | typedef struct { |
| 85 | STRINGLIB_CHAR fill_char; |
| 86 | STRINGLIB_CHAR align; |
| 87 | STRINGLIB_CHAR sign; |
| 88 | Py_ssize_t width; |
| 89 | Py_ssize_t precision; |
| 90 | STRINGLIB_CHAR type; |
| 91 | } InternalFormatSpec; |
| 92 | |
| 93 | /* |
| 94 | ptr points to the start of the format_spec, end points just past its end. |
| 95 | fills in format with the parsed information. |
| 96 | returns 1 on success, 0 on failure. |
| 97 | if failure, sets the exception |
| 98 | */ |
| 99 | static int |
| 100 | parse_internal_render_format_spec(PyObject *format_spec, |
| 101 | InternalFormatSpec *format, |
| 102 | char default_type) |
| 103 | { |
| 104 | STRINGLIB_CHAR *ptr = STRINGLIB_STR(format_spec); |
| 105 | STRINGLIB_CHAR *end = ptr + STRINGLIB_LEN(format_spec); |
| 106 | |
| 107 | /* end-ptr is used throughout this code to specify the length of |
| 108 | the input string */ |
| 109 | |
| 110 | Py_ssize_t specified_width; |
| 111 | |
| 112 | format->fill_char = '\0'; |
| 113 | format->align = '\0'; |
| 114 | format->sign = '\0'; |
| 115 | format->width = -1; |
| 116 | format->precision = -1; |
| 117 | format->type = default_type; |
| 118 | |
| 119 | /* If the second char is an alignment token, |
| 120 | then parse the fill char */ |
| 121 | if (end-ptr >= 2 && is_alignment_token(ptr[1])) { |
| 122 | format->align = ptr[1]; |
| 123 | format->fill_char = ptr[0]; |
| 124 | ptr += 2; |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 125 | } |
| 126 | else if (end-ptr >= 1 && is_alignment_token(ptr[0])) { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 127 | format->align = ptr[0]; |
| 128 | ptr++; |
| 129 | } |
| 130 | |
| 131 | /* Parse the various sign options */ |
| 132 | if (end-ptr >= 1 && is_sign_element(ptr[0])) { |
| 133 | format->sign = ptr[0]; |
| 134 | ptr++; |
| 135 | if (end-ptr >= 1 && ptr[0] == ')') { |
| 136 | ptr++; |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | /* The special case for 0-padding (backwards compat) */ |
| 141 | if (format->fill_char == '\0' && |
| 142 | end-ptr >= 1 && ptr[0] == '0') { |
| 143 | format->fill_char = '0'; |
| 144 | if (format->align == '\0') { |
| 145 | format->align = '='; |
| 146 | } |
| 147 | ptr++; |
| 148 | } |
| 149 | |
| 150 | /* XXX add error checking */ |
| 151 | specified_width = get_integer(&ptr, end, &format->width); |
| 152 | |
| 153 | /* if specified_width is 0, we didn't consume any characters for |
| 154 | the width. in that case, reset the width to -1, because |
| 155 | get_integer() will have set it to zero */ |
| 156 | if (specified_width == 0) { |
| 157 | format->width = -1; |
| 158 | } |
| 159 | |
| 160 | /* Parse field precision */ |
| 161 | if (end-ptr && ptr[0] == '.') { |
| 162 | ptr++; |
| 163 | |
| 164 | /* XXX add error checking */ |
| 165 | specified_width = get_integer(&ptr, end, &format->precision); |
| 166 | |
| 167 | /* not having a precision after a dot is an error */ |
| 168 | if (specified_width == 0) { |
| 169 | PyErr_Format(PyExc_ValueError, |
| 170 | "Format specifier missing precision"); |
| 171 | return 0; |
| 172 | } |
| 173 | |
| 174 | } |
| 175 | |
| 176 | /* Finally, parse the type field */ |
| 177 | |
| 178 | if (end-ptr > 1) { |
| 179 | /* invalid conversion spec */ |
| 180 | PyErr_Format(PyExc_ValueError, "Invalid conversion specification"); |
| 181 | return 0; |
| 182 | } |
| 183 | |
| 184 | if (end-ptr == 1) { |
| 185 | format->type = ptr[0]; |
| 186 | ptr++; |
| 187 | } |
| 188 | |
| 189 | return 1; |
| 190 | } |
| 191 | |
| 192 | |
| 193 | /************************************************************************/ |
| 194 | /*********** common routines for numeric formatting *********************/ |
| 195 | /************************************************************************/ |
| 196 | |
| 197 | /* describes the layout for an integer, see the comment in |
| 198 | _calc_integer_widths() for details */ |
| 199 | typedef struct { |
| 200 | Py_ssize_t n_lpadding; |
| 201 | Py_ssize_t n_spadding; |
| 202 | Py_ssize_t n_rpadding; |
| 203 | char lsign; |
| 204 | Py_ssize_t n_lsign; |
| 205 | char rsign; |
| 206 | Py_ssize_t n_rsign; |
| 207 | Py_ssize_t n_total; /* just a convenience, it's derivable from the |
| 208 | other fields */ |
| 209 | } NumberFieldWidths; |
| 210 | |
| 211 | /* not all fields of format are used. for example, precision is |
| 212 | unused. should this take discrete params in order to be more clear |
| 213 | about what it does? or is passing a single format parameter easier |
| 214 | and more efficient enough to justify a little obfuscation? */ |
| 215 | static void |
| 216 | calc_number_widths(NumberFieldWidths *r, STRINGLIB_CHAR actual_sign, |
| 217 | Py_ssize_t n_digits, const InternalFormatSpec *format) |
| 218 | { |
| 219 | r->n_lpadding = 0; |
| 220 | r->n_spadding = 0; |
| 221 | r->n_rpadding = 0; |
| 222 | r->lsign = '\0'; |
| 223 | r->n_lsign = 0; |
| 224 | r->rsign = '\0'; |
| 225 | r->n_rsign = 0; |
| 226 | |
| 227 | /* the output will look like: |
| 228 | | | |
| 229 | | <lpadding> <lsign> <spadding> <digits> <rsign> <rpadding> | |
| 230 | | | |
| 231 | |
| 232 | lsign and rsign are computed from format->sign and the actual |
| 233 | sign of the number |
| 234 | |
| 235 | digits is already known |
| 236 | |
| 237 | the total width is either given, or computed from the |
| 238 | actual digits |
| 239 | |
| 240 | only one of lpadding, spadding, and rpadding can be non-zero, |
| 241 | and it's calculated from the width and other fields |
| 242 | */ |
| 243 | |
| 244 | /* compute the various parts we're going to write */ |
| 245 | if (format->sign == '+') { |
| 246 | /* always put a + or - */ |
| 247 | r->n_lsign = 1; |
| 248 | r->lsign = (actual_sign == '-' ? '-' : '+'); |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 249 | } |
| 250 | else if (format->sign == '(') { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 251 | if (actual_sign == '-') { |
| 252 | r->n_lsign = 1; |
| 253 | r->lsign = '('; |
| 254 | r->n_rsign = 1; |
| 255 | r->rsign = ')'; |
| 256 | } |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 257 | } |
| 258 | else if (format->sign == ' ') { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 259 | r->n_lsign = 1; |
| 260 | r->lsign = (actual_sign == '-' ? '-' : ' '); |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 261 | } |
| 262 | else { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 263 | /* non specified, or the default (-) */ |
| 264 | if (actual_sign == '-') { |
| 265 | r->n_lsign = 1; |
| 266 | r->lsign = '-'; |
| 267 | } |
| 268 | } |
| 269 | |
| 270 | /* now the number of padding characters */ |
| 271 | if (format->width == -1) { |
| 272 | /* no padding at all, nothing to do */ |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 273 | } |
| 274 | else { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 275 | /* see if any padding is needed */ |
| 276 | if (r->n_lsign + n_digits + r->n_rsign >= format->width) { |
| 277 | /* no padding needed, we're already bigger than the |
| 278 | requested width */ |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 279 | } |
| 280 | else { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 281 | /* determine which of left, space, or right padding is |
| 282 | needed */ |
| 283 | Py_ssize_t padding = format->width - (r->n_lsign + n_digits + r->n_rsign); |
| 284 | if (format->align == '<') |
| 285 | r->n_rpadding = padding; |
| 286 | else if (format->align == '>') |
| 287 | r->n_lpadding = padding; |
| 288 | else if (format->align == '^') { |
| 289 | r->n_lpadding = padding / 2; |
| 290 | r->n_rpadding = padding - r->n_lpadding; |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 291 | } |
| 292 | else |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 293 | /* must be '=' */ |
| 294 | r->n_spadding = padding; |
| 295 | } |
| 296 | } |
| 297 | r->n_total = r->n_lpadding + r->n_lsign + r->n_spadding + |
| 298 | n_digits + r->n_rsign + r->n_rpadding; |
| 299 | } |
| 300 | |
| 301 | /* fill in the non-digit parts of a numbers's string representation, |
| 302 | as determined in _calc_integer_widths(). returns the pointer to |
| 303 | where the digits go. */ |
| 304 | static STRINGLIB_CHAR * |
| 305 | fill_number(STRINGLIB_CHAR *p_buf, const NumberFieldWidths *spec, |
| 306 | Py_ssize_t n_digits, STRINGLIB_CHAR fill_char) |
| 307 | { |
| 308 | STRINGLIB_CHAR* p_digits; |
| 309 | |
| 310 | if (spec->n_lpadding) { |
| 311 | STRINGLIB_FILL(p_buf, fill_char, spec->n_lpadding); |
| 312 | p_buf += spec->n_lpadding; |
| 313 | } |
| 314 | if (spec->n_lsign == 1) { |
| 315 | *p_buf++ = spec->lsign; |
| 316 | } |
| 317 | if (spec->n_spadding) { |
| 318 | STRINGLIB_FILL(p_buf, fill_char, spec->n_spadding); |
| 319 | p_buf += spec->n_spadding; |
| 320 | } |
| 321 | p_digits = p_buf; |
| 322 | p_buf += n_digits; |
| 323 | if (spec->n_rsign == 1) { |
| 324 | *p_buf++ = spec->rsign; |
| 325 | } |
| 326 | if (spec->n_rpadding) { |
| 327 | STRINGLIB_FILL(p_buf, fill_char, spec->n_rpadding); |
| 328 | p_buf += spec->n_rpadding; |
| 329 | } |
| 330 | return p_digits; |
| 331 | } |
| 332 | |
| 333 | /************************************************************************/ |
| 334 | /*********** string formatting ******************************************/ |
| 335 | /************************************************************************/ |
| 336 | |
| 337 | static PyObject * |
| 338 | format_string_internal(PyObject *value, const InternalFormatSpec *format) |
| 339 | { |
| 340 | Py_ssize_t width; /* total field width */ |
| 341 | Py_ssize_t lpad; |
| 342 | STRINGLIB_CHAR *dst; |
| 343 | STRINGLIB_CHAR *src = STRINGLIB_STR(value); |
| 344 | Py_ssize_t len = STRINGLIB_LEN(value); |
| 345 | PyObject *result = NULL; |
| 346 | |
| 347 | /* sign is not allowed on strings */ |
| 348 | if (format->sign != '\0') { |
| 349 | PyErr_SetString(PyExc_ValueError, |
| 350 | "Sign not allowed in string format specifier"); |
| 351 | goto done; |
| 352 | } |
| 353 | |
| 354 | /* '=' alignment not allowed on strings */ |
| 355 | if (format->align == '=') { |
| 356 | PyErr_SetString(PyExc_ValueError, |
| 357 | "'=' alignment not allowed " |
| 358 | "in string format specifier"); |
| 359 | goto done; |
| 360 | } |
| 361 | |
| 362 | /* if precision is specified, output no more that format.precision |
| 363 | characters */ |
| 364 | if (format->precision >= 0 && len >= format->precision) { |
| 365 | len = format->precision; |
| 366 | } |
| 367 | |
| 368 | if (format->width >= 0) { |
| 369 | width = format->width; |
| 370 | |
| 371 | /* but use at least len characters */ |
| 372 | if (len > width) { |
| 373 | width = len; |
| 374 | } |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 375 | } |
| 376 | else { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 377 | /* not specified, use all of the chars and no more */ |
| 378 | width = len; |
| 379 | } |
| 380 | |
| 381 | /* allocate the resulting string */ |
| 382 | result = STRINGLIB_NEW(NULL, width); |
| 383 | if (result == NULL) |
| 384 | goto done; |
| 385 | |
| 386 | /* now write into that space */ |
| 387 | dst = STRINGLIB_STR(result); |
| 388 | |
| 389 | /* figure out how much leading space we need, based on the |
| 390 | aligning */ |
| 391 | if (format->align == '>') |
| 392 | lpad = width - len; |
| 393 | else if (format->align == '^') |
| 394 | lpad = (width - len) / 2; |
| 395 | else |
| 396 | lpad = 0; |
| 397 | |
| 398 | /* if right aligning, increment the destination allow space on the |
| 399 | left */ |
| 400 | memcpy(dst + lpad, src, len * sizeof(STRINGLIB_CHAR)); |
| 401 | |
| 402 | /* do any padding */ |
| 403 | if (width > len) { |
| 404 | STRINGLIB_CHAR fill_char = format->fill_char; |
| 405 | if (fill_char == '\0') { |
| 406 | /* use the default, if not specified */ |
| 407 | fill_char = ' '; |
| 408 | } |
| 409 | |
| 410 | /* pad on left */ |
| 411 | if (lpad) |
| 412 | STRINGLIB_FILL(dst, fill_char, lpad); |
| 413 | |
| 414 | /* pad on right */ |
| 415 | if (width - len - lpad) |
| 416 | STRINGLIB_FILL(dst + len + lpad, fill_char, width - len - lpad); |
| 417 | } |
| 418 | |
| 419 | done: |
| 420 | return result; |
| 421 | } |
| 422 | |
| 423 | |
| 424 | /************************************************************************/ |
| 425 | /*********** long formatting ********************************************/ |
| 426 | /************************************************************************/ |
| 427 | |
| 428 | static PyObject * |
| 429 | format_long_internal(PyObject *value, const InternalFormatSpec *format) |
| 430 | { |
| 431 | PyObject *result = NULL; |
| 432 | int total_leading_chars_to_skip = 0; /* also includes sign, if |
| 433 | present */ |
| 434 | STRINGLIB_CHAR sign = '\0'; |
| 435 | STRINGLIB_CHAR *p; |
| 436 | Py_ssize_t n_digits; /* count of digits need from the computed |
| 437 | string */ |
| 438 | Py_ssize_t len; |
| 439 | Py_ssize_t tmp; |
| 440 | NumberFieldWidths spec; |
| 441 | long x; |
| 442 | |
| 443 | /* no precision allowed on integers */ |
| 444 | if (format->precision != -1) { |
| 445 | PyErr_SetString(PyExc_ValueError, |
| 446 | "Precision not allowed in integer format specifier"); |
| 447 | goto done; |
| 448 | } |
| 449 | |
| 450 | |
| 451 | /* special case for character formatting */ |
| 452 | if (format->type == 'c') { |
| 453 | /* error to specify a sign */ |
| 454 | if (format->sign != '\0') { |
| 455 | PyErr_SetString(PyExc_ValueError, |
| 456 | "Sign not allowed with integer" |
| 457 | " format specifier 'c'"); |
| 458 | goto done; |
| 459 | } |
| 460 | |
| 461 | /* taken from unicodeobject.c formatchar() */ |
| 462 | /* Integer input truncated to a character */ |
| 463 | x = PyInt_AsLong(value); |
| 464 | if (x == -1 && PyErr_Occurred()) |
| 465 | goto done; |
| 466 | #ifdef Py_UNICODE_WIDE |
| 467 | if (x < 0 || x > 0x10ffff) { |
| 468 | PyErr_SetString(PyExc_OverflowError, |
| 469 | "%c arg not in range(0x110000) " |
| 470 | "(wide Python build)"); |
| 471 | goto done; |
| 472 | } |
| 473 | #else |
| 474 | if (x < 0 || x > 0xffff) { |
| 475 | PyErr_SetString(PyExc_OverflowError, |
| 476 | "%c arg not in range(0x10000) " |
| 477 | "(narrow Python build)"); |
| 478 | goto done; |
| 479 | } |
| 480 | #endif |
| 481 | result = STRINGLIB_NEW(NULL, 1); |
| 482 | if (result == NULL) |
| 483 | goto done; |
| 484 | p = STRINGLIB_STR(result); |
| 485 | p[0] = (Py_UNICODE) x; |
| 486 | n_digits = len = 1; |
Eric Smith | 0cb431c | 2007-08-28 01:07:27 +0000 | [diff] [blame^] | 487 | } |
| 488 | else { |
Eric Smith | 8c66326 | 2007-08-25 02:26:07 +0000 | [diff] [blame] | 489 | int base; |
| 490 | int format_leading_chars_to_skip; /* characters added by |
| 491 | PyNumber_ToBase that we |
| 492 | want to skip over. |
| 493 | instead of using them, |
| 494 | we'll compute our |
| 495 | own. */ |
| 496 | /* compute the base and how many characters will be added by |
| 497 | PyNumber_ToBase */ |
| 498 | switch (format->type) { |
| 499 | case 'b': |
| 500 | base = 2; |
| 501 | format_leading_chars_to_skip = 2; /* 0b */ |
| 502 | break; |
| 503 | case 'o': |
| 504 | base = 8; |
| 505 | format_leading_chars_to_skip = 2; /* 0o */ |
| 506 | break; |
| 507 | case 'x': |
| 508 | case 'X': |
| 509 | base = 16; |
| 510 | format_leading_chars_to_skip = 2; /* 0x */ |
| 511 | break; |
| 512 | default: /* shouldn't be needed, but stops a compiler warning */ |
| 513 | case 'd': |
| 514 | base = 10; |
| 515 | format_leading_chars_to_skip = 0; |
| 516 | break; |
| 517 | } |
| 518 | |
| 519 | /* do the hard part, converting to a string in a given base */ |
| 520 | result = PyNumber_ToBase(value, base); |
| 521 | if (result == NULL) |
| 522 | goto done; |
| 523 | |
| 524 | n_digits = STRINGLIB_LEN(result); |
| 525 | len = n_digits; |
| 526 | p = STRINGLIB_STR(result); |
| 527 | |
| 528 | /* if X, convert to uppercase */ |
| 529 | if (format->type == 'X') |
| 530 | for (tmp = 0; tmp < len; tmp++) |
| 531 | p[tmp] = STRINGLIB_TOUPPER(p[tmp]); |
| 532 | |
| 533 | /* is a sign character present in the output? if so, remember it |
| 534 | and skip it */ |
| 535 | sign = p[0]; |
| 536 | if (sign == '-') { |
| 537 | total_leading_chars_to_skip += 1; |
| 538 | n_digits--; |
| 539 | } |
| 540 | |
| 541 | /* skip over the leading digits (0x, 0b, etc.) */ |
| 542 | assert(n_digits >= format_leading_chars_to_skip + 1); |
| 543 | n_digits -= format_leading_chars_to_skip; |
| 544 | total_leading_chars_to_skip += format_leading_chars_to_skip; |
| 545 | } |
| 546 | |
| 547 | calc_number_widths(&spec, sign, n_digits, format); |
| 548 | |
| 549 | /* if the buffer is getting bigger, realloc it. if it's getting |
| 550 | smaller, don't realloc because we need to move the results |
| 551 | around first. realloc after we've done that */ |
| 552 | |
| 553 | if (spec.n_total > len) { |
| 554 | if (STRINGLIB_RESIZE(&result, spec.n_total) < 0) |
| 555 | goto done; |
| 556 | /* recalc, because string might have moved */ |
| 557 | p = STRINGLIB_STR(result); |
| 558 | } |
| 559 | |
| 560 | /* copy the characters into position first, since we're going to |
| 561 | overwrite some of that space */ |
| 562 | /* we need to move if the number of left padding in the output is |
| 563 | different from the number of characters we need to skip */ |
| 564 | if ((spec.n_lpadding + spec.n_lsign + spec.n_spadding) != |
| 565 | total_leading_chars_to_skip) { |
| 566 | memmove(p + (spec.n_lpadding + spec.n_lsign + spec.n_spadding), |
| 567 | p + total_leading_chars_to_skip, |
| 568 | n_digits * sizeof(STRINGLIB_CHAR)); |
| 569 | } |
| 570 | |
| 571 | /* now fill in the non-digit parts */ |
| 572 | fill_number(p, &spec, n_digits, |
| 573 | format->fill_char == '\0' ? ' ' : format->fill_char); |
| 574 | |
| 575 | /* if we're getting smaller, realloc now */ |
| 576 | if (spec.n_total < len) { |
| 577 | if (STRINGLIB_RESIZE(&result, spec.n_total) < 0) |
| 578 | goto done; |
| 579 | } |
| 580 | |
| 581 | done: |
| 582 | return result; |
| 583 | } |
| 584 | |
| 585 | |
| 586 | /************************************************************************/ |
| 587 | /*********** float formatting *******************************************/ |
| 588 | /************************************************************************/ |
| 589 | |
| 590 | /* taken from unicodeobject.c */ |
| 591 | static Py_ssize_t |
| 592 | strtounicode(Py_UNICODE *buffer, const char *charbuffer) |
| 593 | { |
| 594 | register Py_ssize_t i; |
| 595 | Py_ssize_t len = strlen(charbuffer); |
| 596 | for (i = len - 1; i >= 0; i--) |
| 597 | buffer[i] = (Py_UNICODE) charbuffer[i]; |
| 598 | |
| 599 | return len; |
| 600 | } |
| 601 | |
| 602 | /* the callback function to call to do the actual float formatting. |
| 603 | it matches the definition of PyOS_ascii_formatd */ |
| 604 | typedef char* |
| 605 | (*DoubleSnprintfFunction)(char *buffer, size_t buf_len, |
| 606 | const char *format, double d); |
| 607 | |
| 608 | /* just a wrapper to make PyOS_snprintf look like DoubleSnprintfFunction */ |
| 609 | static char* |
| 610 | snprintf_double(char *buffer, size_t buf_len, const char *format, double d) |
| 611 | { |
| 612 | PyOS_snprintf(buffer, buf_len, format, d); |
| 613 | return NULL; |
| 614 | } |
| 615 | |
| 616 | /* see FORMATBUFLEN in unicodeobject.c */ |
| 617 | #define FLOAT_FORMATBUFLEN 120 |
| 618 | |
| 619 | /* much of this is taken from unicodeobject.c */ |
| 620 | /* use type instead of format->type, so that it can be overridden by |
| 621 | format_number() */ |
| 622 | static PyObject * |
| 623 | _format_float(STRINGLIB_CHAR type, PyObject *value, |
| 624 | const InternalFormatSpec *format, |
| 625 | DoubleSnprintfFunction snprintf) |
| 626 | { |
| 627 | /* fmt = '%.' + `prec` + `type` + '%%' |
| 628 | worst case length = 2 + 10 (len of INT_MAX) + 1 + 2 = 15 (use 20)*/ |
| 629 | char fmt[20]; |
| 630 | |
| 631 | /* taken from unicodeobject.c */ |
| 632 | /* Worst case length calc to ensure no buffer overrun: |
| 633 | |
| 634 | 'g' formats: |
| 635 | fmt = %#.<prec>g |
| 636 | buf = '-' + [0-9]*prec + '.' + 'e+' + (longest exp |
| 637 | for any double rep.) |
| 638 | len = 1 + prec + 1 + 2 + 5 = 9 + prec |
| 639 | |
| 640 | 'f' formats: |
| 641 | buf = '-' + [0-9]*x + '.' + [0-9]*prec (with x < 50) |
| 642 | len = 1 + 50 + 1 + prec = 52 + prec |
| 643 | |
| 644 | If prec=0 the effective precision is 1 (the leading digit is |
| 645 | always given), therefore increase the length by one. |
| 646 | |
| 647 | */ |
| 648 | char charbuf[FLOAT_FORMATBUFLEN]; |
| 649 | Py_ssize_t n_digits; |
| 650 | double x; |
| 651 | Py_ssize_t precision = format->precision; |
| 652 | PyObject *result = NULL; |
| 653 | STRINGLIB_CHAR sign; |
| 654 | char* trailing = ""; |
| 655 | STRINGLIB_CHAR *p; |
| 656 | NumberFieldWidths spec; |
| 657 | |
| 658 | #if STRINGLIB_IS_UNICODE |
| 659 | Py_UNICODE unicodebuf[FLOAT_FORMATBUFLEN]; |
| 660 | #endif |
| 661 | |
| 662 | /* first, do the conversion as 8-bit chars, using the platform's |
| 663 | snprintf. then, if needed, convert to unicode. */ |
| 664 | |
| 665 | /* 'F' is the same as 'f', per the PEP */ |
| 666 | if (type == 'F') |
| 667 | type = 'f'; |
| 668 | |
| 669 | x = PyFloat_AsDouble(value); |
| 670 | |
| 671 | if (x == -1.0 && PyErr_Occurred()) |
| 672 | goto done; |
| 673 | |
| 674 | if (type == '%') { |
| 675 | type = 'f'; |
| 676 | x *= 100; |
| 677 | trailing = "%"; |
| 678 | } |
| 679 | |
| 680 | if (precision < 0) |
| 681 | precision = 6; |
| 682 | if (type == 'f' && (fabs(x) / 1e25) >= 1e25) |
| 683 | type = 'g'; |
| 684 | |
| 685 | /* cast "type", because if we're in unicode we need to pass a |
| 686 | 8-bit char. this is safe, because we've restricted what "type" |
| 687 | can be */ |
| 688 | PyOS_snprintf(fmt, sizeof(fmt), "%%.%zd%c", precision, (char)type); |
| 689 | |
| 690 | /* call the passed in function to do the actual formatting */ |
| 691 | snprintf(charbuf, sizeof(charbuf), fmt, x); |
| 692 | |
| 693 | /* adding trailing to fmt with PyOS_snprintf doesn't work, not |
| 694 | sure why. we'll just concatentate it here, no harm done. we |
| 695 | know we can't have a buffer overflow from the fmt size |
| 696 | analysis */ |
| 697 | strcat(charbuf, trailing); |
| 698 | |
| 699 | /* rather than duplicate the code for snprintf for both unicode |
| 700 | and 8 bit strings, we just use the 8 bit version and then |
| 701 | convert to unicode in a separate code path. that's probably |
| 702 | the lesser of 2 evils. */ |
| 703 | #if STRINGLIB_IS_UNICODE |
| 704 | n_digits = strtounicode(unicodebuf, charbuf); |
| 705 | p = unicodebuf; |
| 706 | #else |
| 707 | /* compute the length. I believe this is done because the return |
| 708 | value from snprintf above is unreliable */ |
| 709 | n_digits = strlen(charbuf); |
| 710 | p = charbuf; |
| 711 | #endif |
| 712 | |
| 713 | /* is a sign character present in the output? if so, remember it |
| 714 | and skip it */ |
| 715 | sign = p[0]; |
| 716 | if (sign == '-') { |
| 717 | p++; |
| 718 | n_digits--; |
| 719 | } |
| 720 | |
| 721 | calc_number_widths(&spec, sign, n_digits, format); |
| 722 | |
| 723 | /* allocate a string with enough space */ |
| 724 | result = STRINGLIB_NEW(NULL, spec.n_total); |
| 725 | if (result == NULL) |
| 726 | goto done; |
| 727 | |
| 728 | /* fill in the non-digit parts */ |
| 729 | fill_number(STRINGLIB_STR(result), &spec, n_digits, |
| 730 | format->fill_char == '\0' ? ' ' : format->fill_char); |
| 731 | |
| 732 | /* fill in the digit parts */ |
| 733 | memmove(STRINGLIB_STR(result) + (spec.n_lpadding + spec.n_lsign + spec.n_spadding), |
| 734 | p, |
| 735 | n_digits * sizeof(STRINGLIB_CHAR)); |
| 736 | |
| 737 | done: |
| 738 | return result; |
| 739 | } |
| 740 | |
| 741 | static PyObject * |
| 742 | format_float_internal(PyObject *value, const InternalFormatSpec *format) |
| 743 | { |
| 744 | if (format->type == 'n') |
| 745 | return _format_float('f', value, format, snprintf_double); |
| 746 | else |
| 747 | return _format_float(format->type, value, format, PyOS_ascii_formatd); |
| 748 | } |
| 749 | |
| 750 | /************************************************************************/ |
| 751 | /*********** built in formatters ****************************************/ |
| 752 | /************************************************************************/ |
| 753 | |
| 754 | PyObject * |
| 755 | FORMAT_STRING(PyObject* value, PyObject* args) |
| 756 | { |
| 757 | PyObject *format_spec; |
| 758 | PyObject *tmp = NULL; |
| 759 | PyObject *result = NULL; |
| 760 | InternalFormatSpec format; |
| 761 | |
| 762 | if (!PyArg_ParseTuple(args, "O:__format__", &format_spec)) |
| 763 | goto done; |
| 764 | if (!STRINGLIB_CHECK(format_spec)) { |
| 765 | PyErr_SetString(PyExc_TypeError, STRINGLIB_TYPE_NAME " object required"); |
| 766 | goto done; |
| 767 | } |
| 768 | |
| 769 | /* check for the special case of zero length format spec, make |
| 770 | it equivalent to str(value) */ |
| 771 | if (STRINGLIB_LEN(format_spec) == 0) { |
| 772 | result = STRINGLIB_TOSTR(value); |
| 773 | goto done; |
| 774 | } |
| 775 | |
| 776 | /* parse the format_spec */ |
| 777 | if (!parse_internal_render_format_spec(format_spec, &format, 's')) |
| 778 | goto done; |
| 779 | |
| 780 | /* type conversion? */ |
| 781 | switch (format.type) { |
| 782 | case 's': |
| 783 | /* no type conversion needed, already a string. do the formatting */ |
| 784 | result = format_string_internal(value, &format); |
| 785 | break; |
| 786 | #if 0 |
| 787 | case 'b': |
| 788 | case 'c': |
| 789 | case 'd': |
| 790 | case 'o': |
| 791 | case 'x': |
| 792 | case 'X': |
| 793 | /* convert to integer */ |
| 794 | /* XXX: make a stringlib function to do this when backporting, |
| 795 | since FromUnicode differs from FromString */ |
| 796 | tmp = PyLong_FromUnicode(STRINGLIB_STR(value), STRINGLIB_LEN(value), 0); |
| 797 | if (tmp == NULL) |
| 798 | goto done; |
| 799 | result = format_long_internal(tmp, &format); |
| 800 | break; |
| 801 | |
| 802 | case 'e': |
| 803 | case 'E': |
| 804 | case 'f': |
| 805 | case 'F': |
| 806 | case 'g': |
| 807 | case 'G': |
| 808 | case 'n': |
| 809 | case '%': |
| 810 | /* convert to float */ |
| 811 | tmp = PyFloat_FromString(value); |
| 812 | if (tmp == NULL) |
| 813 | goto done; |
| 814 | result = format_float_internal(tmp, &format); |
| 815 | break; |
| 816 | #endif |
| 817 | default: |
| 818 | /* unknown */ |
| 819 | PyErr_Format(PyExc_ValueError, "Unknown conversion type %c", |
| 820 | format.type); |
| 821 | goto done; |
| 822 | } |
| 823 | |
| 824 | done: |
| 825 | Py_XDECREF(tmp); |
| 826 | return result; |
| 827 | } |
| 828 | |
| 829 | PyObject * |
| 830 | FORMAT_LONG(PyObject* value, PyObject* args) |
| 831 | { |
| 832 | PyObject *format_spec; |
| 833 | PyObject *result = NULL; |
| 834 | PyObject *tmp = NULL; |
| 835 | InternalFormatSpec format; |
| 836 | |
| 837 | if (!PyArg_ParseTuple(args, "O:__format__", &format_spec)) |
| 838 | goto done; |
| 839 | if (!STRINGLIB_CHECK(format_spec)) { |
| 840 | PyErr_SetString(PyExc_TypeError, STRINGLIB_TYPE_NAME " object required"); |
| 841 | goto done; |
| 842 | } |
| 843 | |
| 844 | /* check for the special case of zero length format spec, make |
| 845 | it equivalent to str(value) */ |
| 846 | if (STRINGLIB_LEN(format_spec) == 0) { |
| 847 | result = STRINGLIB_TOSTR(value); |
| 848 | goto done; |
| 849 | } |
| 850 | |
| 851 | /* parse the format_spec */ |
| 852 | if (!parse_internal_render_format_spec(format_spec, &format, 'd')) |
| 853 | goto done; |
| 854 | |
| 855 | /* type conversion? */ |
| 856 | switch (format.type) { |
| 857 | #if 0 |
| 858 | case 's': |
| 859 | /* convert to string/unicode */ |
| 860 | tmp = STRINGLIB_TOSTR(value); |
| 861 | if (tmp == NULL) |
| 862 | goto done; |
| 863 | result = format_string_internal(tmp, &format); |
| 864 | break; |
| 865 | #endif |
| 866 | case 'b': |
| 867 | case 'c': |
| 868 | case 'd': |
| 869 | case 'o': |
| 870 | case 'x': |
| 871 | case 'X': |
| 872 | /* no type conversion needed, already an int. do the formatting */ |
| 873 | result = format_long_internal(value, &format); |
| 874 | break; |
| 875 | |
| 876 | case 'e': |
| 877 | case 'E': |
| 878 | case 'f': |
| 879 | case 'F': |
| 880 | case 'g': |
| 881 | case 'G': |
| 882 | case 'n': |
| 883 | case '%': |
| 884 | /* convert to float */ |
| 885 | tmp = PyNumber_Float(value); |
| 886 | if (tmp == NULL) |
| 887 | goto done; |
| 888 | result = format_float_internal(value, &format); |
| 889 | break; |
| 890 | |
| 891 | default: |
| 892 | /* unknown */ |
| 893 | PyErr_Format(PyExc_ValueError, "Unknown conversion type %c", |
| 894 | format.type); |
| 895 | goto done; |
| 896 | } |
| 897 | |
| 898 | done: |
| 899 | Py_XDECREF(tmp); |
| 900 | return result; |
| 901 | } |
| 902 | |
| 903 | PyObject * |
| 904 | FORMAT_FLOAT(PyObject *value, PyObject *args) |
| 905 | { |
| 906 | PyObject *format_spec; |
| 907 | PyObject *result = NULL; |
| 908 | PyObject *tmp = NULL; |
| 909 | InternalFormatSpec format; |
| 910 | |
| 911 | if (!PyArg_ParseTuple(args, "O:__format__", &format_spec)) |
| 912 | goto done; |
| 913 | if (!STRINGLIB_CHECK(format_spec)) { |
| 914 | PyErr_SetString(PyExc_TypeError, STRINGLIB_TYPE_NAME " object required"); |
| 915 | goto done; |
| 916 | } |
| 917 | |
| 918 | /* check for the special case of zero length format spec, make |
| 919 | it equivalent to str(value) */ |
| 920 | if (STRINGLIB_LEN(format_spec) == 0) { |
| 921 | result = STRINGLIB_TOSTR(value); |
| 922 | goto done; |
| 923 | } |
| 924 | |
| 925 | /* parse the format_spec */ |
| 926 | if (!parse_internal_render_format_spec(format_spec, &format, 'g')) |
| 927 | goto done; |
| 928 | |
| 929 | /* type conversion? */ |
| 930 | switch (format.type) { |
| 931 | #if 0 |
| 932 | case 's': |
| 933 | /* convert to string/unicode */ |
| 934 | tmp = STRINGLIB_TOSTR(value); |
| 935 | if (tmp == NULL) |
| 936 | goto done; |
| 937 | result = format_string_internal(tmp, &format); |
| 938 | break; |
| 939 | #endif |
| 940 | case 'b': |
| 941 | case 'c': |
| 942 | case 'd': |
| 943 | case 'o': |
| 944 | case 'x': |
| 945 | case 'X': |
| 946 | /* convert to integer */ |
| 947 | tmp = PyNumber_Long(value); |
| 948 | if (tmp == NULL) |
| 949 | goto done; |
| 950 | result = format_long_internal(tmp, &format); |
| 951 | break; |
| 952 | |
| 953 | case 'e': |
| 954 | case 'E': |
| 955 | case 'f': |
| 956 | case 'F': |
| 957 | case 'g': |
| 958 | case 'G': |
| 959 | case 'n': |
| 960 | case '%': |
| 961 | /* no conversion, already a float. do the formatting */ |
| 962 | result = format_float_internal(value, &format); |
| 963 | break; |
| 964 | |
| 965 | default: |
| 966 | /* unknown */ |
| 967 | PyErr_Format(PyExc_ValueError, "Unknown conversion type %c", |
| 968 | format.type); |
| 969 | goto done; |
| 970 | } |
| 971 | |
| 972 | done: |
| 973 | Py_XDECREF(tmp); |
| 974 | return result; |
| 975 | } |