1 /* locale.c 2 * 3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others 5 * 6 * You may distribute under the terms of either the GNU General Public 7 * License or the Artistic License, as specified in the README file. 8 * 9 */ 10 11 /* 12 * A Elbereth Gilthoniel, 13 * silivren penna míriel 14 * o menel aglar elenath! 15 * Na-chaered palan-díriel 16 * o galadhremmin ennorath, 17 * Fanuilos, le linnathon 18 * nef aear, si nef aearon! 19 * 20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"] 21 */ 22 23 /* utility functions for handling locale-specific stuff like what 24 * character represents the decimal point. 25 * 26 * All C programs have an underlying locale. Perl code generally doesn't pay 27 * any attention to it except within the scope of a 'use locale'. For most 28 * categories, it accomplishes this by just using different operations if it is 29 * in such scope than if not. However, various libc functions called by Perl 30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that 31 * are used to toggle between the current locale and the C locale depending on 32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is 33 * switched to the C locale for outputting the message unless within the scope 34 * of 'use locale'. 35 * 36 * This code now has multi-thread-safe locale handling on systems that support 37 * that. This is completely transparent to most XS code. On earlier systems, 38 * it would be possible to emulate thread-safe locales, but this likely would 39 * involve a lot of locale switching, and would require XS code changes. 40 * Macros could be written so that the code wouldn't have to know which type of 41 * system is being used. It's unlikely that we would ever do that, since most 42 * modern systems support thread-safe locales, but there was code written to 43 * this end, and is retained, #ifdef'd out. 44 */ 45 46 #include "EXTERN.h" 47 #define PERL_IN_LOCALE_C 48 #include "perl_langinfo.h" 49 #include "perl.h" 50 51 #include "reentr.h" 52 53 #ifdef I_WCHAR 54 # include <wchar.h> 55 #endif 56 #ifdef I_WCTYPE 57 # include <wctype.h> 58 #endif 59 60 /* If the environment says to, we can output debugging information during 61 * initialization. This is done before option parsing, and before any thread 62 * creation, so can be a file-level static */ 63 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT) 64 # define debug_initialization 0 65 # define DEBUG_INITIALIZATION_set(v) 66 #else 67 static bool debug_initialization = FALSE; 68 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v) 69 #endif 70 71 72 /* Returns the Unix errno portion; ignoring any others. This is a macro here 73 * instead of putting it into perl.h, because unclear to khw what should be 74 * done generally. */ 75 #define GET_ERRNO saved_errno 76 77 /* strlen() of a literal string constant. We might want this more general, 78 * but using it in just this file for now. A problem with more generality is 79 * the compiler warnings about comparing unlike signs */ 80 #define STRLENs(s) (sizeof("" s "") - 1) 81 82 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the 83 * return of setlocale(), then this is extremely likely to be the C or POSIX 84 * locale. However, the output of setlocale() is documented to be opaque, but 85 * the odds are extremely small that it would return these two strings for some 86 * other locale. Note that VMS in these two locales includes many non-ASCII 87 * characters as controls and punctuation (below are hex bytes): 88 * cntrl: 84-97 9B-9F 89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD 90 * Oddly, none there are listed as alphas, though some represent alphabetics 91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */ 92 #define isNAME_C_OR_POSIX(name) \ 93 ( (name) != NULL \ 94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \ 95 || strEQ((name), "POSIX"))) 96 97 #ifdef USE_LOCALE 98 99 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far 100 * looked up. This is in the form of a C string: */ 101 102 #define UTF8NESS_SEP "\v" 103 #define UTF8NESS_PREFIX "\f" 104 105 /* So, the string looks like: 106 * 107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0 108 * 109 * where the digit 0 after the \a indicates that the locale starting just 110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */ 111 112 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1); 113 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1); 114 115 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \ 116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0" 117 118 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are 119 * kept there always. The remining portion of the cache is LRU, with the 120 * oldest looked-up locale at the tail end */ 121 122 STATIC char * 123 S_stdize_locale(pTHX_ char *locs) 124 { 125 /* Standardize the locale name from a string returned by 'setlocale', 126 * possibly modifying that string. 127 * 128 * The typical return value of setlocale() is either 129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL 130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL 131 * (the space-separated values represent the various sublocales, 132 * in some unspecified order). This is not handled by this function. 133 * 134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n", 135 * which is harmful for further use of the string in setlocale(). This 136 * function removes the trailing new line and everything up through the '=' 137 * */ 138 139 const char * const s = strchr(locs, '='); 140 bool okay = TRUE; 141 142 PERL_ARGS_ASSERT_STDIZE_LOCALE; 143 144 if (s) { 145 const char * const t = strchr(s, '.'); 146 okay = FALSE; 147 if (t) { 148 const char * const u = strchr(t, '\n'); 149 if (u && (u[1] == 0)) { 150 const STRLEN len = u - s; 151 Move(s + 1, locs, len, char); 152 locs[len] = 0; 153 okay = TRUE; 154 } 155 } 156 } 157 158 if (!okay) 159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs); 160 161 return locs; 162 } 163 164 /* Two parallel arrays; first the locale categories Perl uses on this system; 165 * the second array is their names. These arrays are in mostly arbitrary 166 * order. */ 167 168 const int categories[] = { 169 170 # ifdef USE_LOCALE_NUMERIC 171 LC_NUMERIC, 172 # endif 173 # ifdef USE_LOCALE_CTYPE 174 LC_CTYPE, 175 # endif 176 # ifdef USE_LOCALE_COLLATE 177 LC_COLLATE, 178 # endif 179 # ifdef USE_LOCALE_TIME 180 LC_TIME, 181 # endif 182 # ifdef USE_LOCALE_MESSAGES 183 LC_MESSAGES, 184 # endif 185 # ifdef USE_LOCALE_MONETARY 186 LC_MONETARY, 187 # endif 188 # ifdef USE_LOCALE_ADDRESS 189 LC_ADDRESS, 190 # endif 191 # ifdef USE_LOCALE_IDENTIFICATION 192 LC_IDENTIFICATION, 193 # endif 194 # ifdef USE_LOCALE_MEASUREMENT 195 LC_MEASUREMENT, 196 # endif 197 # ifdef USE_LOCALE_PAPER 198 LC_PAPER, 199 # endif 200 # ifdef USE_LOCALE_TELEPHONE 201 LC_TELEPHONE, 202 # endif 203 # ifdef LC_ALL 204 LC_ALL, 205 # endif 206 -1 /* Placeholder because C doesn't allow a 207 trailing comma, and it would get complicated 208 with all the #ifdef's */ 209 }; 210 211 /* The top-most real element is LC_ALL */ 212 213 const char * const category_names[] = { 214 215 # ifdef USE_LOCALE_NUMERIC 216 "LC_NUMERIC", 217 # endif 218 # ifdef USE_LOCALE_CTYPE 219 "LC_CTYPE", 220 # endif 221 # ifdef USE_LOCALE_COLLATE 222 "LC_COLLATE", 223 # endif 224 # ifdef USE_LOCALE_TIME 225 "LC_TIME", 226 # endif 227 # ifdef USE_LOCALE_MESSAGES 228 "LC_MESSAGES", 229 # endif 230 # ifdef USE_LOCALE_MONETARY 231 "LC_MONETARY", 232 # endif 233 # ifdef USE_LOCALE_ADDRESS 234 "LC_ADDRESS", 235 # endif 236 # ifdef USE_LOCALE_IDENTIFICATION 237 "LC_IDENTIFICATION", 238 # endif 239 # ifdef USE_LOCALE_MEASUREMENT 240 "LC_MEASUREMENT", 241 # endif 242 # ifdef USE_LOCALE_PAPER 243 "LC_PAPER", 244 # endif 245 # ifdef USE_LOCALE_TELEPHONE 246 "LC_TELEPHONE", 247 # endif 248 # ifdef LC_ALL 249 "LC_ALL", 250 # endif 251 NULL /* Placeholder */ 252 }; 253 254 # ifdef LC_ALL 255 256 /* On systems with LC_ALL, it is kept in the highest index position. (-2 257 * to account for the final unused placeholder element.) */ 258 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2) 259 260 # else 261 262 /* On systems without LC_ALL, we pretend it is there, one beyond the real 263 * top element, hence in the unused placeholder element. */ 264 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1) 265 266 # endif 267 268 /* Pretending there is an LC_ALL element just above allows us to avoid most 269 * special cases. Most loops through these arrays in the code below are 270 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work 271 * on either type of system. But the code must be written to not access the 272 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be 273 * checked for at compile time by using the #define LC_ALL_INDEX which is only 274 * defined if we do have LC_ALL. */ 275 276 STATIC const char * 277 S_category_name(const int category) 278 { 279 unsigned int i; 280 281 #ifdef LC_ALL 282 283 if (category == LC_ALL) { 284 return "LC_ALL"; 285 } 286 287 #endif 288 289 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) { 290 if (category == categories[i]) { 291 return category_names[i]; 292 } 293 } 294 295 { 296 const char suffix[] = " (unknown)"; 297 int temp = category; 298 Size_t length = sizeof(suffix) + 1; 299 char * unknown; 300 dTHX; 301 302 if (temp < 0) { 303 length++; 304 temp = - temp; 305 } 306 307 /* Calculate the number of digits */ 308 while (temp >= 10) { 309 temp /= 10; 310 length++; 311 } 312 313 Newx(unknown, length, char); 314 my_snprintf(unknown, length, "%d%s", category, suffix); 315 SAVEFREEPV(unknown); 316 return unknown; 317 } 318 } 319 320 /* Now create LC_foo_INDEX #defines for just those categories on this system */ 321 # ifdef USE_LOCALE_NUMERIC 322 # define LC_NUMERIC_INDEX 0 323 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX 324 # else 325 # define _DUMMY_NUMERIC -1 326 # endif 327 # ifdef USE_LOCALE_CTYPE 328 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1 329 # define _DUMMY_CTYPE LC_CTYPE_INDEX 330 # else 331 # define _DUMMY_CTYPE _DUMMY_NUMERIC 332 # endif 333 # ifdef USE_LOCALE_COLLATE 334 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1 335 # define _DUMMY_COLLATE LC_COLLATE_INDEX 336 # else 337 # define _DUMMY_COLLATE _DUMMY_CTYPE 338 # endif 339 # ifdef USE_LOCALE_TIME 340 # define LC_TIME_INDEX _DUMMY_COLLATE + 1 341 # define _DUMMY_TIME LC_TIME_INDEX 342 # else 343 # define _DUMMY_TIME _DUMMY_COLLATE 344 # endif 345 # ifdef USE_LOCALE_MESSAGES 346 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1 347 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX 348 # else 349 # define _DUMMY_MESSAGES _DUMMY_TIME 350 # endif 351 # ifdef USE_LOCALE_MONETARY 352 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1 353 # define _DUMMY_MONETARY LC_MONETARY_INDEX 354 # else 355 # define _DUMMY_MONETARY _DUMMY_MESSAGES 356 # endif 357 # ifdef USE_LOCALE_ADDRESS 358 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1 359 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX 360 # else 361 # define _DUMMY_ADDRESS _DUMMY_MONETARY 362 # endif 363 # ifdef USE_LOCALE_IDENTIFICATION 364 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1 365 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX 366 # else 367 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS 368 # endif 369 # ifdef USE_LOCALE_MEASUREMENT 370 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1 371 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX 372 # else 373 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION 374 # endif 375 # ifdef USE_LOCALE_PAPER 376 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1 377 # define _DUMMY_PAPER LC_PAPER_INDEX 378 # else 379 # define _DUMMY_PAPER _DUMMY_MEASUREMENT 380 # endif 381 # ifdef USE_LOCALE_TELEPHONE 382 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1 383 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX 384 # else 385 # define _DUMMY_TELEPHONE _DUMMY_PAPER 386 # endif 387 # ifdef LC_ALL 388 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1 389 # endif 390 #endif /* ifdef USE_LOCALE */ 391 392 /* Windows requres a customized base-level setlocale() */ 393 #ifdef WIN32 394 # define my_setlocale(cat, locale) win32_setlocale(cat, locale) 395 #else 396 # define my_setlocale(cat, locale) setlocale(cat, locale) 397 #endif 398 399 #ifndef USE_POSIX_2008_LOCALE 400 401 /* "do_setlocale_c" is intended to be called when the category is a constant 402 * known at compile time; "do_setlocale_r", not known until run time */ 403 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale) 404 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale) 405 # define FIX_GLIBC_LC_MESSAGES_BUG(i) 406 407 #else /* Below uses POSIX 2008 */ 408 409 /* We emulate setlocale with our own function. LC_foo is not valid for the 410 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array 411 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the 412 * proper offset into the array 'category_masks[]'. At runtime, we have to 413 * search through the array (as the actual numbers may not be small contiguous 414 * positive integers which would lend themselves to array lookup). */ 415 # define do_setlocale_c(cat, locale) \ 416 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE) 417 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE) 418 419 # if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES) 420 421 # define FIX_GLIBC_LC_MESSAGES_BUG(i) 422 423 # else /* Invalidate glibc cache of loaded translations, see [perl #134264] */ 424 425 # include <libintl.h> 426 # define FIX_GLIBC_LC_MESSAGES_BUG(i) \ 427 STMT_START { \ 428 if ((i) == LC_MESSAGES_INDEX) { \ 429 textdomain(textdomain(NULL)); \ 430 } \ 431 } STMT_END 432 433 # endif 434 435 /* A third array, parallel to the ones above to map from category to its 436 * equivalent mask */ 437 const int category_masks[] = { 438 # ifdef USE_LOCALE_NUMERIC 439 LC_NUMERIC_MASK, 440 # endif 441 # ifdef USE_LOCALE_CTYPE 442 LC_CTYPE_MASK, 443 # endif 444 # ifdef USE_LOCALE_COLLATE 445 LC_COLLATE_MASK, 446 # endif 447 # ifdef USE_LOCALE_TIME 448 LC_TIME_MASK, 449 # endif 450 # ifdef USE_LOCALE_MESSAGES 451 LC_MESSAGES_MASK, 452 # endif 453 # ifdef USE_LOCALE_MONETARY 454 LC_MONETARY_MASK, 455 # endif 456 # ifdef USE_LOCALE_ADDRESS 457 LC_ADDRESS_MASK, 458 # endif 459 # ifdef USE_LOCALE_IDENTIFICATION 460 LC_IDENTIFICATION_MASK, 461 # endif 462 # ifdef USE_LOCALE_MEASUREMENT 463 LC_MEASUREMENT_MASK, 464 # endif 465 # ifdef USE_LOCALE_PAPER 466 LC_PAPER_MASK, 467 # endif 468 # ifdef USE_LOCALE_TELEPHONE 469 LC_TELEPHONE_MASK, 470 # endif 471 /* LC_ALL can't be turned off by a Configure 472 * option, and in Posix 2008, should always be 473 * here, so compile it in unconditionally. 474 * This could catch some glitches at compile 475 * time */ 476 LC_ALL_MASK 477 }; 478 479 STATIC const char * 480 S_emulate_setlocale(const int category, 481 const char * locale, 482 unsigned int index, 483 const bool is_index_valid 484 ) 485 { 486 /* This function effectively performs a setlocale() on just the current 487 * thread; thus it is thread-safe. It does this by using the POSIX 2008 488 * locale functions to emulate the behavior of setlocale(). Similar to 489 * regular setlocale(), the return from this function points to memory that 490 * can be overwritten by other system calls, so needs to be copied 491 * immediately if you need to retain it. The difference here is that 492 * system calls besides another setlocale() can overwrite it. 493 * 494 * By doing this, most locale-sensitive functions become thread-safe. The 495 * exceptions are mostly those that return a pointer to static memory. 496 * 497 * This function takes the same parameters, 'category' and 'locale', that 498 * the regular setlocale() function does, but it also takes two additional 499 * ones. This is because the 2008 functions don't use a category; instead 500 * they use a corresponding mask. Because this function operates in both 501 * worlds, it may need one or the other or both. This function can 502 * calculate the mask from the input category, but to avoid this 503 * calculation, if the caller knows at compile time what the mask is, it 504 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask 505 * parameter is ignored. 506 * 507 * POSIX 2008, for some sick reason, chose not to provide a method to find 508 * the category name of a locale. Some vendors have created a 509 * querylocale() function to do just that. This function is a lot simpler 510 * to implement on systems that have this. Otherwise, we have to keep 511 * track of what the locale has been set to, so that we can return its 512 * name to emulate setlocale(). It's also possible for C code in some 513 * library to change the locale without us knowing it, though as of 514 * September 2017, there are no occurrences in CPAN of uselocale(). Some 515 * libraries do use setlocale(), but that changes the global locale, and 516 * threads using per-thread locales will just ignore those changes. 517 * Another problem is that without querylocale(), we have to guess at what 518 * was meant by setting a locale of "". We handle this by not actually 519 * ever setting to "" (unless querylocale exists), but to emulate what we 520 * think should happen for "". 521 */ 522 523 int mask; 524 locale_t old_obj; 525 locale_t new_obj; 526 dTHX; 527 528 # ifdef DEBUGGING 529 530 if (DEBUG_Lv_TEST || debug_initialization) { 531 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid); 532 } 533 534 # endif 535 536 /* If the input mask might be incorrect, calculate the correct one */ 537 if (! is_index_valid) { 538 unsigned int i; 539 540 # ifdef DEBUGGING 541 542 if (DEBUG_Lv_TEST || debug_initialization) { 543 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category)); 544 } 545 546 # endif 547 548 for (i = 0; i <= LC_ALL_INDEX; i++) { 549 if (category == categories[i]) { 550 index = i; 551 goto found_index; 552 } 553 } 554 555 /* Here, we don't know about this category, so can't handle it. 556 * Fallback to the early POSIX usages */ 557 Perl_warner(aTHX_ packWARN(WARN_LOCALE), 558 "Unknown locale category %d; can't set it to %s\n", 559 category, locale); 560 return NULL; 561 562 found_index: ; 563 564 # ifdef DEBUGGING 565 566 if (DEBUG_Lv_TEST || debug_initialization) { 567 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category)); 568 } 569 570 # endif 571 572 } 573 574 mask = category_masks[index]; 575 576 # ifdef DEBUGGING 577 578 if (DEBUG_Lv_TEST || debug_initialization) { 579 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask); 580 } 581 582 # endif 583 584 /* If just querying what the existing locale is ... */ 585 if (locale == NULL) { 586 locale_t cur_obj = uselocale((locale_t) 0); 587 588 # ifdef DEBUGGING 589 590 if (DEBUG_Lv_TEST || debug_initialization) { 591 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj); 592 } 593 594 # endif 595 596 if (cur_obj == LC_GLOBAL_LOCALE) { 597 return my_setlocale(category, NULL); 598 } 599 600 # ifdef HAS_QUERYLOCALE 601 602 return (char *) querylocale(mask, cur_obj); 603 604 # else 605 606 /* If this assert fails, adjust the size of curlocales in intrpvar.h */ 607 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX); 608 609 # if defined(_NL_LOCALE_NAME) \ 610 && defined(DEBUGGING) \ 611 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME) 612 /* On systems that accept any locale name, the real underlying locale 613 * is often returned by this internal function, so we can't use it */ 614 { 615 /* Internal glibc for querylocale(), but doesn't handle 616 * empty-string ("") locale properly; who knows what other 617 * glitches. Check for it now, under debug. */ 618 619 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category), 620 uselocale((locale_t) 0)); 621 /* 622 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL"); 623 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index); 624 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]); 625 */ 626 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) { 627 if ( strNE(PL_curlocales[index], temp_name) 628 && ! ( isNAME_C_OR_POSIX(temp_name) 629 && isNAME_C_OR_POSIX(PL_curlocales[index]))) { 630 631 # ifdef USE_C_BACKTRACE 632 633 dump_c_backtrace(Perl_debug_log, 20, 1); 634 635 # endif 636 637 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is" 638 " (%s) and what internal glibc thinks" 639 " (%s)\n", category_names[index], 640 PL_curlocales[index], temp_name); 641 } 642 643 return temp_name; 644 } 645 } 646 647 # endif 648 649 /* Without querylocale(), we have to use our record-keeping we've 650 * done. */ 651 652 if (category != LC_ALL) { 653 654 # ifdef DEBUGGING 655 656 if (DEBUG_Lv_TEST || debug_initialization) { 657 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]); 658 } 659 660 # endif 661 662 return PL_curlocales[index]; 663 } 664 else { /* For LC_ALL */ 665 unsigned int i; 666 Size_t names_len = 0; 667 char * all_string; 668 bool are_all_categories_the_same_locale = TRUE; 669 670 /* If we have a valid LC_ALL value, just return it */ 671 if (PL_curlocales[LC_ALL_INDEX]) { 672 673 # ifdef DEBUGGING 674 675 if (DEBUG_Lv_TEST || debug_initialization) { 676 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]); 677 } 678 679 # endif 680 681 return PL_curlocales[LC_ALL_INDEX]; 682 } 683 684 /* Otherwise, we need to construct a string of name=value pairs. 685 * We use the glibc syntax, like 686 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;... 687 * First calculate the needed size. Along the way, check if all 688 * the locale names are the same */ 689 for (i = 0; i < LC_ALL_INDEX; i++) { 690 691 # ifdef DEBUGGING 692 693 if (DEBUG_Lv_TEST || debug_initialization) { 694 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]); 695 } 696 697 # endif 698 699 names_len += strlen(category_names[i]) 700 + 1 /* '=' */ 701 + strlen(PL_curlocales[i]) 702 + 1; /* ';' */ 703 704 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) { 705 are_all_categories_the_same_locale = FALSE; 706 } 707 } 708 709 /* If they are the same, we don't actually have to construct the 710 * string; we just make the entry in LC_ALL_INDEX valid, and be 711 * that single name */ 712 if (are_all_categories_the_same_locale) { 713 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]); 714 return PL_curlocales[LC_ALL_INDEX]; 715 } 716 717 names_len++; /* Trailing '\0' */ 718 SAVEFREEPV(Newx(all_string, names_len, char)); 719 *all_string = '\0'; 720 721 /* Then fill in the string */ 722 for (i = 0; i < LC_ALL_INDEX; i++) { 723 724 # ifdef DEBUGGING 725 726 if (DEBUG_Lv_TEST || debug_initialization) { 727 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]); 728 } 729 730 # endif 731 732 my_strlcat(all_string, category_names[i], names_len); 733 my_strlcat(all_string, "=", names_len); 734 my_strlcat(all_string, PL_curlocales[i], names_len); 735 my_strlcat(all_string, ";", names_len); 736 } 737 738 # ifdef DEBUGGING 739 740 if (DEBUG_L_TEST || debug_initialization) { 741 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string); 742 } 743 744 #endif 745 746 return all_string; 747 } 748 749 # ifdef EINVAL 750 751 SETERRNO(EINVAL, LIB_INVARG); 752 753 # endif 754 755 return NULL; 756 757 # endif 758 759 } /* End of this being setlocale(LC_foo, NULL) */ 760 761 /* Here, we are switching locales. */ 762 763 # ifndef HAS_QUERYLOCALE 764 765 if (strEQ(locale, "")) { 766 767 /* For non-querylocale() systems, we do the setting of "" ourselves to 768 * be sure that we really know what's going on. We follow the Linux 769 * documented behavior (but if that differs from the actual behavior, 770 * this won't work exactly as the OS implements). We go out and 771 * examine the environment based on our understanding of how the system 772 * works, and use that to figure things out */ 773 774 const char * const lc_all = PerlEnv_getenv("LC_ALL"); 775 776 /* Use any "LC_ALL" environment variable, as it overrides everything 777 * else. */ 778 if (lc_all && strNE(lc_all, "")) { 779 locale = lc_all; 780 } 781 else { 782 783 /* Otherwise, we need to dig deeper. Unless overridden, the 784 * default is the LANG environment variable; if it doesn't exist, 785 * then "C" */ 786 787 const char * default_name; 788 789 default_name = PerlEnv_getenv("LANG"); 790 791 if (! default_name || strEQ(default_name, "")) { 792 default_name = "C"; 793 } 794 795 if (category != LC_ALL) { 796 const char * const name = PerlEnv_getenv(category_names[index]); 797 798 /* Here we are setting a single category. Assume will have the 799 * default name */ 800 locale = default_name; 801 802 /* But then look for an overriding environment variable */ 803 if (name && strNE(name, "")) { 804 locale = name; 805 } 806 } 807 else { 808 bool did_override = FALSE; 809 unsigned int i; 810 811 /* Here, we are getting LC_ALL. Any categories that don't have 812 * a corresponding environment variable set should be set to 813 * LANG, or to "C" if there is no LANG. If no individual 814 * categories differ from this, we can just set LC_ALL. This 815 * is buggy on systems that have extra categories that we don't 816 * know about. If there is an environment variable that sets 817 * that category, we won't know to look for it, and so our use 818 * of LANG or "C" improperly overrides it. On the other hand, 819 * if we don't do what is done here, and there is no 820 * environment variable, the category's locale should be set to 821 * LANG or "C". So there is no good solution. khw thinks the 822 * best is to look at systems to see what categories they have, 823 * and include them, and then to assume that we know the 824 * complete set */ 825 826 for (i = 0; i < LC_ALL_INDEX; i++) { 827 const char * const env_override 828 = PerlEnv_getenv(category_names[i]); 829 const char * this_locale = ( env_override 830 && strNE(env_override, "")) 831 ? env_override 832 : default_name; 833 if (! emulate_setlocale(categories[i], this_locale, i, TRUE)) 834 { 835 return NULL; 836 } 837 838 if (strNE(this_locale, default_name)) { 839 did_override = TRUE; 840 } 841 } 842 843 /* If all the categories are the same, we can set LC_ALL to 844 * that */ 845 if (! did_override) { 846 locale = default_name; 847 } 848 else { 849 850 /* Here, LC_ALL is no longer valid, as some individual 851 * categories don't match it. We call ourselves 852 * recursively, as that will execute the code that 853 * generates the proper locale string for this situation. 854 * We don't do the remainder of this function, as that is 855 * to update our records, and we've just done that for the 856 * individual categories in the loop above, and doing so 857 * would cause LC_ALL to be done as well */ 858 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE); 859 } 860 } 861 } 862 } /* End of this being setlocale(LC_foo, "") */ 863 else if (strchr(locale, ';')) { 864 865 /* LC_ALL may actually incude a conglomeration of various categories. 866 * Without querylocale, this code uses the glibc (as of this writing) 867 * syntax for representing that, but that is not a stable API, and 868 * other platforms do it differently, so we have to handle all cases 869 * ourselves */ 870 871 unsigned int i; 872 const char * s = locale; 873 const char * e = locale + strlen(locale); 874 const char * p = s; 875 const char * category_end; 876 const char * name_start; 877 const char * name_end; 878 879 /* If the string that gives what to set doesn't include all categories, 880 * the omitted ones get set to "C". To get this behavior, first set 881 * all the individual categories to "C", and override the furnished 882 * ones below */ 883 for (i = 0; i < LC_ALL_INDEX; i++) { 884 if (! emulate_setlocale(categories[i], "C", i, TRUE)) { 885 return NULL; 886 } 887 } 888 889 while (s < e) { 890 891 /* Parse through the category */ 892 while (isWORDCHAR(*p)) { 893 p++; 894 } 895 category_end = p; 896 897 if (*p++ != '=') { 898 Perl_croak(aTHX_ 899 "panic: %s: %d: Unexpected character in locale name '%02X", 900 __FILE__, __LINE__, *(p-1)); 901 } 902 903 /* Parse through the locale name */ 904 name_start = p; 905 while (p < e && *p != ';') { 906 if (! isGRAPH(*p)) { 907 Perl_croak(aTHX_ 908 "panic: %s: %d: Unexpected character in locale name '%02X", 909 __FILE__, __LINE__, *(p-1)); 910 } 911 p++; 912 } 913 name_end = p; 914 915 /* Space past the semi-colon */ 916 if (p < e) { 917 p++; 918 } 919 920 /* Find the index of the category name in our lists */ 921 for (i = 0; i < LC_ALL_INDEX; i++) { 922 char * individ_locale; 923 924 /* Keep going if this isn't the index. The strnNE() avoids a 925 * Perl_form(), but would fail if ever a category name could be 926 * a substring of another one, like if there were a 927 * "LC_TIME_DATE" */ 928 if strnNE(s, category_names[i], category_end - s) { 929 continue; 930 } 931 932 /* If this index is for the single category we're changing, we 933 * have found the locale to set it to. */ 934 if (category == categories[i]) { 935 locale = Perl_form(aTHX_ "%.*s", 936 (int) (name_end - name_start), 937 name_start); 938 goto ready_to_set; 939 } 940 941 assert(category == LC_ALL); 942 individ_locale = Perl_form(aTHX_ "%.*s", 943 (int) (name_end - name_start), name_start); 944 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE)) 945 { 946 return NULL; 947 } 948 } 949 950 s = p; 951 } 952 953 /* Here we have set all the individual categories by recursive calls. 954 * These collectively should have fixed up LC_ALL, so can just query 955 * what that now is */ 956 assert(category == LC_ALL); 957 958 return do_setlocale_c(LC_ALL, NULL); 959 } /* End of this being setlocale(LC_ALL, 960 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */ 961 962 ready_to_set: ; 963 964 /* Here at the end of having to deal with the absence of querylocale(). 965 * Some cases have already been fully handled by recursive calls to this 966 * function. But at this point, we haven't dealt with those, but are now 967 * prepared to, knowing what the locale name to set this category to is. 968 * This would have come for free if this system had had querylocale() */ 969 970 # endif /* end of ! querylocale */ 971 972 assert(PL_C_locale_obj); 973 974 /* Switching locales generally entails freeing the current one's space (at 975 * the C library's discretion). We need to stop using that locale before 976 * the switch. So switch to a known locale object that we don't otherwise 977 * mess with. This returns the locale object in effect at the time of the 978 * switch. */ 979 old_obj = uselocale(PL_C_locale_obj); 980 981 # ifdef DEBUGGING 982 983 if (DEBUG_Lv_TEST || debug_initialization) { 984 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj); 985 } 986 987 # endif 988 989 if (! old_obj) { 990 991 # ifdef DEBUGGING 992 993 if (DEBUG_L_TEST || debug_initialization) { 994 dSAVE_ERRNO; 995 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO); 996 RESTORE_ERRNO; 997 } 998 999 # endif 1000 1001 return NULL; 1002 } 1003 1004 # ifdef DEBUGGING 1005 1006 if (DEBUG_Lv_TEST || debug_initialization) { 1007 PerlIO_printf(Perl_debug_log, 1008 "%s:%d: emulate_setlocale now using %p\n", 1009 __FILE__, __LINE__, PL_C_locale_obj); 1010 } 1011 1012 # endif 1013 1014 /* If this call is to switch to the LC_ALL C locale, it already exists, and 1015 * in fact, we already have switched to it (in preparation for what 1016 * normally is to come). But since we're already there, continue to use 1017 * it instead of trying to create a new locale */ 1018 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) { 1019 1020 # ifdef DEBUGGING 1021 1022 if (DEBUG_Lv_TEST || debug_initialization) { 1023 PerlIO_printf(Perl_debug_log, 1024 "%s:%d: will stay in C object\n", __FILE__, __LINE__); 1025 } 1026 1027 # endif 1028 1029 new_obj = PL_C_locale_obj; 1030 1031 /* We already had switched to the C locale in preparation for freeing 1032 * 'old_obj' */ 1033 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) { 1034 freelocale(old_obj); 1035 } 1036 } 1037 else { 1038 /* If we weren't in a thread safe locale, set so that newlocale() below 1039 * which uses 'old_obj', uses an empty one. Same for our reserved C 1040 * object. The latter is defensive coding, so that, even if there is 1041 * some bug, we will never end up trying to modify either of these, as 1042 * if passed to newlocale(), they can be. */ 1043 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) { 1044 old_obj = (locale_t) 0; 1045 } 1046 1047 /* Ready to create a new locale by modification of the exising one */ 1048 new_obj = newlocale(mask, locale, old_obj); 1049 1050 if (! new_obj) { 1051 dSAVE_ERRNO; 1052 1053 # ifdef DEBUGGING 1054 1055 if (DEBUG_L_TEST || debug_initialization) { 1056 PerlIO_printf(Perl_debug_log, 1057 "%s:%d: emulate_setlocale creating new object" 1058 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO); 1059 } 1060 1061 # endif 1062 1063 if (! uselocale(old_obj)) { 1064 1065 # ifdef DEBUGGING 1066 1067 if (DEBUG_L_TEST || debug_initialization) { 1068 PerlIO_printf(Perl_debug_log, 1069 "%s:%d: switching back failed: %d\n", 1070 __FILE__, __LINE__, GET_ERRNO); 1071 } 1072 1073 # endif 1074 1075 } 1076 RESTORE_ERRNO; 1077 return NULL; 1078 } 1079 1080 # ifdef DEBUGGING 1081 1082 if (DEBUG_Lv_TEST || debug_initialization) { 1083 PerlIO_printf(Perl_debug_log, 1084 "%s:%d: emulate_setlocale created %p", 1085 __FILE__, __LINE__, new_obj); 1086 if (old_obj) { 1087 PerlIO_printf(Perl_debug_log, 1088 "; should have freed %p", old_obj); 1089 } 1090 PerlIO_printf(Perl_debug_log, "\n"); 1091 } 1092 1093 # endif 1094 1095 /* And switch into it */ 1096 if (! uselocale(new_obj)) { 1097 dSAVE_ERRNO; 1098 1099 # ifdef DEBUGGING 1100 1101 if (DEBUG_L_TEST || debug_initialization) { 1102 PerlIO_printf(Perl_debug_log, 1103 "%s:%d: emulate_setlocale switching to new object" 1104 " failed\n", __FILE__, __LINE__); 1105 } 1106 1107 # endif 1108 1109 if (! uselocale(old_obj)) { 1110 1111 # ifdef DEBUGGING 1112 1113 if (DEBUG_L_TEST || debug_initialization) { 1114 PerlIO_printf(Perl_debug_log, 1115 "%s:%d: switching back failed: %d\n", 1116 __FILE__, __LINE__, GET_ERRNO); 1117 } 1118 1119 # endif 1120 1121 } 1122 freelocale(new_obj); 1123 RESTORE_ERRNO; 1124 return NULL; 1125 } 1126 } 1127 1128 # ifdef DEBUGGING 1129 1130 if (DEBUG_Lv_TEST || debug_initialization) { 1131 PerlIO_printf(Perl_debug_log, 1132 "%s:%d: emulate_setlocale now using %p\n", 1133 __FILE__, __LINE__, new_obj); 1134 } 1135 1136 # endif 1137 1138 /* We are done, except for updating our records (if the system doesn't keep 1139 * them) and in the case of locale "", we don't actually know what the 1140 * locale that got switched to is, as it came from the environment. So 1141 * have to find it */ 1142 1143 # ifdef HAS_QUERYLOCALE 1144 1145 if (strEQ(locale, "")) { 1146 locale = querylocale(mask, new_obj); 1147 } 1148 1149 # else 1150 1151 /* Here, 'locale' is the return value */ 1152 1153 /* Without querylocale(), we have to update our records */ 1154 1155 if (category == LC_ALL) { 1156 unsigned int i; 1157 1158 /* For LC_ALL, we change all individual categories to correspond */ 1159 /* PL_curlocales is a parallel array, so has same 1160 * length as 'categories' */ 1161 for (i = 0; i <= LC_ALL_INDEX; i++) { 1162 Safefree(PL_curlocales[i]); 1163 PL_curlocales[i] = savepv(locale); 1164 } 1165 1166 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX); 1167 } 1168 else { 1169 1170 /* For a single category, if it's not the same as the one in LC_ALL, we 1171 * nullify LC_ALL */ 1172 1173 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) { 1174 Safefree(PL_curlocales[LC_ALL_INDEX]); 1175 PL_curlocales[LC_ALL_INDEX] = NULL; 1176 } 1177 1178 /* Then update the category's record */ 1179 Safefree(PL_curlocales[index]); 1180 PL_curlocales[index] = savepv(locale); 1181 1182 FIX_GLIBC_LC_MESSAGES_BUG(index); 1183 } 1184 1185 # endif 1186 1187 return locale; 1188 } 1189 1190 #endif /* USE_POSIX_2008_LOCALE */ 1191 1192 #if 0 /* Code that was to emulate thread-safe locales on platforms that 1193 didn't natively support them */ 1194 1195 /* The way this would work is that we would keep a per-thread list of the 1196 * correct locale for that thread. Any operation that was locale-sensitive 1197 * would have to be changed so that it would look like this: 1198 * 1199 * LOCALE_LOCK; 1200 * setlocale to the correct locale for this operation 1201 * do operation 1202 * LOCALE_UNLOCK 1203 * 1204 * This leaves the global locale in the most recently used operation's, but it 1205 * was locked long enough to get the result. If that result is static, it 1206 * needs to be copied before the unlock. 1207 * 1208 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did 1209 * the setup, but are no-ops when not needed, and similarly, 1210 * END_LOCALE_DEPENDENT_OP for the tear-down 1211 * 1212 * But every call to a locale-sensitive function would have to be changed, and 1213 * if a module didn't cooperate by using the mutex, things would break. 1214 * 1215 * This code was abandoned before being completed or tested, and is left as-is 1216 */ 1217 1218 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE) 1219 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE) 1220 1221 STATIC char * 1222 S_locking_setlocale(pTHX_ 1223 const int category, 1224 const char * locale, 1225 int index, 1226 const bool is_index_valid 1227 ) 1228 { 1229 /* This function kind of performs a setlocale() on just the current thread; 1230 * thus it is kind of thread-safe. It does this by keeping a thread-level 1231 * array of the current locales for each category. Every time a locale is 1232 * switched to, it does the switch globally, but updates the thread's 1233 * array. A query as to what the current locale is just returns the 1234 * appropriate element from the array, and doesn't actually call the system 1235 * setlocale(). The saving into the array is done in an uninterruptible 1236 * section of code, so is unaffected by whatever any other threads might be 1237 * doing. 1238 * 1239 * All locale-sensitive operations must work by first starting a critical 1240 * section, then switching to the thread's locale as kept by this function, 1241 * and then doing the operation, then ending the critical section. Thus, 1242 * each gets done in the appropriate locale. simulating thread-safety. 1243 * 1244 * This function takes the same parameters, 'category' and 'locale', that 1245 * the regular setlocale() function does, but it also takes two additional 1246 * ones. This is because as described earlier. If we know on input the 1247 * index corresponding to the category into the array where we store the 1248 * current locales, we don't have to calculate it. If the caller knows at 1249 * compile time what the index is, it can pass it, setting 1250 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored. 1251 * 1252 */ 1253 1254 /* If the input index might be incorrect, calculate the correct one */ 1255 if (! is_index_valid) { 1256 unsigned int i; 1257 1258 if (DEBUG_Lv_TEST || debug_initialization) { 1259 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category); 1260 } 1261 1262 for (i = 0; i <= LC_ALL_INDEX; i++) { 1263 if (category == categories[i]) { 1264 index = i; 1265 goto found_index; 1266 } 1267 } 1268 1269 /* Here, we don't know about this category, so can't handle it. 1270 * XXX best we can do is to unsafely set this 1271 * XXX warning */ 1272 1273 return my_setlocale(category, locale); 1274 1275 found_index: ; 1276 1277 if (DEBUG_Lv_TEST || debug_initialization) { 1278 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index); 1279 } 1280 } 1281 1282 /* For a query, just return what's in our records */ 1283 if (new_locale == NULL) { 1284 return curlocales[index]; 1285 } 1286 1287 1288 /* Otherwise, we need to do the switch, and save the result, all in a 1289 * critical section */ 1290 1291 Safefree(curlocales[[index]]); 1292 1293 /* It might be that this is called from an already-locked section of code. 1294 * We would have to detect and skip the LOCK/UNLOCK if so */ 1295 LOCALE_LOCK; 1296 1297 curlocales[index] = savepv(my_setlocale(category, new_locale)); 1298 1299 if (strEQ(new_locale, "")) { 1300 1301 #ifdef LC_ALL 1302 1303 /* The locale values come from the environment, and may not all be the 1304 * same, so for LC_ALL, we have to update all the others, while the 1305 * mutex is still locked */ 1306 1307 if (category == LC_ALL) { 1308 unsigned int i; 1309 for (i = 0; i < LC_ALL_INDEX) { 1310 curlocales[i] = my_setlocale(categories[i], NULL); 1311 } 1312 } 1313 } 1314 1315 #endif 1316 1317 LOCALE_UNLOCK; 1318 1319 return curlocales[index]; 1320 } 1321 1322 #endif 1323 #ifdef USE_LOCALE 1324 1325 STATIC void 1326 S_set_numeric_radix(pTHX_ const bool use_locale) 1327 { 1328 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If 1329 * TRUE, use the radix character derived from the current locale */ 1330 1331 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \ 1332 || defined(HAS_NL_LANGINFO)) 1333 1334 const char * radix = (use_locale) 1335 ? my_nl_langinfo(RADIXCHAR, FALSE) 1336 /* FALSE => already in dest locale */ 1337 : "."; 1338 1339 sv_setpv(PL_numeric_radix_sv, radix); 1340 1341 /* If this is valid UTF-8 that isn't totally ASCII, and we are in 1342 * a UTF-8 locale, then mark the radix as being in UTF-8 */ 1343 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv), 1344 SvCUR(PL_numeric_radix_sv)) 1345 && _is_cur_LC_category_utf8(LC_NUMERIC)) 1346 { 1347 SvUTF8_on(PL_numeric_radix_sv); 1348 } 1349 1350 # ifdef DEBUGGING 1351 1352 if (DEBUG_L_TEST || debug_initialization) { 1353 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n", 1354 SvPVX(PL_numeric_radix_sv), 1355 cBOOL(SvUTF8(PL_numeric_radix_sv))); 1356 } 1357 1358 # endif 1359 #else 1360 1361 PERL_UNUSED_ARG(use_locale); 1362 1363 #endif /* USE_LOCALE_NUMERIC and can find the radix char */ 1364 1365 } 1366 1367 STATIC void 1368 S_new_numeric(pTHX_ const char *newnum) 1369 { 1370 1371 #ifndef USE_LOCALE_NUMERIC 1372 1373 PERL_UNUSED_ARG(newnum); 1374 1375 #else 1376 1377 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell 1378 * core Perl this and that 'newnum' is the name of the new locale. 1379 * It installs this locale as the current underlying default. 1380 * 1381 * The default locale and the C locale can be toggled between by use of the 1382 * set_numeric_underlying() and set_numeric_standard() functions, which 1383 * should probably not be called directly, but only via macros like 1384 * SET_NUMERIC_STANDARD() in perl.h. 1385 * 1386 * The toggling is necessary mainly so that a non-dot radix decimal point 1387 * character can be output, while allowing internal calculations to use a 1388 * dot. 1389 * 1390 * This sets several interpreter-level variables: 1391 * PL_numeric_name The underlying locale's name: a copy of 'newnum' 1392 * PL_numeric_underlying A boolean indicating if the toggled state is such 1393 * that the current locale is the program's underlying 1394 * locale 1395 * PL_numeric_standard An int indicating if the toggled state is such 1396 * that the current locale is the C locale or 1397 * indistinguishable from the C locale. If non-zero, it 1398 * is in C; if > 1, it means it may not be toggled away 1399 * from C. 1400 * PL_numeric_underlying_is_standard A bool kept by this function 1401 * indicating that the underlying locale and the standard 1402 * C locale are indistinguishable for the purposes of 1403 * LC_NUMERIC. This happens when both of the above two 1404 * variables are true at the same time. (Toggling is a 1405 * no-op under these circumstances.) This variable is 1406 * used to avoid having to recalculate. 1407 */ 1408 1409 char *save_newnum; 1410 1411 if (! newnum) { 1412 Safefree(PL_numeric_name); 1413 PL_numeric_name = NULL; 1414 PL_numeric_standard = TRUE; 1415 PL_numeric_underlying = TRUE; 1416 PL_numeric_underlying_is_standard = TRUE; 1417 return; 1418 } 1419 1420 save_newnum = stdize_locale(savepv(newnum)); 1421 PL_numeric_underlying = TRUE; 1422 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum); 1423 1424 #ifndef TS_W32_BROKEN_LOCALECONV 1425 1426 /* If its name isn't C nor POSIX, it could still be indistinguishable from 1427 * them. But on broken Windows systems calling my_nl_langinfo() for 1428 * THOUSEP can currently (but rarely) cause a race, so avoid doing that, 1429 * and just always change the locale if not C nor POSIX on those systems */ 1430 if (! PL_numeric_standard) { 1431 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR, 1432 FALSE /* Don't toggle locale */ )) 1433 && strEQ("", my_nl_langinfo(THOUSEP, FALSE))); 1434 } 1435 1436 #endif 1437 1438 /* Save the new name if it isn't the same as the previous one, if any */ 1439 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) { 1440 Safefree(PL_numeric_name); 1441 PL_numeric_name = save_newnum; 1442 } 1443 else { 1444 Safefree(save_newnum); 1445 } 1446 1447 PL_numeric_underlying_is_standard = PL_numeric_standard; 1448 1449 # ifdef HAS_POSIX_2008_LOCALE 1450 1451 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK, 1452 PL_numeric_name, 1453 PL_underlying_numeric_obj); 1454 1455 #endif 1456 1457 if (DEBUG_L_TEST || debug_initialization) { 1458 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name); 1459 } 1460 1461 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't 1462 * have to worry about the radix being a non-dot. (Core operations that 1463 * need the underlying locale change to it temporarily). */ 1464 if (PL_numeric_standard) { 1465 set_numeric_radix(0); 1466 } 1467 else { 1468 set_numeric_standard(); 1469 } 1470 1471 #endif /* USE_LOCALE_NUMERIC */ 1472 1473 } 1474 1475 void 1476 Perl_set_numeric_standard(pTHX) 1477 { 1478 1479 #ifdef USE_LOCALE_NUMERIC 1480 1481 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like 1482 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The 1483 * macro avoids calling this routine if toggling isn't necessary according 1484 * to our records (which could be wrong if some XS code has changed the 1485 * locale behind our back) */ 1486 1487 # ifdef DEBUGGING 1488 1489 if (DEBUG_L_TEST || debug_initialization) { 1490 PerlIO_printf(Perl_debug_log, 1491 "Setting LC_NUMERIC locale to standard C\n"); 1492 } 1493 1494 # endif 1495 1496 do_setlocale_c(LC_NUMERIC, "C"); 1497 PL_numeric_standard = TRUE; 1498 PL_numeric_underlying = PL_numeric_underlying_is_standard; 1499 set_numeric_radix(0); 1500 1501 #endif /* USE_LOCALE_NUMERIC */ 1502 1503 } 1504 1505 void 1506 Perl_set_numeric_underlying(pTHX) 1507 { 1508 1509 #ifdef USE_LOCALE_NUMERIC 1510 1511 /* Toggle the LC_NUMERIC locale to the current underlying default. Most 1512 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h 1513 * instead of calling this directly. The macro avoids calling this routine 1514 * if toggling isn't necessary according to our records (which could be 1515 * wrong if some XS code has changed the locale behind our back) */ 1516 1517 # ifdef DEBUGGING 1518 1519 if (DEBUG_L_TEST || debug_initialization) { 1520 PerlIO_printf(Perl_debug_log, 1521 "Setting LC_NUMERIC locale to %s\n", 1522 PL_numeric_name); 1523 } 1524 1525 # endif 1526 1527 do_setlocale_c(LC_NUMERIC, PL_numeric_name); 1528 PL_numeric_standard = PL_numeric_underlying_is_standard; 1529 PL_numeric_underlying = TRUE; 1530 set_numeric_radix(! PL_numeric_standard); 1531 1532 #endif /* USE_LOCALE_NUMERIC */ 1533 1534 } 1535 1536 /* 1537 * Set up for a new ctype locale. 1538 */ 1539 STATIC void 1540 S_new_ctype(pTHX_ const char *newctype) 1541 { 1542 1543 #ifndef USE_LOCALE_CTYPE 1544 1545 PERL_UNUSED_ARG(newctype); 1546 PERL_UNUSED_CONTEXT; 1547 1548 #else 1549 1550 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell 1551 * core Perl this and that 'newctype' is the name of the new locale. 1552 * 1553 * This function sets up the folding arrays for all 256 bytes, assuming 1554 * that tofold() is tolc() since fold case is not a concept in POSIX, 1555 * 1556 * Any code changing the locale (outside this file) should use 1557 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore 1558 * this function should be called directly only from this file and from 1559 * POSIX::setlocale() */ 1560 1561 dVAR; 1562 unsigned int i; 1563 1564 /* Don't check for problems if we are suppressing the warnings */ 1565 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST); 1566 bool maybe_utf8_turkic = FALSE; 1567 1568 PERL_ARGS_ASSERT_NEW_CTYPE; 1569 1570 /* We will replace any bad locale warning with 1) nothing if the new one is 1571 * ok; or 2) a new warning for the bad new locale */ 1572 if (PL_warn_locale) { 1573 SvREFCNT_dec_NN(PL_warn_locale); 1574 PL_warn_locale = NULL; 1575 } 1576 1577 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE); 1578 1579 /* A UTF-8 locale gets standard rules. But note that code still has to 1580 * handle this specially because of the three problematic code points */ 1581 if (PL_in_utf8_CTYPE_locale) { 1582 Copy(PL_fold_latin1, PL_fold_locale, 256, U8); 1583 1584 /* UTF-8 locales can have special handling for 'I' and 'i' if they are 1585 * Turkic. Make sure these two are the only anomalies. (We don't use 1586 * towupper and towlower because they aren't in C89.) */ 1587 1588 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER) 1589 1590 if (towupper('i') == 0x130 && towlower('I') == 0x131) { 1591 1592 #else 1593 1594 if (toupper('i') == 'i' && tolower('I') == 'I') { 1595 1596 #endif 1597 check_for_problems = TRUE; 1598 maybe_utf8_turkic = TRUE; 1599 } 1600 } 1601 1602 /* We don't populate the other lists if a UTF-8 locale, but do check that 1603 * everything works as expected, unless checking turned off */ 1604 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) { 1605 /* Assume enough space for every character being bad. 4 spaces each 1606 * for the 94 printable characters that are output like "'x' "; and 5 1607 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating 1608 * NUL */ 1609 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' }; 1610 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale 1611 to start */ 1612 unsigned int bad_count = 0; /* Count of bad characters */ 1613 1614 for (i = 0; i < 256; i++) { 1615 if (! PL_in_utf8_CTYPE_locale) { 1616 if (isupper(i)) 1617 PL_fold_locale[i] = (U8) tolower(i); 1618 else if (islower(i)) 1619 PL_fold_locale[i] = (U8) toupper(i); 1620 else 1621 PL_fold_locale[i] = (U8) i; 1622 } 1623 1624 /* If checking for locale problems, see if the native ASCII-range 1625 * printables plus \n and \t are in their expected categories in 1626 * the new locale. If not, this could mean big trouble, upending 1627 * Perl's and most programs' assumptions, like having a 1628 * metacharacter with special meaning become a \w. Fortunately, 1629 * it's very rare to find locales that aren't supersets of ASCII 1630 * nowadays. It isn't a problem for most controls to be changed 1631 * into something else; we check only \n and \t, though perhaps \r 1632 * could be an issue as well. */ 1633 if ( check_for_problems 1634 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n')) 1635 { 1636 bool is_bad = FALSE; 1637 char name[4] = { '\0' }; 1638 1639 /* Convert the name into a string */ 1640 if (isGRAPH_A(i)) { 1641 name[0] = i; 1642 name[1] = '\0'; 1643 } 1644 else if (i == '\n') { 1645 my_strlcpy(name, "\\n", sizeof(name)); 1646 } 1647 else if (i == '\t') { 1648 my_strlcpy(name, "\\t", sizeof(name)); 1649 } 1650 else { 1651 assert(i == ' '); 1652 my_strlcpy(name, "' '", sizeof(name)); 1653 } 1654 1655 /* Check each possibe class */ 1656 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) { 1657 is_bad = TRUE; 1658 DEBUG_L(PerlIO_printf(Perl_debug_log, 1659 "isalnum('%s') unexpectedly is %d\n", 1660 name, cBOOL(isalnum(i)))); 1661 } 1662 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) { 1663 is_bad = TRUE; 1664 DEBUG_L(PerlIO_printf(Perl_debug_log, 1665 "isalpha('%s') unexpectedly is %d\n", 1666 name, cBOOL(isalpha(i)))); 1667 } 1668 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) { 1669 is_bad = TRUE; 1670 DEBUG_L(PerlIO_printf(Perl_debug_log, 1671 "isdigit('%s') unexpectedly is %d\n", 1672 name, cBOOL(isdigit(i)))); 1673 } 1674 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) { 1675 is_bad = TRUE; 1676 DEBUG_L(PerlIO_printf(Perl_debug_log, 1677 "isgraph('%s') unexpectedly is %d\n", 1678 name, cBOOL(isgraph(i)))); 1679 } 1680 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) { 1681 is_bad = TRUE; 1682 DEBUG_L(PerlIO_printf(Perl_debug_log, 1683 "islower('%s') unexpectedly is %d\n", 1684 name, cBOOL(islower(i)))); 1685 } 1686 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) { 1687 is_bad = TRUE; 1688 DEBUG_L(PerlIO_printf(Perl_debug_log, 1689 "isprint('%s') unexpectedly is %d\n", 1690 name, cBOOL(isprint(i)))); 1691 } 1692 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) { 1693 is_bad = TRUE; 1694 DEBUG_L(PerlIO_printf(Perl_debug_log, 1695 "ispunct('%s') unexpectedly is %d\n", 1696 name, cBOOL(ispunct(i)))); 1697 } 1698 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) { 1699 is_bad = TRUE; 1700 DEBUG_L(PerlIO_printf(Perl_debug_log, 1701 "isspace('%s') unexpectedly is %d\n", 1702 name, cBOOL(isspace(i)))); 1703 } 1704 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) { 1705 is_bad = TRUE; 1706 DEBUG_L(PerlIO_printf(Perl_debug_log, 1707 "isupper('%s') unexpectedly is %d\n", 1708 name, cBOOL(isupper(i)))); 1709 } 1710 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) { 1711 is_bad = TRUE; 1712 DEBUG_L(PerlIO_printf(Perl_debug_log, 1713 "isxdigit('%s') unexpectedly is %d\n", 1714 name, cBOOL(isxdigit(i)))); 1715 } 1716 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) { 1717 is_bad = TRUE; 1718 DEBUG_L(PerlIO_printf(Perl_debug_log, 1719 "tolower('%s')=0x%x instead of the expected 0x%x\n", 1720 name, tolower(i), (int) toLOWER_A(i))); 1721 } 1722 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) { 1723 is_bad = TRUE; 1724 DEBUG_L(PerlIO_printf(Perl_debug_log, 1725 "toupper('%s')=0x%x instead of the expected 0x%x\n", 1726 name, toupper(i), (int) toUPPER_A(i))); 1727 } 1728 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) { 1729 is_bad = TRUE; 1730 DEBUG_L(PerlIO_printf(Perl_debug_log, 1731 "'\\n' (=%02X) is not a control\n", (int) i)); 1732 } 1733 1734 /* Add to the list; Separate multiple entries with a blank */ 1735 if (is_bad) { 1736 if (bad_count) { 1737 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list)); 1738 } 1739 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list)); 1740 bad_count++; 1741 } 1742 } 1743 } 1744 1745 if (bad_count == 2 && maybe_utf8_turkic) { 1746 bad_count = 0; 1747 *bad_chars_list = '\0'; 1748 PL_fold_locale['I'] = 'I'; 1749 PL_fold_locale['i'] = 'i'; 1750 PL_in_utf8_turkic_locale = TRUE; 1751 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n", 1752 __FILE__, __LINE__, newctype)); 1753 } 1754 else { 1755 PL_in_utf8_turkic_locale = FALSE; 1756 } 1757 1758 # ifdef MB_CUR_MAX 1759 1760 /* We only handle single-byte locales (outside of UTF-8 ones; so if 1761 * this locale requires more than one byte, there are going to be 1762 * problems. */ 1763 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 1764 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n", 1765 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX)); 1766 1767 if ( check_for_problems && MB_CUR_MAX > 1 1768 && ! PL_in_utf8_CTYPE_locale 1769 1770 /* Some platforms return MB_CUR_MAX > 1 for even the "C" 1771 * locale. Just assume that the implementation for them (plus 1772 * for POSIX) is correct and the > 1 value is spurious. (Since 1773 * these are specially handled to never be considered UTF-8 1774 * locales, as long as this is the only problem, everything 1775 * should work fine */ 1776 && strNE(newctype, "C") && strNE(newctype, "POSIX")) 1777 { 1778 multi_byte_locale = TRUE; 1779 } 1780 1781 # endif 1782 1783 /* If we found problems and we want them output, do so */ 1784 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) 1785 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST))) 1786 { 1787 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) { 1788 PL_warn_locale = Perl_newSVpvf(aTHX_ 1789 "Locale '%s' contains (at least) the following characters" 1790 " which have\nunexpected meanings: %s\nThe Perl program" 1791 " will use the expected meanings", 1792 newctype, bad_chars_list); 1793 } 1794 else { 1795 PL_warn_locale = Perl_newSVpvf(aTHX_ 1796 "Locale '%s' may not work well.%s%s%s\n", 1797 newctype, 1798 (multi_byte_locale) 1799 ? " Some characters in it are not recognized by" 1800 " Perl." 1801 : "", 1802 (bad_count) 1803 ? "\nThe following characters (and maybe others)" 1804 " may not have the same meaning as the Perl" 1805 " program expects:\n" 1806 : "", 1807 (bad_count) 1808 ? bad_chars_list 1809 : "" 1810 ); 1811 } 1812 1813 # ifdef HAS_NL_LANGINFO 1814 1815 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s", 1816 /* parameter FALSE is a don't care here */ 1817 my_nl_langinfo(CODESET, FALSE)); 1818 1819 # endif 1820 1821 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n"); 1822 1823 /* If we are actually in the scope of the locale or are debugging, 1824 * output the message now. If not in that scope, we save the 1825 * message to be output at the first operation using this locale, 1826 * if that actually happens. Most programs don't use locales, so 1827 * they are immune to bad ones. */ 1828 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) { 1829 1830 /* The '0' below suppresses a bogus gcc compiler warning */ 1831 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0); 1832 1833 if (IN_LC(LC_CTYPE)) { 1834 SvREFCNT_dec_NN(PL_warn_locale); 1835 PL_warn_locale = NULL; 1836 } 1837 } 1838 } 1839 } 1840 1841 #endif /* USE_LOCALE_CTYPE */ 1842 1843 } 1844 1845 void 1846 Perl__warn_problematic_locale() 1847 { 1848 1849 #ifdef USE_LOCALE_CTYPE 1850 1851 dTHX; 1852 1853 /* Internal-to-core function that outputs the message in PL_warn_locale, 1854 * and then NULLS it. Should be called only through the macro 1855 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */ 1856 1857 if (PL_warn_locale) { 1858 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), 1859 SvPVX(PL_warn_locale), 1860 0 /* dummy to avoid compiler warning */ ); 1861 SvREFCNT_dec_NN(PL_warn_locale); 1862 PL_warn_locale = NULL; 1863 } 1864 1865 #endif 1866 1867 } 1868 1869 STATIC void 1870 S_new_collate(pTHX_ const char *newcoll) 1871 { 1872 1873 #ifndef USE_LOCALE_COLLATE 1874 1875 PERL_UNUSED_ARG(newcoll); 1876 PERL_UNUSED_CONTEXT; 1877 1878 #else 1879 1880 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell 1881 * core Perl this and that 'newcoll' is the name of the new locale. 1882 * 1883 * The design of locale collation is that every locale change is given an 1884 * index 'PL_collation_ix'. The first time a string particpates in an 1885 * operation that requires collation while locale collation is active, it 1886 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That 1887 * magic includes the collation index, and the transformation of the string 1888 * by strxfrm(), q.v. That transformation is used when doing comparisons, 1889 * instead of the string itself. If a string changes, the magic is 1890 * cleared. The next time the locale changes, the index is incremented, 1891 * and so we know during a comparison that the transformation is not 1892 * necessarily still valid, and so is recomputed. Note that if the locale 1893 * changes enough times, the index could wrap (a U32), and it is possible 1894 * that a transformation would improperly be considered valid, leading to 1895 * an unlikely bug */ 1896 1897 if (! newcoll) { 1898 if (PL_collation_name) { 1899 ++PL_collation_ix; 1900 Safefree(PL_collation_name); 1901 PL_collation_name = NULL; 1902 } 1903 PL_collation_standard = TRUE; 1904 is_standard_collation: 1905 PL_collxfrm_base = 0; 1906 PL_collxfrm_mult = 2; 1907 PL_in_utf8_COLLATE_locale = FALSE; 1908 PL_strxfrm_NUL_replacement = '\0'; 1909 PL_strxfrm_max_cp = 0; 1910 return; 1911 } 1912 1913 /* If this is not the same locale as currently, set the new one up */ 1914 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) { 1915 ++PL_collation_ix; 1916 Safefree(PL_collation_name); 1917 PL_collation_name = stdize_locale(savepv(newcoll)); 1918 PL_collation_standard = isNAME_C_OR_POSIX(newcoll); 1919 if (PL_collation_standard) { 1920 goto is_standard_collation; 1921 } 1922 1923 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE); 1924 PL_strxfrm_NUL_replacement = '\0'; 1925 PL_strxfrm_max_cp = 0; 1926 1927 /* A locale collation definition includes primary, secondary, tertiary, 1928 * etc. weights for each character. To sort, the primary weights are 1929 * used, and only if they compare equal, then the secondary weights are 1930 * used, and only if they compare equal, then the tertiary, etc. 1931 * 1932 * strxfrm() works by taking the input string, say ABC, and creating an 1933 * output transformed string consisting of first the primary weights, 1934 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the 1935 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters 1936 * may not have weights at every level. In our example, let's say B 1937 * doesn't have a tertiary weight, and A doesn't have a secondary 1938 * weight. The constructed string is then going to be 1939 * A¹B¹C¹ B²C² A³C³ .... 1940 * This has the desired effect that strcmp() will look at the secondary 1941 * or tertiary weights only if the strings compare equal at all higher 1942 * priority weights. The spaces shown here, like in 1943 * "A¹B¹C¹ A²B²C² " 1944 * are not just for readability. In the general case, these must 1945 * actually be bytes, which we will call here 'separator weights'; and 1946 * they must be smaller than any other weight value, but since these 1947 * are C strings, only the terminating one can be a NUL (some 1948 * implementations may include a non-NUL separator weight just before 1949 * the NUL). Implementations tend to reserve 01 for the separator 1950 * weights. They are needed so that a shorter string's secondary 1951 * weights won't be misconstrued as primary weights of a longer string, 1952 * etc. By making them smaller than any other weight, the shorter 1953 * string will sort first. (Actually, if all secondary weights are 1954 * smaller than all primary ones, there is no need for a separator 1955 * weight between those two levels, etc.) 1956 * 1957 * The length of the transformed string is roughly a linear function of 1958 * the input string. It's not exactly linear because some characters 1959 * don't have weights at all levels. When we call strxfrm() we have to 1960 * allocate some memory to hold the transformed string. The 1961 * calculations below try to find coefficients 'm' and 'b' for this 1962 * locale so that m*x + b equals how much space we need, given the size 1963 * of the input string in 'x'. If we calculate too small, we increase 1964 * the size as needed, and call strxfrm() again, but it is better to 1965 * get it right the first time to avoid wasted expensive string 1966 * transformations. */ 1967 1968 { 1969 /* We use the string below to find how long the tranformation of it 1970 * is. Almost all locales are supersets of ASCII, or at least the 1971 * ASCII letters. We use all of them, half upper half lower, 1972 * because if we used fewer, we might hit just the ones that are 1973 * outliers in a particular locale. Most of the strings being 1974 * collated will contain a preponderance of letters, and even if 1975 * they are above-ASCII, they are likely to have the same number of 1976 * weight levels as the ASCII ones. It turns out that digits tend 1977 * to have fewer levels, and some punctuation has more, but those 1978 * are relatively sparse in text, and khw believes this gives a 1979 * reasonable result, but it could be changed if experience so 1980 * dictates. */ 1981 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz"; 1982 char * x_longer; /* Transformed 'longer' */ 1983 Size_t x_len_longer; /* Length of 'x_longer' */ 1984 1985 char * x_shorter; /* We also transform a substring of 'longer' */ 1986 Size_t x_len_shorter; 1987 1988 /* _mem_collxfrm() is used get the transformation (though here we 1989 * are interested only in its length). It is used because it has 1990 * the intelligence to handle all cases, but to work, it needs some 1991 * values of 'm' and 'b' to get it started. For the purposes of 1992 * this calculation we use a very conservative estimate of 'm' and 1993 * 'b'. This assumes a weight can be multiple bytes, enough to 1994 * hold any UV on the platform, and there are 5 levels, 4 weight 1995 * bytes, and a trailing NUL. */ 1996 PL_collxfrm_base = 5; 1997 PL_collxfrm_mult = 5 * sizeof(UV); 1998 1999 /* Find out how long the transformation really is */ 2000 x_longer = _mem_collxfrm(longer, 2001 sizeof(longer) - 1, 2002 &x_len_longer, 2003 2004 /* We avoid converting to UTF-8 in the 2005 * called function by telling it the 2006 * string is in UTF-8 if the locale is a 2007 * UTF-8 one. Since the string passed 2008 * here is invariant under UTF-8, we can 2009 * claim it's UTF-8 even though it isn't. 2010 * */ 2011 PL_in_utf8_COLLATE_locale); 2012 Safefree(x_longer); 2013 2014 /* Find out how long the transformation of a substring of 'longer' 2015 * is. Together the lengths of these transformations are 2016 * sufficient to calculate 'm' and 'b'. The substring is all of 2017 * 'longer' except the first character. This minimizes the chances 2018 * of being swayed by outliers */ 2019 x_shorter = _mem_collxfrm(longer + 1, 2020 sizeof(longer) - 2, 2021 &x_len_shorter, 2022 PL_in_utf8_COLLATE_locale); 2023 Safefree(x_shorter); 2024 2025 /* If the results are nonsensical for this simple test, the whole 2026 * locale definition is suspect. Mark it so that locale collation 2027 * is not active at all for it. XXX Should we warn? */ 2028 if ( x_len_shorter == 0 2029 || x_len_longer == 0 2030 || x_len_shorter >= x_len_longer) 2031 { 2032 PL_collxfrm_mult = 0; 2033 PL_collxfrm_base = 0; 2034 } 2035 else { 2036 SSize_t base; /* Temporary */ 2037 2038 /* We have both: m * strlen(longer) + b = x_len_longer 2039 * m * strlen(shorter) + b = x_len_shorter; 2040 * subtracting yields: 2041 * m * (strlen(longer) - strlen(shorter)) 2042 * = x_len_longer - x_len_shorter 2043 * But we have set things up so that 'shorter' is 1 byte smaller 2044 * than 'longer'. Hence: 2045 * m = x_len_longer - x_len_shorter 2046 * 2047 * But if something went wrong, make sure the multiplier is at 2048 * least 1. 2049 */ 2050 if (x_len_longer > x_len_shorter) { 2051 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter; 2052 } 2053 else { 2054 PL_collxfrm_mult = 1; 2055 } 2056 2057 /* mx + b = len 2058 * so: b = len - mx 2059 * but in case something has gone wrong, make sure it is 2060 * non-negative */ 2061 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1); 2062 if (base < 0) { 2063 base = 0; 2064 } 2065 2066 /* Add 1 for the trailing NUL */ 2067 PL_collxfrm_base = base + 1; 2068 } 2069 2070 # ifdef DEBUGGING 2071 2072 if (DEBUG_L_TEST || debug_initialization) { 2073 PerlIO_printf(Perl_debug_log, 2074 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, " 2075 "x_len_longer=%zu," 2076 " collate multipler=%zu, collate base=%zu\n", 2077 __FILE__, __LINE__, 2078 PL_in_utf8_COLLATE_locale, 2079 x_len_shorter, x_len_longer, 2080 PL_collxfrm_mult, PL_collxfrm_base); 2081 } 2082 # endif 2083 2084 } 2085 } 2086 2087 #endif /* USE_LOCALE_COLLATE */ 2088 2089 } 2090 2091 #endif 2092 2093 #ifdef WIN32 2094 2095 #define USE_WSETLOCALE 2096 2097 #ifdef USE_WSETLOCALE 2098 2099 STATIC char * 2100 S_wrap_wsetlocale(pTHX_ int category, const char *locale) { 2101 wchar_t *wlocale; 2102 wchar_t *wresult; 2103 char *result; 2104 2105 if (locale) { 2106 int req_size = 2107 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0); 2108 2109 if (!req_size) { 2110 errno = EINVAL; 2111 return NULL; 2112 } 2113 2114 Newx(wlocale, req_size, wchar_t); 2115 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) { 2116 Safefree(wlocale); 2117 errno = EINVAL; 2118 return NULL; 2119 } 2120 } 2121 else { 2122 wlocale = NULL; 2123 } 2124 wresult = _wsetlocale(category, wlocale); 2125 Safefree(wlocale); 2126 if (wresult) { 2127 int req_size = 2128 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL); 2129 Newx(result, req_size, char); 2130 SAVEFREEPV(result); /* is there something better we can do here? */ 2131 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1, 2132 result, req_size, NULL, NULL)) { 2133 errno = EINVAL; 2134 return NULL; 2135 } 2136 } 2137 else { 2138 result = NULL; 2139 } 2140 2141 return result; 2142 } 2143 2144 #endif 2145 2146 STATIC char * 2147 S_win32_setlocale(pTHX_ int category, const char* locale) 2148 { 2149 /* This, for Windows, emulates POSIX setlocale() behavior. There is no 2150 * difference between the two unless the input locale is "", which normally 2151 * means on Windows to get the machine default, which is set via the 2152 * computer's "Regional and Language Options" (or its current equivalent). 2153 * In POSIX, it instead means to find the locale from the user's 2154 * environment. This routine changes the Windows behavior to first look in 2155 * the environment, and, if anything is found, use that instead of going to 2156 * the machine default. If there is no environment override, the machine 2157 * default is used, by calling the real setlocale() with "". 2158 * 2159 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to 2160 * use the particular category's variable if set; otherwise to use the LANG 2161 * variable. */ 2162 2163 bool override_LC_ALL = FALSE; 2164 char * result; 2165 unsigned int i; 2166 2167 if (locale && strEQ(locale, "")) { 2168 2169 # ifdef LC_ALL 2170 2171 locale = PerlEnv_getenv("LC_ALL"); 2172 if (! locale) { 2173 if (category == LC_ALL) { 2174 override_LC_ALL = TRUE; 2175 } 2176 else { 2177 2178 # endif 2179 2180 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) { 2181 if (category == categories[i]) { 2182 locale = PerlEnv_getenv(category_names[i]); 2183 goto found_locale; 2184 } 2185 } 2186 2187 locale = PerlEnv_getenv("LANG"); 2188 if (! locale) { 2189 locale = ""; 2190 } 2191 2192 found_locale: ; 2193 2194 # ifdef LC_ALL 2195 2196 } 2197 } 2198 2199 # endif 2200 2201 } 2202 2203 #ifdef USE_WSETLOCALE 2204 result = S_wrap_wsetlocale(aTHX_ category, locale); 2205 #else 2206 result = setlocale(category, locale); 2207 #endif 2208 DEBUG_L(STMT_START { 2209 dSAVE_ERRNO; 2210 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__, 2211 setlocale_debug_string(category, locale, result)); 2212 RESTORE_ERRNO; 2213 } STMT_END); 2214 2215 if (! override_LC_ALL) { 2216 return result; 2217 } 2218 2219 /* Here the input category was LC_ALL, and we have set it to what is in the 2220 * LANG variable or the system default if there is no LANG. But these have 2221 * lower priority than the other LC_foo variables, so override it for each 2222 * one that is set. (If they are set to "", it means to use the same thing 2223 * we just set LC_ALL to, so can skip) */ 2224 2225 for (i = 0; i < LC_ALL_INDEX; i++) { 2226 result = PerlEnv_getenv(category_names[i]); 2227 if (result && strNE(result, "")) { 2228 #ifdef USE_WSETLOCALE 2229 S_wrap_wsetlocale(aTHX_ categories[i], result); 2230 #else 2231 setlocale(categories[i], result); 2232 #endif 2233 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", 2234 __FILE__, __LINE__, 2235 setlocale_debug_string(categories[i], result, "not captured"))); 2236 } 2237 } 2238 2239 result = setlocale(LC_ALL, NULL); 2240 DEBUG_L(STMT_START { 2241 dSAVE_ERRNO; 2242 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", 2243 __FILE__, __LINE__, 2244 setlocale_debug_string(LC_ALL, NULL, result)); 2245 RESTORE_ERRNO; 2246 } STMT_END); 2247 2248 return result; 2249 } 2250 2251 #endif 2252 2253 /* 2254 2255 =head1 Locale-related functions and macros 2256 2257 =for apidoc Perl_setlocale 2258 2259 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>, 2260 taking the same parameters, and returning the same information, except that it 2261 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will 2262 instead return C<C> if the underlying locale has a non-dot decimal point 2263 character, or a non-empty thousands separator for displaying floating point 2264 numbers. This is because perl keeps that locale category such that it has a 2265 dot and empty separator, changing the locale briefly during the operations 2266 where the underlying one is required. C<Perl_setlocale> knows about this, and 2267 compensates; regular C<setlocale> doesn't. 2268 2269 Another reason it isn't completely a drop-in replacement is that it is 2270 declared to return S<C<const char *>>, whereas the system setlocale omits the 2271 C<const> (presumably because its API was specified long ago, and can't be 2272 updated; it is illegal to change the information C<setlocale> returns; doing 2273 so leads to segfaults.) 2274 2275 Finally, C<Perl_setlocale> works under all circumstances, whereas plain 2276 C<setlocale> can be completely ineffective on some platforms under some 2277 configurations. 2278 2279 C<Perl_setlocale> should not be used to change the locale except on systems 2280 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems, 2281 the system C<setlocale()> is ineffective, returning the wrong information, and 2282 failing to actually change the locale. C<Perl_setlocale>, however works 2283 properly in all circumstances. 2284 2285 The return points to a per-thread static buffer, which is overwritten the next 2286 time C<Perl_setlocale> is called from the same thread. 2287 2288 =cut 2289 2290 */ 2291 2292 const char * 2293 Perl_setlocale(const int category, const char * locale) 2294 { 2295 /* This wraps POSIX::setlocale() */ 2296 2297 #ifndef USE_LOCALE 2298 2299 PERL_UNUSED_ARG(category); 2300 PERL_UNUSED_ARG(locale); 2301 2302 return "C"; 2303 2304 #else 2305 2306 const char * retval; 2307 const char * newlocale; 2308 dSAVEDERRNO; 2309 dTHX; 2310 DECLARATION_FOR_LC_NUMERIC_MANIPULATION; 2311 2312 #ifdef USE_LOCALE_NUMERIC 2313 2314 /* A NULL locale means only query what the current one is. We have the 2315 * LC_NUMERIC name saved, because we are normally switched into the C 2316 * (or equivalent) locale for it. For an LC_ALL query, switch back to get 2317 * the correct results. All other categories don't require special 2318 * handling */ 2319 if (locale == NULL) { 2320 if (category == LC_NUMERIC) { 2321 2322 /* We don't have to copy this return value, as it is a per-thread 2323 * variable, and won't change until a future setlocale */ 2324 return PL_numeric_name; 2325 } 2326 2327 # ifdef LC_ALL 2328 2329 else if (category == LC_ALL) { 2330 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); 2331 } 2332 2333 # endif 2334 2335 } 2336 2337 #endif 2338 2339 retval = save_to_buffer(do_setlocale_r(category, locale), 2340 &PL_setlocale_buf, &PL_setlocale_bufsize, 0); 2341 SAVE_ERRNO; 2342 2343 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL) 2344 2345 if (locale == NULL && category == LC_ALL) { 2346 RESTORE_LC_NUMERIC(); 2347 } 2348 2349 #endif 2350 2351 DEBUG_L(PerlIO_printf(Perl_debug_log, 2352 "%s:%d: %s\n", __FILE__, __LINE__, 2353 setlocale_debug_string(category, locale, retval))); 2354 2355 RESTORE_ERRNO; 2356 2357 if (! retval) { 2358 return NULL; 2359 } 2360 2361 /* If locale == NULL, we are just querying the state */ 2362 if (locale == NULL) { 2363 return retval; 2364 } 2365 2366 /* Now that have switched locales, we have to update our records to 2367 * correspond. */ 2368 2369 switch (category) { 2370 2371 #ifdef USE_LOCALE_CTYPE 2372 2373 case LC_CTYPE: 2374 new_ctype(retval); 2375 break; 2376 2377 #endif 2378 #ifdef USE_LOCALE_COLLATE 2379 2380 case LC_COLLATE: 2381 new_collate(retval); 2382 break; 2383 2384 #endif 2385 #ifdef USE_LOCALE_NUMERIC 2386 2387 case LC_NUMERIC: 2388 new_numeric(retval); 2389 break; 2390 2391 #endif 2392 #ifdef LC_ALL 2393 2394 case LC_ALL: 2395 2396 /* LC_ALL updates all the things we care about. The values may not 2397 * be the same as 'retval', as the locale "" may have set things 2398 * individually */ 2399 2400 # ifdef USE_LOCALE_CTYPE 2401 2402 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL)); 2403 new_ctype(newlocale); 2404 Safefree(newlocale); 2405 2406 # endif /* USE_LOCALE_CTYPE */ 2407 # ifdef USE_LOCALE_COLLATE 2408 2409 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL)); 2410 new_collate(newlocale); 2411 Safefree(newlocale); 2412 2413 # endif 2414 # ifdef USE_LOCALE_NUMERIC 2415 2416 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL)); 2417 new_numeric(newlocale); 2418 Safefree(newlocale); 2419 2420 # endif /* USE_LOCALE_NUMERIC */ 2421 #endif /* LC_ALL */ 2422 2423 default: 2424 break; 2425 } 2426 2427 return retval; 2428 2429 #endif 2430 2431 } 2432 2433 PERL_STATIC_INLINE const char * 2434 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset) 2435 { 2436 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size', 2437 * growing it if necessary */ 2438 2439 Size_t string_size; 2440 2441 PERL_ARGS_ASSERT_SAVE_TO_BUFFER; 2442 2443 if (! string) { 2444 return NULL; 2445 } 2446 2447 string_size = strlen(string) + offset + 1; 2448 2449 if (*buf_size == 0) { 2450 Newx(*buf, string_size, char); 2451 *buf_size = string_size; 2452 } 2453 else if (string_size > *buf_size) { 2454 Renew(*buf, string_size, char); 2455 *buf_size = string_size; 2456 } 2457 2458 Copy(string, *buf + offset, string_size - offset, char); 2459 return *buf; 2460 } 2461 2462 /* 2463 2464 =for apidoc Perl_langinfo 2465 2466 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>, 2467 taking the same C<item> parameter values, and returning the same information. 2468 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks 2469 of Perl's locale handling from your code, and can be used on systems that lack 2470 a native C<nl_langinfo>. 2471 2472 Expanding on these: 2473 2474 =over 2475 2476 =item * 2477 2478 The reason it isn't quite a drop-in replacement is actually an advantage. The 2479 only difference is that it returns S<C<const char *>>, whereas plain 2480 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation) 2481 forbidden to write into the buffer. By declaring this C<const>, the compiler 2482 enforces this restriction, so if it is violated, you know at compilation time, 2483 rather than getting segfaults at runtime. 2484 2485 =item * 2486 2487 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items, 2488 without you having to write extra code. The reason for the extra code would be 2489 because these are from the C<LC_NUMERIC> locale category, which is normally 2490 kept set by Perl so that the radix is a dot, and the separator is the empty 2491 string, no matter what the underlying locale is supposed to be, and so to get 2492 the expected results, you have to temporarily toggle into the underlying 2493 locale, and later toggle back. (You could use plain C<nl_langinfo> and 2494 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get 2495 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C 2496 (or equivalent) locale would break a lot of CPAN, which is expecting the radix 2497 (decimal point) character to be a dot.) 2498 2499 =item * 2500 2501 The system function it replaces can have its static return buffer trashed, 2502 not only by a subesequent call to that function, but by a C<freelocale>, 2503 C<setlocale>, or other locale change. The returned buffer of this function is 2504 not changed until the next call to it, so the buffer is never in a trashed 2505 state. 2506 2507 =item * 2508 2509 Its return buffer is per-thread, so it also is never overwritten by a call to 2510 this function from another thread; unlike the function it replaces. 2511 2512 =item * 2513 2514 But most importantly, it works on systems that don't have C<nl_langinfo>, such 2515 as Windows, hence makes your code more portable. Of the fifty-some possible 2516 items specified by the POSIX 2008 standard, 2517 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>, 2518 only one is completely unimplemented, though on non-Windows platforms, another 2519 significant one is also not implemented). It uses various techniques to 2520 recover the other items, including calling C<L<localeconv(3)>>, and 2521 C<L<strftime(3)>>, both of which are specified in C89, so should be always be 2522 available. Later C<strftime()> versions have additional capabilities; C<""> is 2523 returned for those not available on your system. 2524 2525 It is important to note that when called with an item that is recovered by 2526 using C<localeconv>, the buffer from any previous explicit call to 2527 C<localeconv> will be overwritten. This means you must save that buffer's 2528 contents if you need to access them after a call to this function. (But note 2529 that you might not want to be using C<localeconv()> directly anyway, because of 2530 issues like the ones listed in the second item of this list (above) for 2531 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to 2532 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to 2533 unpack). 2534 2535 The details for those items which may deviate from what this emulation returns 2536 and what a native C<nl_langinfo()> would return are specified in 2537 L<I18N::Langinfo>. 2538 2539 =back 2540 2541 When using C<Perl_langinfo> on systems that don't have a native 2542 C<nl_langinfo()>, you must 2543 2544 #include "perl_langinfo.h" 2545 2546 before the C<perl.h> C<#include>. You can replace your C<langinfo.h> 2547 C<#include> with this one. (Doing it this way keeps out the symbols that plain 2548 C<langinfo.h> would try to import into the namespace for code that doesn't need 2549 it.) 2550 2551 The original impetus for C<Perl_langinfo()> was so that code that needs to 2552 find out the current currency symbol, floating point radix character, or digit 2553 grouping separator can use, on all systems, the simpler and more 2554 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a 2555 pain to make thread-friendly. For other fields returned by C<localeconv>, it 2556 is better to use the methods given in L<perlcall> to call 2557 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly. 2558 2559 =cut 2560 2561 */ 2562 2563 const char * 2564 #ifdef HAS_NL_LANGINFO 2565 Perl_langinfo(const nl_item item) 2566 #else 2567 Perl_langinfo(const int item) 2568 #endif 2569 { 2570 return my_nl_langinfo(item, TRUE); 2571 } 2572 2573 STATIC const char * 2574 #ifdef HAS_NL_LANGINFO 2575 S_my_nl_langinfo(const nl_item item, bool toggle) 2576 #else 2577 S_my_nl_langinfo(const int item, bool toggle) 2578 #endif 2579 { 2580 dTHX; 2581 const char * retval; 2582 2583 #ifdef USE_LOCALE_NUMERIC 2584 2585 /* We only need to toggle into the underlying LC_NUMERIC locale for these 2586 * two items, and only if not already there */ 2587 if (toggle && (( item != RADIXCHAR && item != THOUSEP) 2588 || PL_numeric_underlying)) 2589 2590 #endif /* No toggling needed if not using LC_NUMERIC */ 2591 2592 toggle = FALSE; 2593 2594 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */ 2595 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \ 2596 || ! defined(HAS_POSIX_2008_LOCALE) \ 2597 || ! defined(DUPLOCALE) 2598 2599 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC 2600 * for those items dependent on it. This must be copied to a buffer before 2601 * switching back, as some systems destroy the buffer when setlocale() is 2602 * called */ 2603 2604 { 2605 DECLARATION_FOR_LC_NUMERIC_MANIPULATION; 2606 2607 if (toggle) { 2608 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); 2609 } 2610 2611 LOCALE_LOCK; /* Prevent interference from another thread executing 2612 this code section (the only call to nl_langinfo in 2613 the core) */ 2614 2615 2616 /* Copy to a per-thread buffer, which is also one that won't be 2617 * destroyed by a subsequent setlocale(), such as the 2618 * RESTORE_LC_NUMERIC may do just below. */ 2619 retval = save_to_buffer(nl_langinfo(item), 2620 &PL_langinfo_buf, &PL_langinfo_bufsize, 0); 2621 2622 LOCALE_UNLOCK; 2623 2624 if (toggle) { 2625 RESTORE_LC_NUMERIC(); 2626 } 2627 } 2628 2629 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */ 2630 2631 { 2632 bool do_free = FALSE; 2633 locale_t cur = uselocale((locale_t) 0); 2634 2635 if (cur == LC_GLOBAL_LOCALE) { 2636 cur = duplocale(LC_GLOBAL_LOCALE); 2637 do_free = TRUE; 2638 } 2639 2640 # ifdef USE_LOCALE_NUMERIC 2641 2642 if (toggle) { 2643 if (PL_underlying_numeric_obj) { 2644 cur = PL_underlying_numeric_obj; 2645 } 2646 else { 2647 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur); 2648 do_free = TRUE; 2649 } 2650 } 2651 2652 # endif 2653 2654 /* We have to save it to a buffer, because the freelocale() just below 2655 * can invalidate the internal one */ 2656 retval = save_to_buffer(nl_langinfo_l(item, cur), 2657 &PL_langinfo_buf, &PL_langinfo_bufsize, 0); 2658 2659 if (do_free) { 2660 freelocale(cur); 2661 } 2662 } 2663 2664 # endif 2665 2666 if (strEQ(retval, "")) { 2667 if (item == YESSTR) { 2668 return "yes"; 2669 } 2670 if (item == NOSTR) { 2671 return "no"; 2672 } 2673 } 2674 2675 return retval; 2676 2677 #else /* Below, emulate nl_langinfo as best we can */ 2678 2679 { 2680 2681 # ifdef HAS_LOCALECONV 2682 2683 const struct lconv* lc; 2684 const char * temp; 2685 DECLARATION_FOR_LC_NUMERIC_MANIPULATION; 2686 2687 # ifdef TS_W32_BROKEN_LOCALECONV 2688 2689 const char * save_global; 2690 const char * save_thread; 2691 int needed_size; 2692 char * ptr; 2693 char * e; 2694 char * item_start; 2695 2696 # endif 2697 # endif 2698 # ifdef HAS_STRFTIME 2699 2700 struct tm tm; 2701 bool return_format = FALSE; /* Return the %format, not the value */ 2702 const char * format; 2703 2704 # endif 2705 2706 /* We copy the results to a per-thread buffer, even if not 2707 * multi-threaded. This is in part to simplify this code, and partly 2708 * because we need a buffer anyway for strftime(), and partly because a 2709 * call of localeconv() could otherwise wipe out the buffer, and the 2710 * programmer would not be expecting this, as this is a nl_langinfo() 2711 * substitute after all, so s/he might be thinking their localeconv() 2712 * is safe until another localeconv() call. */ 2713 2714 switch (item) { 2715 Size_t len; 2716 2717 /* This is unimplemented */ 2718 case ERA: /* For use with strftime() %E modifier */ 2719 2720 default: 2721 return ""; 2722 2723 /* We use only an English set, since we don't know any more */ 2724 case YESEXPR: return "^[+1yY]"; 2725 case YESSTR: return "yes"; 2726 case NOEXPR: return "^[-0nN]"; 2727 case NOSTR: return "no"; 2728 2729 case CODESET: 2730 2731 # ifndef WIN32 2732 2733 /* On non-windows, this is unimplemented, in part because of 2734 * inconsistencies between vendors. The Darwin native 2735 * nl_langinfo() implementation simply looks at everything past 2736 * any dot in the name, but that doesn't work for other 2737 * vendors. Many Linux locales that don't have UTF-8 in their 2738 * names really are UTF-8, for example; z/OS locales that do 2739 * have UTF-8 in their names, aren't really UTF-8 */ 2740 return ""; 2741 2742 # else 2743 2744 { /* But on Windows, the name does seem to be consistent, so 2745 use that. */ 2746 const char * p; 2747 const char * first; 2748 Size_t offset = 0; 2749 const char * name = my_setlocale(LC_CTYPE, NULL); 2750 2751 if (isNAME_C_OR_POSIX(name)) { 2752 return "ANSI_X3.4-1968"; 2753 } 2754 2755 /* Find the dot in the locale name */ 2756 first = (const char *) strchr(name, '.'); 2757 if (! first) { 2758 first = name; 2759 goto has_nondigit; 2760 } 2761 2762 /* Look at everything past the dot */ 2763 first++; 2764 p = first; 2765 2766 while (*p) { 2767 if (! isDIGIT(*p)) { 2768 goto has_nondigit; 2769 } 2770 2771 p++; 2772 } 2773 2774 /* Here everything past the dot is a digit. Treat it as a 2775 * code page */ 2776 retval = save_to_buffer("CP", &PL_langinfo_buf, 2777 &PL_langinfo_bufsize, 0); 2778 offset = STRLENs("CP"); 2779 2780 has_nondigit: 2781 2782 retval = save_to_buffer(first, &PL_langinfo_buf, 2783 &PL_langinfo_bufsize, offset); 2784 } 2785 2786 break; 2787 2788 # endif 2789 # ifdef HAS_LOCALECONV 2790 2791 case CRNCYSTR: 2792 2793 /* We don't bother with localeconv_l() because any system that 2794 * has it is likely to also have nl_langinfo() */ 2795 2796 LOCALE_LOCK_V; /* Prevent interference with other threads 2797 using localeconv() */ 2798 2799 # ifdef TS_W32_BROKEN_LOCALECONV 2800 2801 /* This is a workaround for a Windows bug prior to VS 15. 2802 * What we do here is, while locked, switch to the global 2803 * locale so localeconv() works; then switch back just before 2804 * the unlock. This can screw things up if some thread is 2805 * already using the global locale while assuming no other is. 2806 * A different workaround would be to call GetCurrencyFormat on 2807 * a known value, and parse it; patches welcome 2808 * 2809 * We have to use LC_ALL instead of LC_MONETARY because of 2810 * another bug in Windows */ 2811 2812 save_thread = savepv(my_setlocale(LC_ALL, NULL)); 2813 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE); 2814 save_global= savepv(my_setlocale(LC_ALL, NULL)); 2815 my_setlocale(LC_ALL, save_thread); 2816 2817 # endif 2818 2819 lc = localeconv(); 2820 if ( ! lc 2821 || ! lc->currency_symbol 2822 || strEQ("", lc->currency_symbol)) 2823 { 2824 LOCALE_UNLOCK_V; 2825 return ""; 2826 } 2827 2828 /* Leave the first spot empty to be filled in below */ 2829 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf, 2830 &PL_langinfo_bufsize, 1); 2831 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, "")) 2832 { /* khw couldn't figure out how the localedef specifications 2833 would show that the $ should replace the radix; this is 2834 just a guess as to how it might work.*/ 2835 PL_langinfo_buf[0] = '.'; 2836 } 2837 else if (lc->p_cs_precedes) { 2838 PL_langinfo_buf[0] = '-'; 2839 } 2840 else { 2841 PL_langinfo_buf[0] = '+'; 2842 } 2843 2844 # ifdef TS_W32_BROKEN_LOCALECONV 2845 2846 my_setlocale(LC_ALL, save_global); 2847 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); 2848 my_setlocale(LC_ALL, save_thread); 2849 Safefree(save_global); 2850 Safefree(save_thread); 2851 2852 # endif 2853 2854 LOCALE_UNLOCK_V; 2855 break; 2856 2857 # ifdef TS_W32_BROKEN_LOCALECONV 2858 2859 case RADIXCHAR: 2860 2861 /* For this, we output a known simple floating point number to 2862 * a buffer, and parse it, looking for the radix */ 2863 2864 if (toggle) { 2865 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); 2866 } 2867 2868 if (PL_langinfo_bufsize < 10) { 2869 PL_langinfo_bufsize = 10; 2870 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char); 2871 } 2872 2873 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize, 2874 "%.1f", 1.5); 2875 if (needed_size >= (int) PL_langinfo_bufsize) { 2876 PL_langinfo_bufsize = needed_size + 1; 2877 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char); 2878 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize, 2879 "%.1f", 1.5); 2880 assert(needed_size < (int) PL_langinfo_bufsize); 2881 } 2882 2883 ptr = PL_langinfo_buf; 2884 e = PL_langinfo_buf + PL_langinfo_bufsize; 2885 2886 /* Find the '1' */ 2887 while (ptr < e && *ptr != '1') { 2888 ptr++; 2889 } 2890 ptr++; 2891 2892 /* Find the '5' */ 2893 item_start = ptr; 2894 while (ptr < e && *ptr != '5') { 2895 ptr++; 2896 } 2897 2898 /* Everything in between is the radix string */ 2899 if (ptr >= e) { 2900 PL_langinfo_buf[0] = '?'; 2901 PL_langinfo_buf[1] = '\0'; 2902 } 2903 else { 2904 *ptr = '\0'; 2905 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char); 2906 } 2907 2908 if (toggle) { 2909 RESTORE_LC_NUMERIC(); 2910 } 2911 2912 retval = PL_langinfo_buf; 2913 break; 2914 2915 # else 2916 2917 case RADIXCHAR: /* No special handling needed */ 2918 2919 # endif 2920 2921 case THOUSEP: 2922 2923 if (toggle) { 2924 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING(); 2925 } 2926 2927 LOCALE_LOCK_V; /* Prevent interference with other threads 2928 using localeconv() */ 2929 2930 # ifdef TS_W32_BROKEN_LOCALECONV 2931 2932 /* This should only be for the thousands separator. A 2933 * different work around would be to use GetNumberFormat on a 2934 * known value and parse the result to find the separator */ 2935 save_thread = savepv(my_setlocale(LC_ALL, NULL)); 2936 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE); 2937 save_global = savepv(my_setlocale(LC_ALL, NULL)); 2938 my_setlocale(LC_ALL, save_thread); 2939 # if 0 2940 /* This is the start of code that for broken Windows replaces 2941 * the above and below code, and instead calls 2942 * GetNumberFormat() and then would parse that to find the 2943 * thousands separator. It needs to handle UTF-16 vs -8 2944 * issues. */ 2945 2946 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize); 2947 DEBUG_L(PerlIO_printf(Perl_debug_log, 2948 "%s: %d: return from GetNumber, count=%d, val=%s\n", 2949 __FILE__, __LINE__, needed_size, PL_langinfo_buf)); 2950 2951 # endif 2952 # endif 2953 2954 lc = localeconv(); 2955 if (! lc) { 2956 temp = ""; 2957 } 2958 else { 2959 temp = (item == RADIXCHAR) 2960 ? lc->decimal_point 2961 : lc->thousands_sep; 2962 if (! temp) { 2963 temp = ""; 2964 } 2965 } 2966 2967 retval = save_to_buffer(temp, &PL_langinfo_buf, 2968 &PL_langinfo_bufsize, 0); 2969 2970 # ifdef TS_W32_BROKEN_LOCALECONV 2971 2972 my_setlocale(LC_ALL, save_global); 2973 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); 2974 my_setlocale(LC_ALL, save_thread); 2975 Safefree(save_global); 2976 Safefree(save_thread); 2977 2978 # endif 2979 2980 LOCALE_UNLOCK_V; 2981 2982 if (toggle) { 2983 RESTORE_LC_NUMERIC(); 2984 } 2985 2986 break; 2987 2988 # endif 2989 # ifdef HAS_STRFTIME 2990 2991 /* These are defined by C89, so we assume that strftime supports 2992 * them, and so are returned unconditionally; they may not be what 2993 * the locale actually says, but should give good enough results 2994 * for someone using them as formats (as opposed to trying to parse 2995 * them to figure out what the locale says). The other format 2996 * items are actually tested to verify they work on the platform */ 2997 case D_FMT: return "%x"; 2998 case T_FMT: return "%X"; 2999 case D_T_FMT: return "%c"; 3000 3001 /* These formats are only available in later strfmtime's */ 3002 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM: 3003 3004 /* The rest can be gotten from most versions of strftime(). */ 3005 case ABDAY_1: case ABDAY_2: case ABDAY_3: 3006 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7: 3007 case ALT_DIGITS: 3008 case AM_STR: case PM_STR: 3009 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4: 3010 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8: 3011 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12: 3012 case DAY_1: case DAY_2: case DAY_3: case DAY_4: 3013 case DAY_5: case DAY_6: case DAY_7: 3014 case MON_1: case MON_2: case MON_3: case MON_4: 3015 case MON_5: case MON_6: case MON_7: case MON_8: 3016 case MON_9: case MON_10: case MON_11: case MON_12: 3017 3018 LOCALE_LOCK; 3019 3020 init_tm(&tm); /* Precaution against core dumps */ 3021 tm.tm_sec = 30; 3022 tm.tm_min = 30; 3023 tm.tm_hour = 6; 3024 tm.tm_year = 2017 - 1900; 3025 tm.tm_wday = 0; 3026 tm.tm_mon = 0; 3027 switch (item) { 3028 default: 3029 LOCALE_UNLOCK; 3030 Perl_croak(aTHX_ 3031 "panic: %s: %d: switch case: %d problem", 3032 __FILE__, __LINE__, item); 3033 NOT_REACHED; /* NOTREACHED */ 3034 3035 case PM_STR: tm.tm_hour = 18; 3036 case AM_STR: 3037 format = "%p"; 3038 break; 3039 3040 case ABDAY_7: tm.tm_wday++; 3041 case ABDAY_6: tm.tm_wday++; 3042 case ABDAY_5: tm.tm_wday++; 3043 case ABDAY_4: tm.tm_wday++; 3044 case ABDAY_3: tm.tm_wday++; 3045 case ABDAY_2: tm.tm_wday++; 3046 case ABDAY_1: 3047 format = "%a"; 3048 break; 3049 3050 case DAY_7: tm.tm_wday++; 3051 case DAY_6: tm.tm_wday++; 3052 case DAY_5: tm.tm_wday++; 3053 case DAY_4: tm.tm_wday++; 3054 case DAY_3: tm.tm_wday++; 3055 case DAY_2: tm.tm_wday++; 3056 case DAY_1: 3057 format = "%A"; 3058 break; 3059 3060 case ABMON_12: tm.tm_mon++; 3061 case ABMON_11: tm.tm_mon++; 3062 case ABMON_10: tm.tm_mon++; 3063 case ABMON_9: tm.tm_mon++; 3064 case ABMON_8: tm.tm_mon++; 3065 case ABMON_7: tm.tm_mon++; 3066 case ABMON_6: tm.tm_mon++; 3067 case ABMON_5: tm.tm_mon++; 3068 case ABMON_4: tm.tm_mon++; 3069 case ABMON_3: tm.tm_mon++; 3070 case ABMON_2: tm.tm_mon++; 3071 case ABMON_1: 3072 format = "%b"; 3073 break; 3074 3075 case MON_12: tm.tm_mon++; 3076 case MON_11: tm.tm_mon++; 3077 case MON_10: tm.tm_mon++; 3078 case MON_9: tm.tm_mon++; 3079 case MON_8: tm.tm_mon++; 3080 case MON_7: tm.tm_mon++; 3081 case MON_6: tm.tm_mon++; 3082 case MON_5: tm.tm_mon++; 3083 case MON_4: tm.tm_mon++; 3084 case MON_3: tm.tm_mon++; 3085 case MON_2: tm.tm_mon++; 3086 case MON_1: 3087 format = "%B"; 3088 break; 3089 3090 case T_FMT_AMPM: 3091 format = "%r"; 3092 return_format = TRUE; 3093 break; 3094 3095 case ERA_D_FMT: 3096 format = "%Ex"; 3097 return_format = TRUE; 3098 break; 3099 3100 case ERA_T_FMT: 3101 format = "%EX"; 3102 return_format = TRUE; 3103 break; 3104 3105 case ERA_D_T_FMT: 3106 format = "%Ec"; 3107 return_format = TRUE; 3108 break; 3109 3110 case ALT_DIGITS: 3111 tm.tm_wday = 0; 3112 format = "%Ow"; /* Find the alternate digit for 0 */ 3113 break; 3114 } 3115 3116 /* We can't use my_strftime() because it doesn't look at 3117 * tm_wday */ 3118 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize, 3119 format, &tm)) 3120 { 3121 /* A zero return means one of: 3122 * a) there wasn't enough space in PL_langinfo_buf 3123 * b) the format, like a plain %p, returns empty 3124 * c) it was an illegal format, though some 3125 * implementations of strftime will just return the 3126 * illegal format as a plain character sequence. 3127 * 3128 * To quickly test for case 'b)', try again but precede 3129 * the format with a plain character. If that result is 3130 * still empty, the problem is either 'a)' or 'c)' */ 3131 3132 Size_t format_size = strlen(format) + 1; 3133 Size_t mod_size = format_size + 1; 3134 char * mod_format; 3135 char * temp_result; 3136 3137 Newx(mod_format, mod_size, char); 3138 Newx(temp_result, PL_langinfo_bufsize, char); 3139 *mod_format = ' '; 3140 my_strlcpy(mod_format + 1, format, mod_size); 3141 len = strftime(temp_result, 3142 PL_langinfo_bufsize, 3143 mod_format, &tm); 3144 Safefree(mod_format); 3145 Safefree(temp_result); 3146 3147 /* If 'len' is non-zero, it means that we had a case like 3148 * %p which means the current locale doesn't use a.m. or 3149 * p.m., and that is valid */ 3150 if (len == 0) { 3151 3152 /* Here, still didn't work. If we get well beyond a 3153 * reasonable size, bail out to prevent an infinite 3154 * loop. */ 3155 3156 if (PL_langinfo_bufsize > 100 * format_size) { 3157 *PL_langinfo_buf = '\0'; 3158 } 3159 else { 3160 /* Double the buffer size to retry; Add 1 in case 3161 * original was 0, so we aren't stuck at 0. */ 3162 PL_langinfo_bufsize *= 2; 3163 PL_langinfo_bufsize++; 3164 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char); 3165 continue; 3166 } 3167 } 3168 3169 break; 3170 } 3171 3172 /* Here, we got a result. 3173 * 3174 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the 3175 * alternate format for wday 0. If the value is the same as 3176 * the normal 0, there isn't an alternate, so clear the buffer. 3177 * */ 3178 if ( item == ALT_DIGITS 3179 && strEQ(PL_langinfo_buf, "0")) 3180 { 3181 *PL_langinfo_buf = '\0'; 3182 } 3183 3184 /* ALT_DIGITS is problematic. Experiments on it showed that 3185 * strftime() did not always work properly when going from 3186 * alt-9 to alt-10. Only a few locales have this item defined, 3187 * and in all of them on Linux that khw was able to find, 3188 * nl_langinfo() merely returned the alt-0 character, possibly 3189 * doubled. Most Unicode digits are in blocks of 10 3190 * consecutive code points, so that is sufficient information 3191 * for those scripts, as we can infer alt-1, alt-2, .... But 3192 * for a Japanese locale, a CJK ideographic 0 is returned, and 3193 * the CJK digits are not in code point order, so you can't 3194 * really infer anything. The localedef for this locale did 3195 * specify the succeeding digits, so that strftime() works 3196 * properly on them, without needing to infer anything. But 3197 * the nl_langinfo() return did not give sufficient information 3198 * for the caller to understand what's going on. So until 3199 * there is evidence that it should work differently, this 3200 * returns the alt-0 string for ALT_DIGITS. 3201 * 3202 * wday was chosen because its range is all a single digit. 3203 * Things like tm_sec have two digits as the minimum: '00' */ 3204 3205 LOCALE_UNLOCK; 3206 3207 retval = PL_langinfo_buf; 3208 3209 /* If to return the format, not the value, overwrite the buffer 3210 * with it. But some strftime()s will keep the original format 3211 * if illegal, so change those to "" */ 3212 if (return_format) { 3213 if (strEQ(PL_langinfo_buf, format)) { 3214 *PL_langinfo_buf = '\0'; 3215 } 3216 else { 3217 retval = save_to_buffer(format, &PL_langinfo_buf, 3218 &PL_langinfo_bufsize, 0); 3219 } 3220 } 3221 3222 break; 3223 3224 # endif 3225 3226 } 3227 } 3228 3229 return retval; 3230 3231 #endif 3232 3233 } 3234 3235 /* 3236 * Initialize locale awareness. 3237 */ 3238 int 3239 Perl_init_i18nl10n(pTHX_ int printwarn) 3240 { 3241 /* printwarn is 3242 * 3243 * 0 if not to output warning when setup locale is bad 3244 * 1 if to output warning based on value of PERL_BADLANG 3245 * >1 if to output regardless of PERL_BADLANG 3246 * 3247 * returns 3248 * 1 = set ok or not applicable, 3249 * 0 = fallback to a locale of lower priority 3250 * -1 = fallback to all locales failed, not even to the C locale 3251 * 3252 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is 3253 * set, debugging information is output. 3254 * 3255 * This looks more complicated than it is, mainly due to the #ifdefs. 3256 * 3257 * We try to set LC_ALL to the value determined by the environment. If 3258 * there is no LC_ALL on this platform, we try the individual categories we 3259 * know about. If this works, we are done. 3260 * 3261 * But if it doesn't work, we have to do something else. We search the 3262 * environment variables ourselves instead of relying on the system to do 3263 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we 3264 * think there is one), and the ultimate fallback "C". This is all done in 3265 * the same loop as above to avoid duplicating code, but it makes things 3266 * more complex. The 'trial_locales' array is initialized with just one 3267 * element; it causes the behavior described in the paragraph above this to 3268 * happen. If that fails, we add elements to 'trial_locales', and do extra 3269 * loop iterations to cause the behavior described in this paragraph. 3270 * 3271 * On Ultrix, the locale MUST come from the environment, so there is 3272 * preliminary code to set it. I (khw) am not sure that it is necessary, 3273 * and that this couldn't be folded into the loop, but barring any real 3274 * platforms to test on, it's staying as-is 3275 * 3276 * A slight complication is that in embedded Perls, the locale may already 3277 * be set-up, and we don't want to get it from the normal environment 3278 * variables. This is handled by having a special environment variable 3279 * indicate we're in this situation. We simply set setlocale's 2nd 3280 * parameter to be a NULL instead of "". That indicates to setlocale that 3281 * it is not to change anything, but to return the current value, 3282 * effectively initializing perl's db to what the locale already is. 3283 * 3284 * We play the same trick with NULL if a LC_ALL succeeds. We call 3285 * setlocale() on the individual categores with NULL to get their existing 3286 * values for our db, instead of trying to change them. 3287 * */ 3288 3289 dVAR; 3290 3291 int ok = 1; 3292 3293 #ifndef USE_LOCALE 3294 3295 PERL_UNUSED_ARG(printwarn); 3296 3297 #else /* USE_LOCALE */ 3298 # ifdef __GLIBC__ 3299 3300 const char * const language = PerlEnv_getenv("LANGUAGE"); 3301 3302 # endif 3303 3304 /* NULL uses the existing already set up locale */ 3305 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT")) 3306 ? NULL 3307 : ""; 3308 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */ 3309 unsigned int trial_locales_count; 3310 const char * const lc_all = PerlEnv_getenv("LC_ALL"); 3311 const char * const lang = PerlEnv_getenv("LANG"); 3312 bool setlocale_failure = FALSE; 3313 unsigned int i; 3314 3315 /* A later getenv() could zap this, so only use here */ 3316 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG"); 3317 3318 const bool locwarn = (printwarn > 1 3319 || ( printwarn 3320 && ( ! bad_lang_use_once 3321 || ( 3322 /* disallow with "" or "0" */ 3323 *bad_lang_use_once 3324 && strNE("0", bad_lang_use_once))))); 3325 3326 /* setlocale() return vals; not copied so must be looked at immediately */ 3327 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1]; 3328 3329 /* current locale for given category; should have been copied so aren't 3330 * volatile */ 3331 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1]; 3332 3333 # ifdef WIN32 3334 3335 /* In some systems you can find out the system default locale 3336 * and use that as the fallback locale. */ 3337 # define SYSTEM_DEFAULT_LOCALE 3338 # endif 3339 # ifdef SYSTEM_DEFAULT_LOCALE 3340 3341 const char *system_default_locale = NULL; 3342 3343 # endif 3344 3345 # ifndef DEBUGGING 3346 # define DEBUG_LOCALE_INIT(a,b,c) 3347 # else 3348 3349 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT"))); 3350 3351 # define DEBUG_LOCALE_INIT(category, locale, result) \ 3352 STMT_START { \ 3353 if (debug_initialization) { \ 3354 PerlIO_printf(Perl_debug_log, \ 3355 "%s:%d: %s\n", \ 3356 __FILE__, __LINE__, \ 3357 setlocale_debug_string(category, \ 3358 locale, \ 3359 result)); \ 3360 } \ 3361 } STMT_END 3362 3363 /* Make sure the parallel arrays are properly set up */ 3364 # ifdef USE_LOCALE_NUMERIC 3365 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC); 3366 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC")); 3367 # ifdef USE_POSIX_2008_LOCALE 3368 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK); 3369 # endif 3370 # endif 3371 # ifdef USE_LOCALE_CTYPE 3372 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE); 3373 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE")); 3374 # ifdef USE_POSIX_2008_LOCALE 3375 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK); 3376 # endif 3377 # endif 3378 # ifdef USE_LOCALE_COLLATE 3379 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE); 3380 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE")); 3381 # ifdef USE_POSIX_2008_LOCALE 3382 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK); 3383 # endif 3384 # endif 3385 # ifdef USE_LOCALE_TIME 3386 assert(categories[LC_TIME_INDEX] == LC_TIME); 3387 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME")); 3388 # ifdef USE_POSIX_2008_LOCALE 3389 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK); 3390 # endif 3391 # endif 3392 # ifdef USE_LOCALE_MESSAGES 3393 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES); 3394 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES")); 3395 # ifdef USE_POSIX_2008_LOCALE 3396 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK); 3397 # endif 3398 # endif 3399 # ifdef USE_LOCALE_MONETARY 3400 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY); 3401 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY")); 3402 # ifdef USE_POSIX_2008_LOCALE 3403 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK); 3404 # endif 3405 # endif 3406 # ifdef USE_LOCALE_ADDRESS 3407 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS); 3408 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS")); 3409 # ifdef USE_POSIX_2008_LOCALE 3410 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK); 3411 # endif 3412 # endif 3413 # ifdef USE_LOCALE_IDENTIFICATION 3414 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION); 3415 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION")); 3416 # ifdef USE_POSIX_2008_LOCALE 3417 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK); 3418 # endif 3419 # endif 3420 # ifdef USE_LOCALE_MEASUREMENT 3421 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT); 3422 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT")); 3423 # ifdef USE_POSIX_2008_LOCALE 3424 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK); 3425 # endif 3426 # endif 3427 # ifdef USE_LOCALE_PAPER 3428 assert(categories[LC_PAPER_INDEX] == LC_PAPER); 3429 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER")); 3430 # ifdef USE_POSIX_2008_LOCALE 3431 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK); 3432 # endif 3433 # endif 3434 # ifdef USE_LOCALE_TELEPHONE 3435 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE); 3436 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE")); 3437 # ifdef USE_POSIX_2008_LOCALE 3438 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK); 3439 # endif 3440 # endif 3441 # ifdef LC_ALL 3442 assert(categories[LC_ALL_INDEX] == LC_ALL); 3443 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL")); 3444 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX); 3445 # ifdef USE_POSIX_2008_LOCALE 3446 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK); 3447 # endif 3448 # endif 3449 # endif /* DEBUGGING */ 3450 3451 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for 3452 * why these particular incantations are used. */ 3453 #ifdef HAS_MBRLEN 3454 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps)); 3455 #endif 3456 #ifdef HAS_MBRTOWC 3457 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps)); 3458 #endif 3459 #ifdef HAS_WCTOMBR 3460 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps); 3461 #endif 3462 3463 /* Initialize the cache of the program's UTF-8ness for the always known 3464 * locales C and POSIX */ 3465 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness, 3466 sizeof(PL_locale_utf8ness)); 3467 3468 # ifdef USE_THREAD_SAFE_LOCALE 3469 # ifdef WIN32 3470 3471 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); 3472 3473 # endif 3474 # endif 3475 # ifdef USE_POSIX_2008_LOCALE 3476 3477 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0); 3478 if (! PL_C_locale_obj) { 3479 Perl_croak_nocontext( 3480 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno); 3481 } 3482 if (DEBUG_Lv_TEST || debug_initialization) { 3483 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj); 3484 } 3485 3486 # endif 3487 3488 # ifdef USE_LOCALE_NUMERIC 3489 3490 PL_numeric_radix_sv = newSVpvs("."); 3491 3492 # endif 3493 3494 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE) 3495 3496 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */ 3497 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL)); 3498 3499 # endif 3500 # ifdef LOCALE_ENVIRON_REQUIRED 3501 3502 /* 3503 * Ultrix setlocale(..., "") fails if there are no environment 3504 * variables from which to get a locale name. 3505 */ 3506 3507 # ifndef LC_ALL 3508 # error Ultrix without LC_ALL not implemented 3509 # else 3510 3511 { 3512 bool done = FALSE; 3513 if (lang) { 3514 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init); 3515 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]); 3516 if (sl_result[LC_ALL_INDEX]) 3517 done = TRUE; 3518 else 3519 setlocale_failure = TRUE; 3520 } 3521 if (! setlocale_failure) { 3522 const char * locale_param; 3523 for (i = 0; i < LC_ALL_INDEX; i++) { 3524 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i]))) 3525 ? setlocale_init 3526 : NULL; 3527 sl_result[i] = do_setlocale_r(categories[i], locale_param); 3528 if (! sl_result[i]) { 3529 setlocale_failure = TRUE; 3530 } 3531 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]); 3532 } 3533 } 3534 } 3535 3536 # endif /* LC_ALL */ 3537 # endif /* LOCALE_ENVIRON_REQUIRED */ 3538 3539 /* We try each locale in the list until we get one that works, or exhaust 3540 * the list. Normally the loop is executed just once. But if setting the 3541 * locale fails, inside the loop we add fallback trials to the array and so 3542 * will execute the loop multiple times */ 3543 trial_locales[0] = setlocale_init; 3544 trial_locales_count = 1; 3545 3546 for (i= 0; i < trial_locales_count; i++) { 3547 const char * trial_locale = trial_locales[i]; 3548 3549 if (i > 0) { 3550 3551 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED 3552 * when i==0, but I (khw) don't think that behavior makes much 3553 * sense */ 3554 setlocale_failure = FALSE; 3555 3556 # ifdef SYSTEM_DEFAULT_LOCALE 3557 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */ 3558 3559 /* On Windows machines, an entry of "" after the 0th means to use 3560 * the system default locale, which we now proceed to get. */ 3561 if (strEQ(trial_locale, "")) { 3562 unsigned int j; 3563 3564 /* Note that this may change the locale, but we are going to do 3565 * that anyway just below */ 3566 system_default_locale = do_setlocale_c(LC_ALL, ""); 3567 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale); 3568 3569 /* Skip if invalid or if it's already on the list of locales to 3570 * try */ 3571 if (! system_default_locale) { 3572 goto next_iteration; 3573 } 3574 for (j = 0; j < trial_locales_count; j++) { 3575 if (strEQ(system_default_locale, trial_locales[j])) { 3576 goto next_iteration; 3577 } 3578 } 3579 3580 trial_locale = system_default_locale; 3581 } 3582 # else 3583 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32 3584 # endif 3585 # endif /* SYSTEM_DEFAULT_LOCALE */ 3586 3587 } /* For i > 0 */ 3588 3589 # ifdef LC_ALL 3590 3591 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale); 3592 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]); 3593 if (! sl_result[LC_ALL_INDEX]) { 3594 setlocale_failure = TRUE; 3595 } 3596 else { 3597 /* Since LC_ALL succeeded, it should have changed all the other 3598 * categories it can to its value; so we massage things so that the 3599 * setlocales below just return their category's current values. 3600 * This adequately handles the case in NetBSD where LC_COLLATE may 3601 * not be defined for a locale, and setting it individually will 3602 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to 3603 * the POSIX locale. */ 3604 trial_locale = NULL; 3605 } 3606 3607 # endif /* LC_ALL */ 3608 3609 if (! setlocale_failure) { 3610 unsigned int j; 3611 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) { 3612 curlocales[j] 3613 = savepv(do_setlocale_r(categories[j], trial_locale)); 3614 if (! curlocales[j]) { 3615 setlocale_failure = TRUE; 3616 } 3617 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]); 3618 } 3619 3620 if (! setlocale_failure) { /* All succeeded */ 3621 break; /* Exit trial_locales loop */ 3622 } 3623 } 3624 3625 /* Here, something failed; will need to try a fallback. */ 3626 ok = 0; 3627 3628 if (i == 0) { 3629 unsigned int j; 3630 3631 if (locwarn) { /* Output failure info only on the first one */ 3632 3633 # ifdef LC_ALL 3634 3635 PerlIO_printf(Perl_error_log, 3636 "perl: warning: Setting locale failed.\n"); 3637 3638 # else /* !LC_ALL */ 3639 3640 PerlIO_printf(Perl_error_log, 3641 "perl: warning: Setting locale failed for the categories:\n\t"); 3642 3643 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) { 3644 if (! curlocales[j]) { 3645 PerlIO_printf(Perl_error_log, category_names[j]); 3646 } 3647 else { 3648 Safefree(curlocales[j]); 3649 } 3650 } 3651 3652 # endif /* LC_ALL */ 3653 3654 PerlIO_printf(Perl_error_log, 3655 "perl: warning: Please check that your locale settings:\n"); 3656 3657 # ifdef __GLIBC__ 3658 3659 PerlIO_printf(Perl_error_log, 3660 "\tLANGUAGE = %c%s%c,\n", 3661 language ? '"' : '(', 3662 language ? language : "unset", 3663 language ? '"' : ')'); 3664 # endif 3665 3666 PerlIO_printf(Perl_error_log, 3667 "\tLC_ALL = %c%s%c,\n", 3668 lc_all ? '"' : '(', 3669 lc_all ? lc_all : "unset", 3670 lc_all ? '"' : ')'); 3671 3672 # if defined(USE_ENVIRON_ARRAY) 3673 3674 { 3675 char **e; 3676 3677 /* Look through the environment for any variables of the 3678 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was 3679 * already handled above. These are assumed to be locale 3680 * settings. Output them and their values. */ 3681 for (e = environ; *e; e++) { 3682 const STRLEN prefix_len = sizeof("LC_") - 1; 3683 STRLEN uppers_len; 3684 3685 if ( strBEGINs(*e, "LC_") 3686 && ! strBEGINs(*e, "LC_ALL=") 3687 && (uppers_len = strspn(*e + prefix_len, 3688 "ABCDEFGHIJKLMNOPQRSTUVWXYZ")) 3689 && ((*e)[prefix_len + uppers_len] == '=')) 3690 { 3691 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n", 3692 (int) (prefix_len + uppers_len), *e, 3693 *e + prefix_len + uppers_len + 1); 3694 } 3695 } 3696 } 3697 3698 # else 3699 3700 PerlIO_printf(Perl_error_log, 3701 "\t(possibly more locale environment variables)\n"); 3702 3703 # endif 3704 3705 PerlIO_printf(Perl_error_log, 3706 "\tLANG = %c%s%c\n", 3707 lang ? '"' : '(', 3708 lang ? lang : "unset", 3709 lang ? '"' : ')'); 3710 3711 PerlIO_printf(Perl_error_log, 3712 " are supported and installed on your system.\n"); 3713 } 3714 3715 /* Calculate what fallback locales to try. We have avoided this 3716 * until we have to, because failure is quite unlikely. This will 3717 * usually change the upper bound of the loop we are in. 3718 * 3719 * Since the system's default way of setting the locale has not 3720 * found one that works, We use Perl's defined ordering: LC_ALL, 3721 * LANG, and the C locale. We don't try the same locale twice, so 3722 * don't add to the list if already there. (On POSIX systems, the 3723 * LC_ALL element will likely be a repeat of the 0th element "", 3724 * but there's no harm done by doing it explicitly. 3725 * 3726 * Note that this tries the LC_ALL environment variable even on 3727 * systems which have no LC_ALL locale setting. This may or may 3728 * not have been originally intentional, but there's no real need 3729 * to change the behavior. */ 3730 if (lc_all) { 3731 for (j = 0; j < trial_locales_count; j++) { 3732 if (strEQ(lc_all, trial_locales[j])) { 3733 goto done_lc_all; 3734 } 3735 } 3736 trial_locales[trial_locales_count++] = lc_all; 3737 } 3738 done_lc_all: 3739 3740 if (lang) { 3741 for (j = 0; j < trial_locales_count; j++) { 3742 if (strEQ(lang, trial_locales[j])) { 3743 goto done_lang; 3744 } 3745 } 3746 trial_locales[trial_locales_count++] = lang; 3747 } 3748 done_lang: 3749 3750 # if defined(WIN32) && defined(LC_ALL) 3751 3752 /* For Windows, we also try the system default locale before "C". 3753 * (If there exists a Windows without LC_ALL we skip this because 3754 * it gets too complicated. For those, the "C" is the next 3755 * fallback possibility). The "" is the same as the 0th element of 3756 * the array, but the code at the loop above knows to treat it 3757 * differently when not the 0th */ 3758 trial_locales[trial_locales_count++] = ""; 3759 3760 # endif 3761 3762 for (j = 0; j < trial_locales_count; j++) { 3763 if (strEQ("C", trial_locales[j])) { 3764 goto done_C; 3765 } 3766 } 3767 trial_locales[trial_locales_count++] = "C"; 3768 3769 done_C: ; 3770 } /* end of first time through the loop */ 3771 3772 # ifdef WIN32 3773 3774 next_iteration: ; 3775 3776 # endif 3777 3778 } /* end of looping through the trial locales */ 3779 3780 if (ok < 1) { /* If we tried to fallback */ 3781 const char* msg; 3782 if (! setlocale_failure) { /* fallback succeeded */ 3783 msg = "Falling back to"; 3784 } 3785 else { /* fallback failed */ 3786 unsigned int j; 3787 3788 /* We dropped off the end of the loop, so have to decrement i to 3789 * get back to the value the last time through */ 3790 i--; 3791 3792 ok = -1; 3793 msg = "Failed to fall back to"; 3794 3795 /* To continue, we should use whatever values we've got */ 3796 3797 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) { 3798 Safefree(curlocales[j]); 3799 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL)); 3800 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]); 3801 } 3802 } 3803 3804 if (locwarn) { 3805 const char * description; 3806 const char * name = ""; 3807 if (strEQ(trial_locales[i], "C")) { 3808 description = "the standard locale"; 3809 name = "C"; 3810 } 3811 3812 # ifdef SYSTEM_DEFAULT_LOCALE 3813 3814 else if (strEQ(trial_locales[i], "")) { 3815 description = "the system default locale"; 3816 if (system_default_locale) { 3817 name = system_default_locale; 3818 } 3819 } 3820 3821 # endif /* SYSTEM_DEFAULT_LOCALE */ 3822 3823 else { 3824 description = "a fallback locale"; 3825 name = trial_locales[i]; 3826 } 3827 if (name && strNE(name, "")) { 3828 PerlIO_printf(Perl_error_log, 3829 "perl: warning: %s %s (\"%s\").\n", msg, description, name); 3830 } 3831 else { 3832 PerlIO_printf(Perl_error_log, 3833 "perl: warning: %s %s.\n", msg, description); 3834 } 3835 } 3836 } /* End of tried to fallback */ 3837 3838 /* Done with finding the locales; update our records */ 3839 3840 # ifdef USE_LOCALE_CTYPE 3841 3842 new_ctype(curlocales[LC_CTYPE_INDEX]); 3843 3844 # endif 3845 # ifdef USE_LOCALE_COLLATE 3846 3847 new_collate(curlocales[LC_COLLATE_INDEX]); 3848 3849 # endif 3850 # ifdef USE_LOCALE_NUMERIC 3851 3852 new_numeric(curlocales[LC_NUMERIC_INDEX]); 3853 3854 # endif 3855 3856 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) { 3857 3858 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE) 3859 3860 /* This caches whether each category's locale is UTF-8 or not. This 3861 * may involve changing the locale. It is ok to do this at 3862 * initialization time before any threads have started, but not later 3863 * unless thread-safe operations are used. 3864 * Caching means that if the program heeds our dictate not to change 3865 * locales in threaded applications, this data will remain valid, and 3866 * it may get queried without having to change locales. If the 3867 * environment is such that all categories have the same locale, this 3868 * isn't needed, as the code will not change the locale; but this 3869 * handles the uncommon case where the environment has disparate 3870 * locales for the categories */ 3871 (void) _is_cur_LC_category_utf8(categories[i]); 3872 3873 # endif 3874 3875 Safefree(curlocales[i]); 3876 } 3877 3878 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE) 3879 3880 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE 3881 * locale is UTF-8. The call to new_ctype() just above has already 3882 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If 3883 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE}) 3884 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on 3885 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */ 3886 PL_utf8locale = PL_in_utf8_CTYPE_locale; 3887 3888 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO. 3889 This is an alternative to using the -C command line switch 3890 (the -C if present will override this). */ 3891 { 3892 const char *p = PerlEnv_getenv("PERL_UNICODE"); 3893 PL_unicode = p ? parse_unicode_opts(&p) : 0; 3894 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG) 3895 PL_utf8cache = -1; 3896 } 3897 3898 # endif 3899 #endif /* USE_LOCALE */ 3900 #ifdef DEBUGGING 3901 3902 /* So won't continue to output stuff */ 3903 DEBUG_INITIALIZATION_set(FALSE); 3904 3905 #endif 3906 3907 return ok; 3908 } 3909 3910 #ifdef USE_LOCALE_COLLATE 3911 3912 char * 3913 Perl__mem_collxfrm(pTHX_ const char *input_string, 3914 STRLEN len, /* Length of 'input_string' */ 3915 STRLEN *xlen, /* Set to length of returned string 3916 (not including the collation index 3917 prefix) */ 3918 bool utf8 /* Is the input in UTF-8? */ 3919 ) 3920 { 3921 3922 /* _mem_collxfrm() is a bit like strxfrm() but with two important 3923 * differences. First, it handles embedded NULs. Second, it allocates a bit 3924 * more memory than needed for the transformed data itself. The real 3925 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to 3926 * the length of that, and doesn't include the collation index size. 3927 * Please see sv_collxfrm() to see how this is used. */ 3928 3929 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix) 3930 3931 char * s = (char *) input_string; 3932 STRLEN s_strlen = strlen(input_string); 3933 char *xbuf = NULL; 3934 STRLEN xAlloc; /* xalloc is a reserved word in VC */ 3935 STRLEN length_in_chars; 3936 bool first_time = TRUE; /* Cleared after first loop iteration */ 3937 3938 PERL_ARGS_ASSERT__MEM_COLLXFRM; 3939 3940 /* Must be NUL-terminated */ 3941 assert(*(input_string + len) == '\0'); 3942 3943 /* If this locale has defective collation, skip */ 3944 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) { 3945 DEBUG_L(PerlIO_printf(Perl_debug_log, 3946 "_mem_collxfrm: locale's collation is defective\n")); 3947 goto bad; 3948 } 3949 3950 /* Replace any embedded NULs with the control that sorts before any others. 3951 * This will give as good as possible results on strings that don't 3952 * otherwise contain that character, but otherwise there may be 3953 * less-than-perfect results with that character and NUL. This is 3954 * unavoidable unless we replace strxfrm with our own implementation. */ 3955 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded 3956 NUL */ 3957 char * e = s + len; 3958 char * sans_nuls; 3959 STRLEN sans_nuls_len; 3960 int try_non_controls; 3961 char this_replacement_char[] = "?\0"; /* Room for a two-byte string, 3962 making sure 2nd byte is NUL. 3963 */ 3964 STRLEN this_replacement_len; 3965 3966 /* If we don't know what non-NUL control character sorts lowest for 3967 * this locale, find it */ 3968 if (PL_strxfrm_NUL_replacement == '\0') { 3969 int j; 3970 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also 3971 includes the collation index 3972 prefixed. */ 3973 3974 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n")); 3975 3976 /* Unlikely, but it may be that no control will work to replace 3977 * NUL, in which case we instead look for any character. Controls 3978 * are preferred because collation order is, in general, context 3979 * sensitive, with adjoining characters affecting the order, and 3980 * controls are less likely to have such interactions, allowing the 3981 * NUL-replacement to stand on its own. (Another way to look at it 3982 * is to imagine what would happen if the NUL were replaced by a 3983 * combining character; it wouldn't work out all that well.) */ 3984 for (try_non_controls = 0; 3985 try_non_controls < 2; 3986 try_non_controls++) 3987 { 3988 /* Look through all legal code points (NUL isn't) */ 3989 for (j = 1; j < 256; j++) { 3990 char * x; /* j's xfrm plus collation index */ 3991 STRLEN x_len; /* length of 'x' */ 3992 STRLEN trial_len = 1; 3993 char cur_source[] = { '\0', '\0' }; 3994 3995 /* Skip non-controls the first time through the loop. The 3996 * controls in a UTF-8 locale are the L1 ones */ 3997 if (! try_non_controls && (PL_in_utf8_COLLATE_locale) 3998 ? ! isCNTRL_L1(j) 3999 : ! isCNTRL_LC(j)) 4000 { 4001 continue; 4002 } 4003 4004 /* Create a 1-char string of the current code point */ 4005 cur_source[0] = (char) j; 4006 4007 /* Then transform it */ 4008 x = _mem_collxfrm(cur_source, trial_len, &x_len, 4009 0 /* The string is not in UTF-8 */); 4010 4011 /* Ignore any character that didn't successfully transform. 4012 * */ 4013 if (! x) { 4014 continue; 4015 } 4016 4017 /* If this character's transformation is lower than 4018 * the current lowest, this one becomes the lowest */ 4019 if ( cur_min_x == NULL 4020 || strLT(x + COLLXFRM_HDR_LEN, 4021 cur_min_x + COLLXFRM_HDR_LEN)) 4022 { 4023 PL_strxfrm_NUL_replacement = j; 4024 Safefree(cur_min_x); 4025 cur_min_x = x; 4026 } 4027 else { 4028 Safefree(x); 4029 } 4030 } /* end of loop through all 255 characters */ 4031 4032 /* Stop looking if found */ 4033 if (cur_min_x) { 4034 break; 4035 } 4036 4037 /* Unlikely, but possible, if there aren't any controls that 4038 * work in the locale, repeat the loop, looking for any 4039 * character that works */ 4040 DEBUG_L(PerlIO_printf(Perl_debug_log, 4041 "_mem_collxfrm: No control worked. Trying non-controls\n")); 4042 } /* End of loop to try first the controls, then any char */ 4043 4044 if (! cur_min_x) { 4045 DEBUG_L(PerlIO_printf(Perl_debug_log, 4046 "_mem_collxfrm: Couldn't find any character to replace" 4047 " embedded NULs in locale %s with", PL_collation_name)); 4048 goto bad; 4049 } 4050 4051 DEBUG_L(PerlIO_printf(Perl_debug_log, 4052 "_mem_collxfrm: Replacing embedded NULs in locale %s with " 4053 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement)); 4054 4055 Safefree(cur_min_x); 4056 } /* End of determining the character that is to replace NULs */ 4057 4058 /* If the replacement is variant under UTF-8, it must match the 4059 * UTF8-ness of the original */ 4060 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) { 4061 this_replacement_char[0] = 4062 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement); 4063 this_replacement_char[1] = 4064 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement); 4065 this_replacement_len = 2; 4066 } 4067 else { 4068 this_replacement_char[0] = PL_strxfrm_NUL_replacement; 4069 /* this_replacement_char[1] = '\0' was done at initialization */ 4070 this_replacement_len = 1; 4071 } 4072 4073 /* The worst case length for the replaced string would be if every 4074 * character in it is NUL. Multiply that by the length of each 4075 * replacement, and allow for a trailing NUL */ 4076 sans_nuls_len = (len * this_replacement_len) + 1; 4077 Newx(sans_nuls, sans_nuls_len, char); 4078 *sans_nuls = '\0'; 4079 4080 /* Replace each NUL with the lowest collating control. Loop until have 4081 * exhausted all the NULs */ 4082 while (s + s_strlen < e) { 4083 my_strlcat(sans_nuls, s, sans_nuls_len); 4084 4085 /* Do the actual replacement */ 4086 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len); 4087 4088 /* Move past the input NUL */ 4089 s += s_strlen + 1; 4090 s_strlen = strlen(s); 4091 } 4092 4093 /* And add anything that trails the final NUL */ 4094 my_strlcat(sans_nuls, s, sans_nuls_len); 4095 4096 /* Switch so below we transform this modified string */ 4097 s = sans_nuls; 4098 len = strlen(s); 4099 } /* End of replacing NULs */ 4100 4101 /* Make sure the UTF8ness of the string and locale match */ 4102 if (utf8 != PL_in_utf8_COLLATE_locale) { 4103 /* XXX convert above Unicode to 10FFFF? */ 4104 const char * const t = s; /* Temporary so we can later find where the 4105 input was */ 4106 4107 /* Here they don't match. Change the string's to be what the locale is 4108 * expecting */ 4109 4110 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */ 4111 s = (char *) bytes_to_utf8((const U8 *) s, &len); 4112 utf8 = TRUE; 4113 } 4114 else { /* locale is not UTF-8; but input is; downgrade the input */ 4115 4116 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8); 4117 4118 /* If the downgrade was successful we are done, but if the input 4119 * contains things that require UTF-8 to represent, have to do 4120 * damage control ... */ 4121 if (UNLIKELY(utf8)) { 4122 4123 /* What we do is construct a non-UTF-8 string with 4124 * 1) the characters representable by a single byte converted 4125 * to be so (if necessary); 4126 * 2) and the rest converted to collate the same as the 4127 * highest collating representable character. That makes 4128 * them collate at the end. This is similar to how we 4129 * handle embedded NULs, but we use the highest collating 4130 * code point instead of the smallest. Like the NUL case, 4131 * this isn't perfect, but is the best we can reasonably 4132 * do. Every above-255 code point will sort the same as 4133 * the highest-sorting 0-255 code point. If that code 4134 * point can combine in a sequence with some other code 4135 * points for weight calculations, us changing something to 4136 * be it can adversely affect the results. But in most 4137 * cases, it should work reasonably. And note that this is 4138 * really an illegal situation: using code points above 255 4139 * on a locale where only 0-255 are valid. If two strings 4140 * sort entirely equal, then the sort order for the 4141 * above-255 code points will be in code point order. */ 4142 4143 utf8 = FALSE; 4144 4145 /* If we haven't calculated the code point with the maximum 4146 * collating order for this locale, do so now */ 4147 if (! PL_strxfrm_max_cp) { 4148 int j; 4149 4150 /* The current transformed string that collates the 4151 * highest (except it also includes the prefixed collation 4152 * index. */ 4153 char * cur_max_x = NULL; 4154 4155 /* Look through all legal code points (NUL isn't) */ 4156 for (j = 1; j < 256; j++) { 4157 char * x; 4158 STRLEN x_len; 4159 char cur_source[] = { '\0', '\0' }; 4160 4161 /* Create a 1-char string of the current code point */ 4162 cur_source[0] = (char) j; 4163 4164 /* Then transform it */ 4165 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE); 4166 4167 /* If something went wrong (which it shouldn't), just 4168 * ignore this code point */ 4169 if (! x) { 4170 continue; 4171 } 4172 4173 /* If this character's transformation is higher than 4174 * the current highest, this one becomes the highest */ 4175 if ( cur_max_x == NULL 4176 || strGT(x + COLLXFRM_HDR_LEN, 4177 cur_max_x + COLLXFRM_HDR_LEN)) 4178 { 4179 PL_strxfrm_max_cp = j; 4180 Safefree(cur_max_x); 4181 cur_max_x = x; 4182 } 4183 else { 4184 Safefree(x); 4185 } 4186 } 4187 4188 if (! cur_max_x) { 4189 DEBUG_L(PerlIO_printf(Perl_debug_log, 4190 "_mem_collxfrm: Couldn't find any character to" 4191 " replace above-Latin1 chars in locale %s with", 4192 PL_collation_name)); 4193 goto bad; 4194 } 4195 4196 DEBUG_L(PerlIO_printf(Perl_debug_log, 4197 "_mem_collxfrm: highest 1-byte collating character" 4198 " in locale %s is 0x%02X\n", 4199 PL_collation_name, 4200 PL_strxfrm_max_cp)); 4201 4202 Safefree(cur_max_x); 4203 } 4204 4205 /* Here we know which legal code point collates the highest. 4206 * We are ready to construct the non-UTF-8 string. The length 4207 * will be at least 1 byte smaller than the input string 4208 * (because we changed at least one 2-byte character into a 4209 * single byte), but that is eaten up by the trailing NUL */ 4210 Newx(s, len, char); 4211 4212 { 4213 STRLEN i; 4214 STRLEN d= 0; 4215 char * e = (char *) t + len; 4216 4217 for (i = 0; i < len; i+= UTF8SKIP(t + i)) { 4218 U8 cur_char = t[i]; 4219 if (UTF8_IS_INVARIANT(cur_char)) { 4220 s[d++] = cur_char; 4221 } 4222 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) { 4223 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]); 4224 } 4225 else { /* Replace illegal cp with highest collating 4226 one */ 4227 s[d++] = PL_strxfrm_max_cp; 4228 } 4229 } 4230 s[d++] = '\0'; 4231 Renew(s, d, char); /* Free up unused space */ 4232 } 4233 } 4234 } 4235 4236 /* Here, we have constructed a modified version of the input. It could 4237 * be that we already had a modified copy before we did this version. 4238 * If so, that copy is no longer needed */ 4239 if (t != input_string) { 4240 Safefree(t); 4241 } 4242 } 4243 4244 length_in_chars = (utf8) 4245 ? utf8_length((U8 *) s, (U8 *) s + len) 4246 : len; 4247 4248 /* The first element in the output is the collation id, used by 4249 * sv_collxfrm(); then comes the space for the transformed string. The 4250 * equation should give us a good estimate as to how much is needed */ 4251 xAlloc = COLLXFRM_HDR_LEN 4252 + PL_collxfrm_base 4253 + (PL_collxfrm_mult * length_in_chars); 4254 Newx(xbuf, xAlloc, char); 4255 if (UNLIKELY(! xbuf)) { 4256 DEBUG_L(PerlIO_printf(Perl_debug_log, 4257 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc)); 4258 goto bad; 4259 } 4260 4261 /* Store the collation id */ 4262 *(U32*)xbuf = PL_collation_ix; 4263 4264 /* Then the transformation of the input. We loop until successful, or we 4265 * give up */ 4266 for (;;) { 4267 4268 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN); 4269 4270 /* If the transformed string occupies less space than we told strxfrm() 4271 * was available, it means it successfully transformed the whole 4272 * string. */ 4273 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) { 4274 4275 /* Some systems include a trailing NUL in the returned length. 4276 * Ignore it, using a loop in case multiple trailing NULs are 4277 * returned. */ 4278 while ( (*xlen) > 0 4279 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0') 4280 { 4281 (*xlen)--; 4282 } 4283 4284 /* If the first try didn't get it, it means our prediction was low. 4285 * Modify the coefficients so that we predict a larger value in any 4286 * future transformations */ 4287 if (! first_time) { 4288 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */ 4289 STRLEN computed_guess = PL_collxfrm_base 4290 + (PL_collxfrm_mult * length_in_chars); 4291 4292 /* On zero-length input, just keep current slope instead of 4293 * dividing by 0 */ 4294 const STRLEN new_m = (length_in_chars != 0) 4295 ? needed / length_in_chars 4296 : PL_collxfrm_mult; 4297 4298 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 4299 "%s: %d: initial size of %zu bytes for a length " 4300 "%zu string was insufficient, %zu needed\n", 4301 __FILE__, __LINE__, 4302 computed_guess, length_in_chars, needed)); 4303 4304 /* If slope increased, use it, but discard this result for 4305 * length 1 strings, as we can't be sure that it's a real slope 4306 * change */ 4307 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) { 4308 4309 # ifdef DEBUGGING 4310 4311 STRLEN old_m = PL_collxfrm_mult; 4312 STRLEN old_b = PL_collxfrm_base; 4313 4314 # endif 4315 4316 PL_collxfrm_mult = new_m; 4317 PL_collxfrm_base = 1; /* +1 For trailing NUL */ 4318 computed_guess = PL_collxfrm_base 4319 + (PL_collxfrm_mult * length_in_chars); 4320 if (computed_guess < needed) { 4321 PL_collxfrm_base += needed - computed_guess; 4322 } 4323 4324 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 4325 "%s: %d: slope is now %zu; was %zu, base " 4326 "is now %zu; was %zu\n", 4327 __FILE__, __LINE__, 4328 PL_collxfrm_mult, old_m, 4329 PL_collxfrm_base, old_b)); 4330 } 4331 else { /* Slope didn't change, but 'b' did */ 4332 const STRLEN new_b = needed 4333 - computed_guess 4334 + PL_collxfrm_base; 4335 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 4336 "%s: %d: base is now %zu; was %zu\n", 4337 __FILE__, __LINE__, 4338 new_b, PL_collxfrm_base)); 4339 PL_collxfrm_base = new_b; 4340 } 4341 } 4342 4343 break; 4344 } 4345 4346 if (UNLIKELY(*xlen >= PERL_INT_MAX)) { 4347 DEBUG_L(PerlIO_printf(Perl_debug_log, 4348 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n", 4349 *xlen, PERL_INT_MAX)); 4350 goto bad; 4351 } 4352 4353 /* A well-behaved strxfrm() returns exactly how much space it needs 4354 * (usually not including the trailing NUL) when it fails due to not 4355 * enough space being provided. Assume that this is the case unless 4356 * it's been proven otherwise */ 4357 if (LIKELY(PL_strxfrm_is_behaved) && first_time) { 4358 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1; 4359 } 4360 else { /* Here, either: 4361 * 1) The strxfrm() has previously shown bad behavior; or 4362 * 2) It isn't the first time through the loop, which means 4363 * that the strxfrm() is now showing bad behavior, because 4364 * we gave it what it said was needed in the previous 4365 * iteration, and it came back saying it needed still more. 4366 * (Many versions of cygwin fit this. When the buffer size 4367 * isn't sufficient, they return the input size instead of 4368 * how much is needed.) 4369 * Increase the buffer size by a fixed percentage and try again. 4370 * */ 4371 xAlloc += (xAlloc / 4) + 1; 4372 PL_strxfrm_is_behaved = FALSE; 4373 4374 # ifdef DEBUGGING 4375 4376 if (DEBUG_Lv_TEST || debug_initialization) { 4377 PerlIO_printf(Perl_debug_log, 4378 "_mem_collxfrm required more space than previously calculated" 4379 " for locale %s, trying again with new guess=%d+%zu\n", 4380 PL_collation_name, (int) COLLXFRM_HDR_LEN, 4381 xAlloc - COLLXFRM_HDR_LEN); 4382 } 4383 4384 # endif 4385 4386 } 4387 4388 Renew(xbuf, xAlloc, char); 4389 if (UNLIKELY(! xbuf)) { 4390 DEBUG_L(PerlIO_printf(Perl_debug_log, 4391 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc)); 4392 goto bad; 4393 } 4394 4395 first_time = FALSE; 4396 } 4397 4398 4399 # ifdef DEBUGGING 4400 4401 if (DEBUG_Lv_TEST || debug_initialization) { 4402 4403 print_collxfrm_input_and_return(s, s + len, xlen, utf8); 4404 PerlIO_printf(Perl_debug_log, "Its xfrm is:"); 4405 PerlIO_printf(Perl_debug_log, "%s\n", 4406 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN, 4407 *xlen, 1)); 4408 } 4409 4410 # endif 4411 4412 /* Free up unneeded space; retain ehough for trailing NUL */ 4413 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char); 4414 4415 if (s != input_string) { 4416 Safefree(s); 4417 } 4418 4419 return xbuf; 4420 4421 bad: 4422 4423 # ifdef DEBUGGING 4424 4425 if (DEBUG_Lv_TEST || debug_initialization) { 4426 print_collxfrm_input_and_return(s, s + len, NULL, utf8); 4427 } 4428 4429 # endif 4430 4431 Safefree(xbuf); 4432 if (s != input_string) { 4433 Safefree(s); 4434 } 4435 *xlen = 0; 4436 4437 return NULL; 4438 } 4439 4440 # ifdef DEBUGGING 4441 4442 STATIC void 4443 S_print_collxfrm_input_and_return(pTHX_ 4444 const char * const s, 4445 const char * const e, 4446 const STRLEN * const xlen, 4447 const bool is_utf8) 4448 { 4449 4450 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN; 4451 4452 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ", 4453 (UV)PL_collation_ix); 4454 if (xlen) { 4455 PerlIO_printf(Perl_debug_log, "%zu", *xlen); 4456 } 4457 else { 4458 PerlIO_printf(Perl_debug_log, "NULL"); 4459 } 4460 PerlIO_printf(Perl_debug_log, " for locale '%s', string='", 4461 PL_collation_name); 4462 print_bytes_for_locale(s, e, is_utf8); 4463 4464 PerlIO_printf(Perl_debug_log, "'\n"); 4465 } 4466 4467 # endif /* DEBUGGING */ 4468 #endif /* USE_LOCALE_COLLATE */ 4469 #ifdef USE_LOCALE 4470 # ifdef DEBUGGING 4471 4472 STATIC void 4473 S_print_bytes_for_locale(pTHX_ 4474 const char * const s, 4475 const char * const e, 4476 const bool is_utf8) 4477 { 4478 const char * t = s; 4479 bool prev_was_printable = TRUE; 4480 bool first_time = TRUE; 4481 4482 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE; 4483 4484 while (t < e) { 4485 UV cp = (is_utf8) 4486 ? utf8_to_uvchr_buf((U8 *) t, e, NULL) 4487 : * (U8 *) t; 4488 if (isPRINT(cp)) { 4489 if (! prev_was_printable) { 4490 PerlIO_printf(Perl_debug_log, " "); 4491 } 4492 PerlIO_printf(Perl_debug_log, "%c", (U8) cp); 4493 prev_was_printable = TRUE; 4494 } 4495 else { 4496 if (! first_time) { 4497 PerlIO_printf(Perl_debug_log, " "); 4498 } 4499 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp); 4500 prev_was_printable = FALSE; 4501 } 4502 t += (is_utf8) ? UTF8SKIP(t) : 1; 4503 first_time = FALSE; 4504 } 4505 } 4506 4507 # endif /* #ifdef DEBUGGING */ 4508 4509 STATIC const char * 4510 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale) 4511 { 4512 /* Changes the locale for LC_'switch_category" to that of 4513 * LC_'template_category', if they aren't already the same. If not NULL, 4514 * 'template_locale' is the locale that 'template_category' is in. 4515 * 4516 * Returns a copy of the name of the original locale for 'switch_category' 4517 * so can be switched back to with the companion function 4518 * restore_switched_locale(), (NULL if no restoral is necessary.) */ 4519 4520 char * restore_to_locale = NULL; 4521 4522 if (switch_category == template_category) { /* No changes needed */ 4523 return NULL; 4524 } 4525 4526 /* Find the original locale of the category we may need to change, so that 4527 * it can be restored to later */ 4528 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category, 4529 NULL))); 4530 if (! restore_to_locale) { 4531 Perl_croak(aTHX_ 4532 "panic: %s: %d: Could not find current %s locale, errno=%d\n", 4533 __FILE__, __LINE__, category_name(switch_category), errno); 4534 } 4535 4536 /* If the locale of the template category wasn't passed in, find it now */ 4537 if (template_locale == NULL) { 4538 template_locale = do_setlocale_r(template_category, NULL); 4539 if (! template_locale) { 4540 Perl_croak(aTHX_ 4541 "panic: %s: %d: Could not find current %s locale, errno=%d\n", 4542 __FILE__, __LINE__, category_name(template_category), errno); 4543 } 4544 } 4545 4546 /* It the locales are the same, there's nothing to do */ 4547 if (strEQ(restore_to_locale, template_locale)) { 4548 Safefree(restore_to_locale); 4549 4550 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n", 4551 category_name(switch_category), template_locale)); 4552 4553 return NULL; 4554 } 4555 4556 /* Finally, change the locale to the template one */ 4557 if (! do_setlocale_r(switch_category, template_locale)) { 4558 Perl_croak(aTHX_ 4559 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n", 4560 __FILE__, __LINE__, category_name(switch_category), 4561 template_locale, errno); 4562 } 4563 4564 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n", 4565 category_name(switch_category), template_locale)); 4566 4567 return restore_to_locale; 4568 } 4569 4570 STATIC void 4571 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale) 4572 { 4573 /* Restores the locale for LC_'category' to 'original_locale' (which is a 4574 * copy that will be freed by this function), or do nothing if the latter 4575 * parameter is NULL */ 4576 4577 if (original_locale == NULL) { 4578 return; 4579 } 4580 4581 if (! do_setlocale_r(category, original_locale)) { 4582 Perl_croak(aTHX_ 4583 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n", 4584 __FILE__, __LINE__, 4585 category_name(category), original_locale, errno); 4586 } 4587 4588 Safefree(original_locale); 4589 } 4590 4591 /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */ 4592 #define CUR_LC_BUFFER_SIZE 64 4593 4594 bool 4595 Perl__is_cur_LC_category_utf8(pTHX_ int category) 4596 { 4597 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE 4598 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have 4599 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence 4600 * could give the wrong result. The result will very likely be correct for 4601 * languages that have commonly used non-ASCII characters, but for notably 4602 * English, it comes down to if the locale's name ends in something like 4603 * "UTF-8". It errs on the side of not being a UTF-8 locale. 4604 * 4605 * If the platform is early C89, not containing mbtowc(), or we are 4606 * compiled to not pay attention to LC_CTYPE, this employs heuristics. 4607 * These work very well for non-Latin locales or those whose currency 4608 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at 4609 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend 4610 * on the name containing UTF-8 or not. */ 4611 4612 /* Name of current locale corresponding to the input category */ 4613 const char *save_input_locale = NULL; 4614 4615 bool is_utf8 = FALSE; /* The return value */ 4616 4617 /* The variables below are for the cache of previous lookups using this 4618 * function. The cache is a C string, described at the definition for 4619 * 'C_and_POSIX_utf8ness'. 4620 * 4621 * The first part of the cache is fixed, for the C and POSIX locales. The 4622 * varying part starts just after them. */ 4623 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness); 4624 4625 Size_t utf8ness_cache_size; /* Size of the varying portion */ 4626 Size_t input_name_len; /* Length in bytes of save_input_locale */ 4627 Size_t input_name_len_with_overhead; /* plus extra chars used to store 4628 the name in the cache */ 4629 char * delimited; /* The name plus the delimiters used to store 4630 it in the cache */ 4631 char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */ 4632 char * name_pos; /* position of 'delimited' in the cache, or 0 4633 if not there */ 4634 4635 4636 # ifdef LC_ALL 4637 4638 assert(category != LC_ALL); 4639 4640 # endif 4641 4642 /* Get the desired category's locale */ 4643 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL))); 4644 if (! save_input_locale) { 4645 Perl_croak(aTHX_ 4646 "panic: %s: %d: Could not find current %s locale, errno=%d\n", 4647 __FILE__, __LINE__, category_name(category), errno); 4648 } 4649 4650 DEBUG_L(PerlIO_printf(Perl_debug_log, 4651 "Current locale for %s is %s\n", 4652 category_name(category), save_input_locale)); 4653 4654 input_name_len = strlen(save_input_locale); 4655 4656 /* In our cache, each name is accompanied by two delimiters and a single 4657 * utf8ness digit */ 4658 input_name_len_with_overhead = input_name_len + 3; 4659 4660 if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) { 4661 /* we can use the buffer, avoid a malloc */ 4662 delimited = buffer; 4663 } else { /* need a malloc */ 4664 /* Allocate and populate space for a copy of the name surrounded by the 4665 * delimiters */ 4666 Newx(delimited, input_name_len_with_overhead, char); 4667 } 4668 4669 delimited[0] = UTF8NESS_SEP[0]; 4670 Copy(save_input_locale, delimited + 1, input_name_len, char); 4671 delimited[input_name_len+1] = UTF8NESS_PREFIX[0]; 4672 delimited[input_name_len+2] = '\0'; 4673 4674 /* And see if that is in the cache */ 4675 name_pos = instr(PL_locale_utf8ness, delimited); 4676 if (name_pos) { 4677 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0'; 4678 4679 # ifdef DEBUGGING 4680 4681 if (DEBUG_Lv_TEST || debug_initialization) { 4682 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n", 4683 save_input_locale, is_utf8); 4684 } 4685 4686 # endif 4687 4688 /* And, if not already in that position, move it to the beginning of 4689 * the non-constant portion of the list, since it is the most recently 4690 * used. (We don't have to worry about overflow, since just moving 4691 * existing names around) */ 4692 if (name_pos > utf8ness_cache) { 4693 Move(utf8ness_cache, 4694 utf8ness_cache + input_name_len_with_overhead, 4695 name_pos - utf8ness_cache, char); 4696 Copy(delimited, 4697 utf8ness_cache, 4698 input_name_len_with_overhead - 1, char); 4699 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0'; 4700 } 4701 4702 /* free only when not using the buffer */ 4703 if ( delimited != buffer ) Safefree(delimited); 4704 Safefree(save_input_locale); 4705 return is_utf8; 4706 } 4707 4708 /* Here we don't have stored the utf8ness for the input locale. We have to 4709 * calculate it */ 4710 4711 # if defined(USE_LOCALE_CTYPE) \ 4712 && ( defined(HAS_NL_LANGINFO) \ 4713 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC))) 4714 4715 { 4716 const char *original_ctype_locale 4717 = switch_category_locale_to_template(LC_CTYPE, 4718 category, 4719 save_input_locale); 4720 4721 /* Here the current LC_CTYPE is set to the locale of the category whose 4722 * information is desired. This means that nl_langinfo() and mbtowc() 4723 * should give the correct results */ 4724 4725 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding 4726 calling the functions if we have this */ 4727 4728 /* Standard UTF-8 needs at least 4 bytes to represent the maximum 4729 * Unicode code point. */ 4730 4731 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n", 4732 __FILE__, __LINE__, (int) MB_CUR_MAX)); 4733 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) { 4734 is_utf8 = FALSE; 4735 restore_switched_locale(LC_CTYPE, original_ctype_locale); 4736 goto finish_and_return; 4737 } 4738 4739 # endif 4740 # if defined(HAS_NL_LANGINFO) 4741 4742 { /* The task is easiest if the platform has this POSIX 2001 function. 4743 Except on some platforms it can wrongly return "", so have to have 4744 a fallback. And it can return that it's UTF-8, even if there are 4745 variances from that. For example, Turkish locales may use the 4746 alternate dotted I rules, and sometimes it appears to be a 4747 defective locale definition. XXX We should probably check for 4748 these in the Latin1 range and warn (but on glibc, requires 4749 iswalnum() etc. due to their not handling 80-FF correctly */ 4750 const char *codeset = my_nl_langinfo(CODESET, FALSE); 4751 /* FALSE => already in dest locale */ 4752 4753 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 4754 "\tnllanginfo returned CODESET '%s'\n", codeset)); 4755 4756 if (codeset && strNE(codeset, "")) { 4757 4758 /* If the implementation of foldEQ() somehow were 4759 * to change to not go byte-by-byte, this could 4760 * read past end of string, as only one length is 4761 * checked. But currently, a premature NUL will 4762 * compare false, and it will stop there */ 4763 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8")) 4764 || foldEQ(codeset, STR_WITH_LEN("UTF8"))); 4765 4766 DEBUG_L(PerlIO_printf(Perl_debug_log, 4767 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n", 4768 codeset, is_utf8)); 4769 restore_switched_locale(LC_CTYPE, original_ctype_locale); 4770 goto finish_and_return; 4771 } 4772 } 4773 4774 # endif 4775 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC) 4776 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a 4777 * late adder to C89, so very likely to have it. However, testing has 4778 * shown that, like nl_langinfo() above, there are locales that are not 4779 * strictly UTF-8 that this will return that they are */ 4780 4781 { 4782 wchar_t wc; 4783 int len; 4784 dSAVEDERRNO; 4785 4786 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS) 4787 4788 mbstate_t ps; 4789 4790 # endif 4791 4792 /* mbrtowc() and mbtowc() convert a byte string to a wide 4793 * character. Feed a byte string to one of them and check that the 4794 * result is the expected Unicode code point */ 4795 4796 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS) 4797 /* Prefer this function if available, as it's reentrant */ 4798 4799 memset(&ps, 0, sizeof(ps));; 4800 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift 4801 state */ 4802 SETERRNO(0, 0); 4803 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps); 4804 SAVE_ERRNO; 4805 4806 # else 4807 4808 LOCALE_LOCK; 4809 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */ 4810 SETERRNO(0, 0); 4811 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8)); 4812 SAVE_ERRNO; 4813 LOCALE_UNLOCK; 4814 4815 # endif 4816 4817 RESTORE_ERRNO; 4818 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 4819 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n", 4820 len, (unsigned int) wc, GET_ERRNO)); 4821 4822 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8) 4823 && wc == (wchar_t) UNICODE_REPLACEMENT); 4824 } 4825 4826 # endif 4827 4828 restore_switched_locale(LC_CTYPE, original_ctype_locale); 4829 goto finish_and_return; 4830 } 4831 4832 # else 4833 4834 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next 4835 * try looking at the currency symbol to see if it disambiguates 4836 * things. Often that will be in the native script, and if the symbol 4837 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII 4838 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8 4839 * string being valid UTF-8 are quite small */ 4840 4841 # ifdef USE_LOCALE_MONETARY 4842 4843 /* If have LC_MONETARY, we can look at the currency symbol. Often that 4844 * will be in the native script. We do this one first because there is 4845 * just one string to examine, so potentially avoids work */ 4846 4847 { 4848 const char *original_monetary_locale 4849 = switch_category_locale_to_template(LC_MONETARY, 4850 category, 4851 save_input_locale); 4852 bool only_ascii = FALSE; 4853 const U8 * currency_string 4854 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE); 4855 /* 2nd param not relevant for this item */ 4856 const U8 * first_variant; 4857 4858 assert( *currency_string == '-' 4859 || *currency_string == '+' 4860 || *currency_string == '.'); 4861 4862 currency_string++; 4863 4864 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant)) 4865 { 4866 DEBUG_L(PerlIO_printf(Perl_debug_log, "Couldn't get currency symbol for %s, or contains only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale)); 4867 only_ascii = TRUE; 4868 } 4869 else { 4870 is_utf8 = is_strict_utf8_string(first_variant, 0); 4871 } 4872 4873 restore_switched_locale(LC_MONETARY, original_monetary_locale); 4874 4875 if (! only_ascii) { 4876 4877 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8; 4878 * otherwise assume the locale is UTF-8 if and only if the symbol 4879 * is non-ascii UTF-8. */ 4880 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n", 4881 save_input_locale, is_utf8)); 4882 goto finish_and_return; 4883 } 4884 } 4885 4886 # endif /* USE_LOCALE_MONETARY */ 4887 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME) 4888 4889 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try 4890 * the names of the months and weekdays, timezone, and am/pm indicator */ 4891 { 4892 const char *original_time_locale 4893 = switch_category_locale_to_template(LC_TIME, 4894 category, 4895 save_input_locale); 4896 int hour = 10; 4897 bool is_dst = FALSE; 4898 int dom = 1; 4899 int month = 0; 4900 int i; 4901 char * formatted_time; 4902 4903 /* Here the current LC_TIME is set to the locale of the category 4904 * whose information is desired. Look at all the days of the week and 4905 * month names, and the timezone and am/pm indicator for UTF-8 variant 4906 * characters. The first such a one found will tell us if the locale 4907 * is UTF-8 or not */ 4908 4909 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */ 4910 formatted_time = my_strftime("%A %B %Z %p", 4911 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst); 4912 if ( ! formatted_time 4913 || is_utf8_invariant_string((U8 *) formatted_time, 0)) 4914 { 4915 4916 /* Here, we didn't find a non-ASCII. Try the next time through 4917 * with the complemented dst and am/pm, and try with the next 4918 * weekday. After we have gotten all weekdays, try the next 4919 * month */ 4920 is_dst = ! is_dst; 4921 hour = (hour + 12) % 24; 4922 dom++; 4923 if (i > 6) { 4924 month++; 4925 } 4926 continue; 4927 } 4928 4929 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8; 4930 * false otherwise. But first, restore LC_TIME to its original 4931 * locale if we changed it */ 4932 restore_switched_locale(LC_TIME, original_time_locale); 4933 4934 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n", 4935 save_input_locale, 4936 is_utf8_string((U8 *) formatted_time, 0))); 4937 is_utf8 = is_utf8_string((U8 *) formatted_time, 0); 4938 goto finish_and_return; 4939 } 4940 4941 /* Falling off the end of the loop indicates all the names were just 4942 * ASCII. Go on to the next test. If we changed it, restore LC_TIME 4943 * to its original locale */ 4944 restore_switched_locale(LC_TIME, original_time_locale); 4945 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "All time-related words for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale)); 4946 } 4947 4948 # endif 4949 4950 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST) 4951 4952 /* This code is ifdefd out because it was found to not be necessary in testing 4953 * on our dromedary test machine, which has over 700 locales. There, this 4954 * added no value to looking at the currency symbol and the time strings. I 4955 * left it in so as to avoid rewriting it if real-world experience indicates 4956 * that dromedary is an outlier. Essentially, instead of returning abpve if we 4957 * haven't found illegal utf8, we continue on and examine all the strerror() 4958 * messages on the platform for utf8ness. If all are ASCII, we still don't 4959 * know the answer; but otherwise we have a pretty good indication of the 4960 * utf8ness. The reason this doesn't help much is that the messages may not 4961 * have been translated into the locale. The currency symbol and time strings 4962 * are much more likely to have been translated. */ 4963 { 4964 int e; 4965 bool non_ascii = FALSE; 4966 const char *original_messages_locale 4967 = switch_category_locale_to_template(LC_MESSAGES, 4968 category, 4969 save_input_locale); 4970 const char * errmsg = NULL; 4971 4972 /* Here the current LC_MESSAGES is set to the locale of the category 4973 * whose information is desired. Look through all the messages. We 4974 * can't use Strerror() here because it may expand to code that 4975 * segfaults in miniperl */ 4976 4977 for (e = 0; e <= sys_nerr; e++) { 4978 errno = 0; 4979 errmsg = sys_errlist[e]; 4980 if (errno || !errmsg) { 4981 break; 4982 } 4983 errmsg = savepv(errmsg); 4984 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) { 4985 non_ascii = TRUE; 4986 is_utf8 = is_utf8_string((U8 *) errmsg, 0); 4987 break; 4988 } 4989 } 4990 Safefree(errmsg); 4991 4992 restore_switched_locale(LC_MESSAGES, original_messages_locale); 4993 4994 if (non_ascii) { 4995 4996 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid, 4997 * any non-ascii means it is one; otherwise we assume it isn't */ 4998 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n", 4999 save_input_locale, 5000 is_utf8)); 5001 goto finish_and_return; 5002 } 5003 5004 DEBUG_L(PerlIO_printf(Perl_debug_log, "All error messages for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale)); 5005 } 5006 5007 # endif 5008 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a 5009 UTF-8 locale */ 5010 5011 /* As a last resort, look at the locale name to see if it matches 5012 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the 5013 * return of setlocale(), is actually defined to be opaque, so we can't 5014 * really rely on the absence of various substrings in the name to indicate 5015 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to 5016 * be a UTF-8 locale. Similarly for the other common names */ 5017 5018 { 5019 const Size_t final_pos = strlen(save_input_locale) - 1; 5020 5021 if (final_pos >= 3) { 5022 const char *name = save_input_locale; 5023 5024 /* Find next 'U' or 'u' and look from there */ 5025 while ((name += strcspn(name, "Uu") + 1) 5026 <= save_input_locale + final_pos - 2) 5027 { 5028 if ( isALPHA_FOLD_NE(*name, 't') 5029 || isALPHA_FOLD_NE(*(name + 1), 'f')) 5030 { 5031 continue; 5032 } 5033 name += 2; 5034 if (*(name) == '-') { 5035 if ((name > save_input_locale + final_pos - 1)) { 5036 break; 5037 } 5038 name++; 5039 } 5040 if (*(name) == '8') { 5041 DEBUG_L(PerlIO_printf(Perl_debug_log, 5042 "Locale %s ends with UTF-8 in name\n", 5043 save_input_locale)); 5044 is_utf8 = TRUE; 5045 goto finish_and_return; 5046 } 5047 } 5048 DEBUG_L(PerlIO_printf(Perl_debug_log, 5049 "Locale %s doesn't end with UTF-8 in name\n", 5050 save_input_locale)); 5051 } 5052 5053 # ifdef WIN32 5054 5055 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */ 5056 if (memENDs(save_input_locale, final_pos, "65001")) { 5057 DEBUG_L(PerlIO_printf(Perl_debug_log, 5058 "Locale %s ends with 65001 in name, is UTF-8 locale\n", 5059 save_input_locale)); 5060 is_utf8 = TRUE; 5061 goto finish_and_return; 5062 } 5063 5064 # endif 5065 } 5066 # endif 5067 5068 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But 5069 * since we are about to return FALSE anyway, there is no point in doing 5070 * this extra work */ 5071 5072 # if 0 5073 if (instr(save_input_locale, "8859")) { 5074 DEBUG_L(PerlIO_printf(Perl_debug_log, 5075 "Locale %s has 8859 in name, not UTF-8 locale\n", 5076 save_input_locale)); 5077 is_utf8 = FALSE; 5078 goto finish_and_return; 5079 } 5080 # endif 5081 5082 DEBUG_L(PerlIO_printf(Perl_debug_log, 5083 "Assuming locale %s is not a UTF-8 locale\n", 5084 save_input_locale)); 5085 is_utf8 = FALSE; 5086 5087 # endif /* the code that is compiled when no modern LC_CTYPE */ 5088 5089 finish_and_return: 5090 5091 /* Cache this result so we don't have to go through all this next time. */ 5092 utf8ness_cache_size = sizeof(PL_locale_utf8ness) 5093 - (utf8ness_cache - PL_locale_utf8ness); 5094 5095 /* But we can't save it if it is too large for the total space available */ 5096 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) { 5097 Size_t utf8ness_cache_len = strlen(utf8ness_cache); 5098 5099 /* Here it can fit, but we may need to clear out the oldest cached 5100 * result(s) to do so. Check */ 5101 if (utf8ness_cache_len + input_name_len_with_overhead 5102 >= utf8ness_cache_size) 5103 { 5104 /* Here we have to clear something out to make room for this. 5105 * Start looking at the rightmost place where it could fit and find 5106 * the beginning of the entry that extends past that. */ 5107 char * cutoff = (char *) my_memrchr(utf8ness_cache, 5108 UTF8NESS_SEP[0], 5109 utf8ness_cache_size 5110 - input_name_len_with_overhead); 5111 5112 assert(cutoff); 5113 assert(cutoff >= utf8ness_cache); 5114 5115 /* This and all subsequent entries must be removed */ 5116 *cutoff = '\0'; 5117 utf8ness_cache_len = strlen(utf8ness_cache); 5118 } 5119 5120 /* Make space for the new entry */ 5121 Move(utf8ness_cache, 5122 utf8ness_cache + input_name_len_with_overhead, 5123 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char); 5124 5125 /* And insert it */ 5126 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char); 5127 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0'; 5128 5129 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1] & ~1) != '0') { 5130 Perl_croak(aTHX_ 5131 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu," 5132 " inserted_name=%s, its_len=%zu\n", 5133 __FILE__, __LINE__, 5134 PL_locale_utf8ness, strlen(PL_locale_utf8ness), 5135 delimited, input_name_len_with_overhead); 5136 } 5137 } 5138 5139 # ifdef DEBUGGING 5140 5141 if (DEBUG_Lv_TEST) { 5142 const char * s = PL_locale_utf8ness; 5143 5144 /* Audit the structure */ 5145 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) { 5146 const char *e; 5147 5148 if (*s != UTF8NESS_SEP[0]) { 5149 Perl_croak(aTHX_ 5150 "panic: %s: %d: Corrupt utf8ness_cache: missing" 5151 " separator %.*s<-- HERE %s\n", 5152 __FILE__, __LINE__, 5153 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness, 5154 s); 5155 } 5156 s++; 5157 e = strchr(s, UTF8NESS_PREFIX[0]); 5158 if (! e) { 5159 e = PL_locale_utf8ness + strlen(PL_locale_utf8ness); 5160 Perl_croak(aTHX_ 5161 "panic: %s: %d: Corrupt utf8ness_cache: missing" 5162 " separator %.*s<-- HERE %s\n", 5163 __FILE__, __LINE__, 5164 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness, 5165 e); 5166 } 5167 e++; 5168 if (*e != '0' && *e != '1') { 5169 Perl_croak(aTHX_ 5170 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness" 5171 " must be [01] %.*s<-- HERE %s\n", 5172 __FILE__, __LINE__, 5173 (int) (e + 1 - PL_locale_utf8ness), 5174 PL_locale_utf8ness, e + 1); 5175 } 5176 if (ninstr(PL_locale_utf8ness, s, s-1, e)) { 5177 Perl_croak(aTHX_ 5178 "panic: %s: %d: Corrupt utf8ness_cache: entry" 5179 " has duplicate %.*s<-- HERE %s\n", 5180 __FILE__, __LINE__, 5181 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness, 5182 e); 5183 } 5184 s = e + 1; 5185 } 5186 } 5187 5188 if (DEBUG_Lv_TEST || debug_initialization) { 5189 5190 PerlIO_printf(Perl_debug_log, 5191 "PL_locale_utf8ness is now %s; returning %d\n", 5192 PL_locale_utf8ness, is_utf8); 5193 } 5194 5195 # endif 5196 5197 /* free only when not using the buffer */ 5198 if ( delimited != buffer ) Safefree(delimited); 5199 Safefree(save_input_locale); 5200 return is_utf8; 5201 } 5202 5203 #endif 5204 5205 bool 5206 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category) 5207 { 5208 dVAR; 5209 /* Internal function which returns if we are in the scope of a pragma that 5210 * enables the locale category 'category'. 'compiling' should indicate if 5211 * this is during the compilation phase (TRUE) or not (FALSE). */ 5212 5213 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop; 5214 5215 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0); 5216 if (! these_categories || these_categories == &PL_sv_placeholder) { 5217 return FALSE; 5218 } 5219 5220 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get 5221 * a valid unsigned */ 5222 assert(category >= -1); 5223 return cBOOL(SvUV(these_categories) & (1U << (category + 1))); 5224 } 5225 5226 char * 5227 Perl_my_strerror(pTHX_ const int errnum) 5228 { 5229 /* Returns a mortalized copy of the text of the error message associated 5230 * with 'errnum'. It uses the current locale's text unless the platform 5231 * doesn't have the LC_MESSAGES category or we are not being called from 5232 * within the scope of 'use locale'. In the former case, it uses whatever 5233 * strerror returns; in the latter case it uses the text from the C locale. 5234 * 5235 * The function just calls strerror(), but temporarily switches, if needed, 5236 * to the C locale */ 5237 5238 char *errstr; 5239 dVAR; 5240 5241 #ifndef USE_LOCALE_MESSAGES 5242 5243 /* If platform doesn't have messages category, we don't do any switching to 5244 * the C locale; we just use whatever strerror() returns */ 5245 5246 errstr = savepv(Strerror(errnum)); 5247 5248 #else /* Has locale messages */ 5249 5250 const bool within_locale_scope = IN_LC(LC_MESSAGES); 5251 5252 # ifndef USE_ITHREADS 5253 5254 /* This function is trivial without threads. */ 5255 if (within_locale_scope) { 5256 errstr = savepv(strerror(errnum)); 5257 } 5258 else { 5259 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL)); 5260 5261 do_setlocale_c(LC_MESSAGES, "C"); 5262 errstr = savepv(strerror(errnum)); 5263 do_setlocale_c(LC_MESSAGES, save_locale); 5264 Safefree(save_locale); 5265 } 5266 5267 # elif defined(USE_POSIX_2008_LOCALE) \ 5268 && defined(HAS_STRERROR_L) \ 5269 && defined(HAS_DUPLOCALE) 5270 5271 /* This function is also trivial if we don't have to worry about thread 5272 * safety and have strerror_l(), as it handles the switch of locales so we 5273 * don't have to deal with that. We don't have to worry about thread 5274 * safety if strerror_r() is also available. Both it and strerror_l() are 5275 * thread-safe. Plain strerror() isn't thread safe. But on threaded 5276 * builds when strerror_r() is available, the apparent call to strerror() 5277 * below is actually a macro that behind-the-scenes calls strerror_r(). */ 5278 5279 # ifdef HAS_STRERROR_R 5280 5281 if (within_locale_scope) { 5282 errstr = savepv(strerror(errnum)); 5283 } 5284 else { 5285 errstr = savepv(strerror_l(errnum, PL_C_locale_obj)); 5286 } 5287 5288 # else 5289 5290 /* Here we have strerror_l(), but not strerror_r() and we are on a 5291 * threaded-build. We use strerror_l() for everything, constructing a 5292 * locale to pass to it if necessary */ 5293 5294 bool do_free = FALSE; 5295 locale_t locale_to_use; 5296 5297 if (within_locale_scope) { 5298 locale_to_use = uselocale((locale_t) 0); 5299 if (locale_to_use == LC_GLOBAL_LOCALE) { 5300 locale_to_use = duplocale(LC_GLOBAL_LOCALE); 5301 do_free = TRUE; 5302 } 5303 } 5304 else { /* Use C locale if not within 'use locale' scope */ 5305 locale_to_use = PL_C_locale_obj; 5306 } 5307 5308 errstr = savepv(strerror_l(errnum, locale_to_use)); 5309 5310 if (do_free) { 5311 freelocale(locale_to_use); 5312 } 5313 5314 # endif 5315 # else /* Doesn't have strerror_l() */ 5316 5317 const char * save_locale = NULL; 5318 bool locale_is_C = FALSE; 5319 5320 /* We have a critical section to prevent another thread from executing this 5321 * same code at the same time. (On thread-safe perls, the LOCK is a 5322 * no-op.) Since this is the only place in core that changes LC_MESSAGES 5323 * (unless the user has called setlocale(), this works to prevent races. */ 5324 LOCALE_LOCK; 5325 5326 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 5327 "my_strerror called with errnum %d\n", errnum)); 5328 if (! within_locale_scope) { 5329 save_locale = do_setlocale_c(LC_MESSAGES, NULL); 5330 if (! save_locale) { 5331 Perl_croak(aTHX_ 5332 "panic: %s: %d: Could not find current LC_MESSAGES locale," 5333 " errno=%d\n", __FILE__, __LINE__, errno); 5334 } 5335 else { 5336 locale_is_C = isNAME_C_OR_POSIX(save_locale); 5337 5338 /* Switch to the C locale if not already in it */ 5339 if (! locale_is_C) { 5340 5341 /* The setlocale() just below likely will zap 'save_locale', so 5342 * create a copy. */ 5343 save_locale = savepv(save_locale); 5344 do_setlocale_c(LC_MESSAGES, "C"); 5345 } 5346 } 5347 } /* end of ! within_locale_scope */ 5348 else { 5349 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n", 5350 __FILE__, __LINE__)); 5351 } 5352 5353 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 5354 "Any locale change has been done; about to call Strerror\n")); 5355 errstr = savepv(Strerror(errnum)); 5356 5357 if (! within_locale_scope) { 5358 if (save_locale && ! locale_is_C) { 5359 if (! do_setlocale_c(LC_MESSAGES, save_locale)) { 5360 Perl_croak(aTHX_ 5361 "panic: %s: %d: setlocale restore failed, errno=%d\n", 5362 __FILE__, __LINE__, errno); 5363 } 5364 Safefree(save_locale); 5365 } 5366 } 5367 5368 LOCALE_UNLOCK; 5369 5370 # endif /* End of doesn't have strerror_l */ 5371 # ifdef DEBUGGING 5372 5373 if (DEBUG_Lv_TEST) { 5374 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '"); 5375 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0); 5376 PerlIO_printf(Perl_debug_log, "'\n"); 5377 } 5378 5379 # endif 5380 #endif /* End of does have locale messages */ 5381 5382 SAVEFREEPV(errstr); 5383 return errstr; 5384 } 5385 5386 /* 5387 5388 =for apidoc switch_to_global_locale 5389 5390 On systems without locale support, or on typical single-threaded builds, or on 5391 platforms that do not support per-thread locale operations, this function does 5392 nothing. On such systems that do have locale support, only a locale global to 5393 the whole program is available. 5394 5395 On multi-threaded builds on systems that do have per-thread locale operations, 5396 this function converts the thread it is running in to use the global locale. 5397 This is for code that has not yet or cannot be updated to handle multi-threaded 5398 locale operation. As long as only a single thread is so-converted, everything 5399 works fine, as all the other threads continue to ignore the global one, so only 5400 this thread looks at it. 5401 5402 However, on Windows systems this isn't quite true prior to Visual Studio 15, 5403 at which point Microsoft fixed a bug. A race can occur if you use the 5404 following operations on earlier Windows platforms: 5405 5406 =over 5407 5408 =item L<POSIX::localeconv|POSIX/localeconv> 5409 5410 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP> 5411 5412 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP> 5413 5414 =back 5415 5416 The first item is not fixable (except by upgrading to a later Visual Studio 5417 release), but it would be possible to work around the latter two items by using 5418 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches 5419 welcome. 5420 5421 Without this function call, threads that use the L<C<setlocale(3)>> system 5422 function will not work properly, as all the locale-sensitive functions will 5423 look at the per-thread locale, and C<setlocale> will have no effect on this 5424 thread. 5425 5426 Perl code should convert to either call 5427 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system 5428 C<setlocale>) or use the methods given in L<perlcall> to call 5429 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly 5430 handle all cases of single- vs multi-thread, POSIX 2008-supported or not. 5431 5432 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can 5433 continue to work if this function is called before transferring control to the 5434 library. 5435 5436 Upon return from the code that needs to use the global locale, 5437 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe 5438 multi-thread operation. 5439 5440 =cut 5441 */ 5442 5443 void 5444 Perl_switch_to_global_locale() 5445 { 5446 5447 #ifdef USE_THREAD_SAFE_LOCALE 5448 # ifdef WIN32 5449 5450 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE); 5451 5452 # else 5453 # ifdef HAS_QUERYLOCALE 5454 5455 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0))); 5456 5457 # else 5458 5459 { 5460 unsigned int i; 5461 5462 for (i = 0; i < LC_ALL_INDEX; i++) { 5463 setlocale(categories[i], do_setlocale_r(categories[i], NULL)); 5464 } 5465 } 5466 5467 # endif 5468 5469 uselocale(LC_GLOBAL_LOCALE); 5470 5471 # endif 5472 #endif 5473 5474 } 5475 5476 /* 5477 5478 =for apidoc sync_locale 5479 5480 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or 5481 change the locale (though changing the locale is antisocial and dangerous on 5482 multi-threaded systems that don't have multi-thread safe locale operations. 5483 (See L<perllocale/Multi-threaded operation>). Using the system 5484 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries 5485 called from XS, such as C<Gtk> do so, and this can't be changed. When the 5486 locale is changed by XS code that didn't use 5487 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the 5488 locale has changed. Use this function to do so, before returning to Perl. 5489 5490 The return value is a boolean: TRUE if the global locale at the time of call 5491 was in effect; and FALSE if a per-thread locale was in effect. This can be 5492 used by the caller that needs to restore things as-they-were to decide whether 5493 or not to call 5494 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>. 5495 5496 =cut 5497 */ 5498 5499 bool 5500 Perl_sync_locale() 5501 { 5502 5503 #ifndef USE_LOCALE 5504 5505 return TRUE; 5506 5507 #else 5508 5509 const char * newlocale; 5510 dTHX; 5511 5512 # ifdef USE_POSIX_2008_LOCALE 5513 5514 bool was_in_global_locale = FALSE; 5515 locale_t cur_obj = uselocale((locale_t) 0); 5516 5517 /* On Windows, unless the foreign code has turned off the thread-safe 5518 * locale setting, any plain setlocale() will have affected what we see, so 5519 * no need to worry. Otherwise, If the foreign code has done a plain 5520 * setlocale(), it will only affect the global locale on POSIX systems, but 5521 * will affect the */ 5522 if (cur_obj == LC_GLOBAL_LOCALE) { 5523 5524 # ifdef HAS_QUERY_LOCALE 5525 5526 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL)); 5527 5528 # else 5529 5530 unsigned int i; 5531 5532 /* We can't trust that we can read the LC_ALL format on the 5533 * platform, so do them individually */ 5534 for (i = 0; i < LC_ALL_INDEX; i++) { 5535 do_setlocale_r(categories[i], setlocale(categories[i], NULL)); 5536 } 5537 5538 # endif 5539 5540 was_in_global_locale = TRUE; 5541 } 5542 5543 # else 5544 5545 bool was_in_global_locale = TRUE; 5546 5547 # endif 5548 # ifdef USE_LOCALE_CTYPE 5549 5550 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL)); 5551 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 5552 "%s:%d: %s\n", __FILE__, __LINE__, 5553 setlocale_debug_string(LC_CTYPE, NULL, newlocale))); 5554 new_ctype(newlocale); 5555 Safefree(newlocale); 5556 5557 # endif /* USE_LOCALE_CTYPE */ 5558 # ifdef USE_LOCALE_COLLATE 5559 5560 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL)); 5561 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 5562 "%s:%d: %s\n", __FILE__, __LINE__, 5563 setlocale_debug_string(LC_COLLATE, NULL, newlocale))); 5564 new_collate(newlocale); 5565 Safefree(newlocale); 5566 5567 # endif 5568 # ifdef USE_LOCALE_NUMERIC 5569 5570 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL)); 5571 DEBUG_Lv(PerlIO_printf(Perl_debug_log, 5572 "%s:%d: %s\n", __FILE__, __LINE__, 5573 setlocale_debug_string(LC_NUMERIC, NULL, newlocale))); 5574 new_numeric(newlocale); 5575 Safefree(newlocale); 5576 5577 # endif /* USE_LOCALE_NUMERIC */ 5578 5579 return was_in_global_locale; 5580 5581 #endif 5582 5583 } 5584 5585 #if defined(DEBUGGING) && defined(USE_LOCALE) 5586 5587 STATIC char * 5588 S_setlocale_debug_string(const int category, /* category number, 5589 like LC_ALL */ 5590 const char* const locale, /* locale name */ 5591 5592 /* return value from setlocale() when attempting to 5593 * set 'category' to 'locale' */ 5594 const char* const retval) 5595 { 5596 /* Returns a pointer to a NUL-terminated string in static storage with 5597 * added text about the info passed in. This is not thread safe and will 5598 * be overwritten by the next call, so this should be used just to 5599 * formulate a string to immediately print or savepv() on. */ 5600 5601 /* initialise to a non-null value to keep it out of BSS and so keep 5602 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */ 5603 static char ret[256] = "If you can read this, thank your buggy C" 5604 " library strlcpy(), and change your hints file" 5605 " to undef it"; 5606 5607 my_strlcpy(ret, "setlocale(", sizeof(ret)); 5608 my_strlcat(ret, category_name(category), sizeof(ret)); 5609 my_strlcat(ret, ", ", sizeof(ret)); 5610 5611 if (locale) { 5612 my_strlcat(ret, "\"", sizeof(ret)); 5613 my_strlcat(ret, locale, sizeof(ret)); 5614 my_strlcat(ret, "\"", sizeof(ret)); 5615 } 5616 else { 5617 my_strlcat(ret, "NULL", sizeof(ret)); 5618 } 5619 5620 my_strlcat(ret, ") returned ", sizeof(ret)); 5621 5622 if (retval) { 5623 my_strlcat(ret, "\"", sizeof(ret)); 5624 my_strlcat(ret, retval, sizeof(ret)); 5625 my_strlcat(ret, "\"", sizeof(ret)); 5626 } 5627 else { 5628 my_strlcat(ret, "NULL", sizeof(ret)); 5629 } 5630 5631 assert(strlen(ret) < sizeof(ret)); 5632 5633 return ret; 5634 } 5635 5636 #endif 5637 5638 void 5639 Perl_thread_locale_init() 5640 { 5641 /* Called from a thread on startup*/ 5642 5643 #ifdef USE_THREAD_SAFE_LOCALE 5644 5645 dTHX_DEBUGGING; 5646 5647 /* C starts the new thread in the global C locale. If we are thread-safe, 5648 * we want to not be in the global locale */ 5649 5650 DEBUG_L(PerlIO_printf(Perl_debug_log, 5651 "%s:%d: new thread, initial locale is %s; calling setlocale\n", 5652 __FILE__, __LINE__, setlocale(LC_ALL, NULL))); 5653 5654 # ifdef WIN32 5655 5656 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE); 5657 5658 # else 5659 5660 Perl_setlocale(LC_ALL, "C"); 5661 5662 # endif 5663 #endif 5664 5665 } 5666 5667 void 5668 Perl_thread_locale_term() 5669 { 5670 /* Called from a thread as it gets ready to terminate */ 5671 5672 #ifdef USE_THREAD_SAFE_LOCALE 5673 5674 /* C starts the new thread in the global C locale. If we are thread-safe, 5675 * we want to not be in the global locale */ 5676 5677 # ifndef WIN32 5678 5679 { /* Free up */ 5680 dVAR; 5681 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE); 5682 if (cur_obj != LC_GLOBAL_LOCALE && cur_obj != PL_C_locale_obj) { 5683 freelocale(cur_obj); 5684 } 5685 } 5686 5687 # endif 5688 #endif 5689 5690 } 5691 5692 /* 5693 * ex: set ts=8 sts=4 sw=4 et: 5694 */ 5695