3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
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.
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!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
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
38 #define PERL_IN_LOCALE_C
39 #include "perl_langinfo.h"
44 /* If the environment says to, we can output debugging information during
45 * initialization. This is done before option parsing, and before any thread
46 * creation, so can be a file-level static */
48 # ifdef PERL_GLOBAL_STRUCT
49 /* no global syms allowed */
50 # define debug_initialization 0
51 # define DEBUG_INITIALIZATION_set(v)
53 static bool debug_initialization = FALSE;
54 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
58 /* strlen() of a literal string constant. XXX We might want this more general,
59 * but using it in just this file for now */
60 #define STRLENs(s) (sizeof("" s "") - 1)
62 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
63 * return of setlocale(), then this is extremely likely to be the C or POSIX
64 * locale. However, the output of setlocale() is documented to be opaque, but
65 * the odds are extremely small that it would return these two strings for some
66 * other locale. Note that VMS in these two locales includes many non-ASCII
67 * characters as controls and punctuation (below are hex bytes):
69 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
70 * Oddly, none there are listed as alphas, though some represent alphabetics
71 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
72 #define isNAME_C_OR_POSIX(name) \
74 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
75 || strEQ((name), "POSIX")))
80 * Standardize the locale name from a string returned by 'setlocale', possibly
81 * modifying that string.
83 * The typical return value of setlocale() is either
84 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
85 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
86 * (the space-separated values represent the various sublocales,
87 * in some unspecified order). This is not handled by this function.
89 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
90 * which is harmful for further use of the string in setlocale(). This
91 * function removes the trailing new line and everything up through the '='
95 S_stdize_locale(pTHX_ char *locs)
97 const char * const s = strchr(locs, '=');
100 PERL_ARGS_ASSERT_STDIZE_LOCALE;
103 const char * const t = strchr(s, '.');
106 const char * const u = strchr(t, '\n');
107 if (u && (u[1] == 0)) {
108 const STRLEN len = u - s;
109 Move(s + 1, locs, len, char);
117 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
122 /* Two parallel arrays; first the locale categories Perl uses on this system;
123 * the second array is their names. These arrays are in mostly arbitrary
126 const int categories[] = {
128 # ifdef USE_LOCALE_NUMERIC
131 # ifdef USE_LOCALE_CTYPE
134 # ifdef USE_LOCALE_COLLATE
137 # ifdef USE_LOCALE_TIME
140 # ifdef USE_LOCALE_MESSAGES
143 # ifdef USE_LOCALE_MONETARY
149 -1 /* Placeholder because C doesn't allow a
150 trailing comma, and it would get complicated
151 with all the #ifdef's */
154 /* The top-most real element is LC_ALL */
156 const char * category_names[] = {
158 # ifdef USE_LOCALE_NUMERIC
161 # ifdef USE_LOCALE_CTYPE
164 # ifdef USE_LOCALE_COLLATE
167 # ifdef USE_LOCALE_TIME
170 # ifdef USE_LOCALE_MESSAGES
173 # ifdef USE_LOCALE_MONETARY
179 NULL /* Placeholder */
184 /* On systems with LC_ALL, it is kept in the highest index position. (-2
185 * to account for the final unused placeholder element.) */
186 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
190 /* On systems without LC_ALL, we pretend it is there, one beyond the real
191 * top element, hence in the unused placeholder element. */
192 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
196 /* Pretending there is an LC_ALL element just above allows us to avoid most
197 * special cases. Most loops through these arrays in the code below are
198 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
199 * on either type of system. But the code must be written to not access the
200 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
201 * checked for at compile time by using the #define LC_ALL_INDEX which is only
202 * defined if we do have LC_ALL. */
204 /* Now create LC_foo_INDEX #defines for just those categories on this system */
205 # ifdef USE_LOCALE_NUMERIC
206 # define LC_NUMERIC_INDEX 0
207 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
209 # define _DUMMY_NUMERIC -1
211 # ifdef USE_LOCALE_CTYPE
212 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
213 # define _DUMMY_CTYPE LC_CTYPE_INDEX
215 # define _DUMMY_CTYPE _DUMMY_NUMERIC
217 # ifdef USE_LOCALE_COLLATE
218 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
219 # define _DUMMY_COLLATE LC_COLLATE_INDEX
221 # define _DUMMY_COLLATE _DUMMY_COLLATE
223 # ifdef USE_LOCALE_TIME
224 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
225 # define _DUMMY_TIME LC_TIME_INDEX
227 # define _DUMMY_TIME _DUMMY_COLLATE
229 # ifdef USE_LOCALE_MESSAGES
230 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
231 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
233 # define _DUMMY_MESSAGES _DUMMY_TIME
235 # ifdef USE_LOCALE_MONETARY
236 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
237 # define _DUMMY_MONETARY LC_MONETARY_INDEX
239 # define _DUMMY_MONETARY _DUMMY_MESSAGES
242 # define LC_ALL_INDEX _DUMMY_MONETARY + 1
244 #endif /* ifdef USE_LOCALE */
246 /* Windows requres a customized base-level setlocale() */
248 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
250 # define my_setlocale(cat, locale) setlocale(cat, locale)
253 /* Just placeholders for now. "_c" is intended to be called when the category
254 * is a constant known at compile time; "_r", not known until run time */
255 # define do_setlocale_c(category, locale) my_setlocale(category, locale)
256 # define do_setlocale_r(category, locale) my_setlocale(category, locale)
259 S_set_numeric_radix(pTHX_ const bool use_locale)
261 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
262 * TRUE, use the radix character derived from the current locale */
264 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
265 || defined(HAS_NL_LANGINFO))
267 /* We only set up the radix SV if we are to use a locale radix ... */
269 const char * radix = my_nl_langinfo(PERL_RADIXCHAR, FALSE);
270 /* FALSE => already in dest locale */
272 /* ... and the character being used isn't a dot */
273 if (strNE(radix, ".")) {
274 if (PL_numeric_radix_sv) {
275 sv_setpv(PL_numeric_radix_sv, radix);
278 PL_numeric_radix_sv = newSVpv(radix, 0);
281 if ( ! is_utf8_invariant_string(
282 (U8 *) SvPVX(PL_numeric_radix_sv), SvCUR(PL_numeric_radix_sv))
284 (U8 *) SvPVX(PL_numeric_radix_sv), SvCUR(PL_numeric_radix_sv))
285 && _is_cur_LC_category_utf8(LC_NUMERIC))
287 SvUTF8_on(PL_numeric_radix_sv);
293 SvREFCNT_dec(PL_numeric_radix_sv);
294 PL_numeric_radix_sv = NULL;
300 if (DEBUG_L_TEST || debug_initialization) {
301 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
302 (PL_numeric_radix_sv)
303 ? SvPVX(PL_numeric_radix_sv)
305 (PL_numeric_radix_sv)
306 ? cBOOL(SvUTF8(PL_numeric_radix_sv))
311 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
317 Perl_new_numeric(pTHX_ const char *newnum)
320 #ifndef USE_LOCALE_NUMERIC
322 PERL_UNUSED_ARG(newnum);
326 /* Called after all libc setlocale() calls affecting LC_NUMERIC, to tell
327 * core Perl this and that 'newnum' is the name of the new locale.
328 * It installs this locale as the current underlying default.
330 * The default locale and the C locale can be toggled between by use of the
331 * set_numeric_underlying() and set_numeric_standard() functions, which
332 * should probably not be called directly, but only via macros like
333 * SET_NUMERIC_STANDARD() in perl.h.
335 * The toggling is necessary mainly so that a non-dot radix decimal point
336 * character can be output, while allowing internal calculations to use a
339 * This sets several interpreter-level variables:
340 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
341 * PL_numeric_underlying A boolean indicating if the toggled state is such
342 * that the current locale is the program's underlying
344 * PL_numeric_standard An int indicating if the toggled state is such
345 * that the current locale is the C locale. If non-zero,
346 * it is in C; if > 1, it means it may not be toggled away
348 * Note that both of the last two variables can be true at the same time,
349 * if the underlying locale is C. (Toggling is a no-op under these
352 * Any code changing the locale (outside this file) should use
353 * POSIX::setlocale, which calls this function. Therefore this function
354 * should be called directly only from this file and from
355 * POSIX::setlocale() */
360 Safefree(PL_numeric_name);
361 PL_numeric_name = NULL;
362 PL_numeric_standard = TRUE;
363 PL_numeric_underlying = TRUE;
367 save_newnum = stdize_locale(savepv(newnum));
369 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
370 PL_numeric_underlying = TRUE;
372 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
373 Safefree(PL_numeric_name);
374 PL_numeric_name = save_newnum;
377 Safefree(save_newnum);
380 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
381 * have to worry about the radix being a non-dot. (Core operations that
382 * need the underlying locale change to it temporarily). */
383 set_numeric_standard();
385 #endif /* USE_LOCALE_NUMERIC */
390 Perl_set_numeric_standard(pTHX)
393 #ifdef USE_LOCALE_NUMERIC
395 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
396 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
397 * macro avoids calling this routine if toggling isn't necessary according
398 * to our records (which could be wrong if some XS code has changed the
399 * locale behind our back) */
401 do_setlocale_c(LC_NUMERIC, "C");
402 PL_numeric_standard = TRUE;
403 PL_numeric_underlying = isNAME_C_OR_POSIX(PL_numeric_name);
404 set_numeric_radix(0);
408 if (DEBUG_L_TEST || debug_initialization) {
409 PerlIO_printf(Perl_debug_log,
410 "LC_NUMERIC locale now is standard C\n");
414 #endif /* USE_LOCALE_NUMERIC */
419 Perl_set_numeric_underlying(pTHX)
422 #ifdef USE_LOCALE_NUMERIC
424 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
425 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
426 * instead of calling this directly. The macro avoids calling this routine
427 * if toggling isn't necessary according to our records (which could be
428 * wrong if some XS code has changed the locale behind our back) */
430 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
431 PL_numeric_standard = isNAME_C_OR_POSIX(PL_numeric_name);
432 PL_numeric_underlying = TRUE;
433 set_numeric_radix(1);
437 if (DEBUG_L_TEST || debug_initialization) {
438 PerlIO_printf(Perl_debug_log,
439 "LC_NUMERIC locale now is %s\n",
444 #endif /* USE_LOCALE_NUMERIC */
449 * Set up for a new ctype locale.
452 S_new_ctype(pTHX_ const char *newctype)
455 #ifndef USE_LOCALE_CTYPE
457 PERL_ARGS_ASSERT_NEW_CTYPE;
458 PERL_UNUSED_ARG(newctype);
463 /* Called after all libc setlocale() calls affecting LC_CTYPE, to tell
464 * core Perl this and that 'newctype' is the name of the new locale.
466 * This function sets up the folding arrays for all 256 bytes, assuming
467 * that tofold() is tolc() since fold case is not a concept in POSIX,
469 * Any code changing the locale (outside this file) should use
470 * POSIX::setlocale, which calls this function. Therefore this function
471 * should be called directly only from this file and from
472 * POSIX::setlocale() */
477 PERL_ARGS_ASSERT_NEW_CTYPE;
479 /* We will replace any bad locale warning with 1) nothing if the new one is
480 * ok; or 2) a new warning for the bad new locale */
481 if (PL_warn_locale) {
482 SvREFCNT_dec_NN(PL_warn_locale);
483 PL_warn_locale = NULL;
486 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
488 /* A UTF-8 locale gets standard rules. But note that code still has to
489 * handle this specially because of the three problematic code points */
490 if (PL_in_utf8_CTYPE_locale) {
491 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
494 /* Assume enough space for every character being bad. 4 spaces each
495 * for the 94 printable characters that are output like "'x' "; and 5
496 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
498 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
500 /* Don't check for problems if we are suppressing the warnings */
501 bool check_for_problems = ckWARN_d(WARN_LOCALE)
502 || UNLIKELY(DEBUG_L_TEST);
503 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
505 unsigned int bad_count = 0; /* Count of bad characters */
507 for (i = 0; i < 256; i++) {
508 if (isUPPER_LC((U8) i))
509 PL_fold_locale[i] = (U8) toLOWER_LC((U8) i);
510 else if (isLOWER_LC((U8) i))
511 PL_fold_locale[i] = (U8) toUPPER_LC((U8) i);
513 PL_fold_locale[i] = (U8) i;
515 /* If checking for locale problems, see if the native ASCII-range
516 * printables plus \n and \t are in their expected categories in
517 * the new locale. If not, this could mean big trouble, upending
518 * Perl's and most programs' assumptions, like having a
519 * metacharacter with special meaning become a \w. Fortunately,
520 * it's very rare to find locales that aren't supersets of ASCII
521 * nowadays. It isn't a problem for most controls to be changed
522 * into something else; we check only \n and \t, though perhaps \r
523 * could be an issue as well. */
524 if ( check_for_problems
525 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
527 if (( isALPHANUMERIC_A(i) && ! isALPHANUMERIC_LC(i))
528 || (isPUNCT_A(i) && ! isPUNCT_LC(i))
529 || (isBLANK_A(i) && ! isBLANK_LC(i))
530 || (i == '\n' && ! isCNTRL_LC(i)))
532 if (bad_count) { /* Separate multiple entries with a
534 bad_chars_list[bad_count++] = ' ';
536 bad_chars_list[bad_count++] = '\'';
538 bad_chars_list[bad_count++] = (char) i;
541 bad_chars_list[bad_count++] = '\\';
543 bad_chars_list[bad_count++] = 'n';
547 bad_chars_list[bad_count++] = 't';
550 bad_chars_list[bad_count++] = '\'';
551 bad_chars_list[bad_count] = '\0';
558 /* We only handle single-byte locales (outside of UTF-8 ones; so if
559 * this locale requires more than one byte, there are going to be
561 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
562 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
563 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
565 if (check_for_problems && MB_CUR_MAX > 1
567 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
568 * locale. Just assume that the implementation for them (plus
569 * for POSIX) is correct and the > 1 value is spurious. (Since
570 * these are specially handled to never be considered UTF-8
571 * locales, as long as this is the only problem, everything
572 * should work fine */
573 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
575 multi_byte_locale = TRUE;
580 if (bad_count || multi_byte_locale) {
581 PL_warn_locale = Perl_newSVpvf(aTHX_
582 "Locale '%s' may not work well.%s%s%s\n",
585 ? " Some characters in it are not recognized by"
589 ? "\nThe following characters (and maybe others)"
590 " may not have the same meaning as the Perl"
591 " program expects:\n"
597 /* If we are actually in the scope of the locale or are debugging,
598 * output the message now. If not in that scope, we save the
599 * message to be output at the first operation using this locale,
600 * if that actually happens. Most programs don't use locales, so
601 * they are immune to bad ones. */
602 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
604 /* We have to save 'newctype' because the setlocale() just
605 * below may destroy it. The next setlocale() further down
606 * should restore it properly so that the intermediate change
607 * here is transparent to this function's caller */
608 const char * const badlocale = savepv(newctype);
610 do_setlocale_c(LC_CTYPE, "C");
612 /* The '0' below suppresses a bogus gcc compiler warning */
613 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
615 do_setlocale_c(LC_CTYPE, badlocale);
618 if (IN_LC(LC_CTYPE)) {
619 SvREFCNT_dec_NN(PL_warn_locale);
620 PL_warn_locale = NULL;
626 #endif /* USE_LOCALE_CTYPE */
631 Perl__warn_problematic_locale()
634 #ifdef USE_LOCALE_CTYPE
638 /* Internal-to-core function that outputs the message in PL_warn_locale,
639 * and then NULLS it. Should be called only through the macro
640 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
642 if (PL_warn_locale) {
643 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
644 SvPVX(PL_warn_locale),
645 0 /* dummy to avoid compiler warning */ );
646 SvREFCNT_dec_NN(PL_warn_locale);
647 PL_warn_locale = NULL;
655 S_new_collate(pTHX_ const char *newcoll)
658 #ifndef USE_LOCALE_COLLATE
660 PERL_UNUSED_ARG(newcoll);
665 /* Called after all libc setlocale() calls affecting LC_COLLATE, to tell
666 * core Perl this and that 'newcoll' is the name of the new locale.
668 * The design of locale collation is that every locale change is given an
669 * index 'PL_collation_ix'. The first time a string particpates in an
670 * operation that requires collation while locale collation is active, it
671 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
672 * magic includes the collation index, and the transformation of the string
673 * by strxfrm(), q.v. That transformation is used when doing comparisons,
674 * instead of the string itself. If a string changes, the magic is
675 * cleared. The next time the locale changes, the index is incremented,
676 * and so we know during a comparison that the transformation is not
677 * necessarily still valid, and so is recomputed. Note that if the locale
678 * changes enough times, the index could wrap (a U32), and it is possible
679 * that a transformation would improperly be considered valid, leading to
683 if (PL_collation_name) {
685 Safefree(PL_collation_name);
686 PL_collation_name = NULL;
688 PL_collation_standard = TRUE;
689 is_standard_collation:
690 PL_collxfrm_base = 0;
691 PL_collxfrm_mult = 2;
692 PL_in_utf8_COLLATE_locale = FALSE;
693 PL_strxfrm_NUL_replacement = '\0';
694 PL_strxfrm_max_cp = 0;
698 /* If this is not the same locale as currently, set the new one up */
699 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
701 Safefree(PL_collation_name);
702 PL_collation_name = stdize_locale(savepv(newcoll));
703 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
704 if (PL_collation_standard) {
705 goto is_standard_collation;
708 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
709 PL_strxfrm_NUL_replacement = '\0';
710 PL_strxfrm_max_cp = 0;
712 /* A locale collation definition includes primary, secondary, tertiary,
713 * etc. weights for each character. To sort, the primary weights are
714 * used, and only if they compare equal, then the secondary weights are
715 * used, and only if they compare equal, then the tertiary, etc.
717 * strxfrm() works by taking the input string, say ABC, and creating an
718 * output transformed string consisting of first the primary weights,
719 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
720 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
721 * may not have weights at every level. In our example, let's say B
722 * doesn't have a tertiary weight, and A doesn't have a secondary
723 * weight. The constructed string is then going to be
724 * A¹B¹C¹ B²C² A³C³ ....
725 * This has the desired effect that strcmp() will look at the secondary
726 * or tertiary weights only if the strings compare equal at all higher
727 * priority weights. The spaces shown here, like in
729 * are not just for readability. In the general case, these must
730 * actually be bytes, which we will call here 'separator weights'; and
731 * they must be smaller than any other weight value, but since these
732 * are C strings, only the terminating one can be a NUL (some
733 * implementations may include a non-NUL separator weight just before
734 * the NUL). Implementations tend to reserve 01 for the separator
735 * weights. They are needed so that a shorter string's secondary
736 * weights won't be misconstrued as primary weights of a longer string,
737 * etc. By making them smaller than any other weight, the shorter
738 * string will sort first. (Actually, if all secondary weights are
739 * smaller than all primary ones, there is no need for a separator
740 * weight between those two levels, etc.)
742 * The length of the transformed string is roughly a linear function of
743 * the input string. It's not exactly linear because some characters
744 * don't have weights at all levels. When we call strxfrm() we have to
745 * allocate some memory to hold the transformed string. The
746 * calculations below try to find coefficients 'm' and 'b' for this
747 * locale so that m*x + b equals how much space we need, given the size
748 * of the input string in 'x'. If we calculate too small, we increase
749 * the size as needed, and call strxfrm() again, but it is better to
750 * get it right the first time to avoid wasted expensive string
751 * transformations. */
754 /* We use the string below to find how long the tranformation of it
755 * is. Almost all locales are supersets of ASCII, or at least the
756 * ASCII letters. We use all of them, half upper half lower,
757 * because if we used fewer, we might hit just the ones that are
758 * outliers in a particular locale. Most of the strings being
759 * collated will contain a preponderance of letters, and even if
760 * they are above-ASCII, they are likely to have the same number of
761 * weight levels as the ASCII ones. It turns out that digits tend
762 * to have fewer levels, and some punctuation has more, but those
763 * are relatively sparse in text, and khw believes this gives a
764 * reasonable result, but it could be changed if experience so
766 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
767 char * x_longer; /* Transformed 'longer' */
768 Size_t x_len_longer; /* Length of 'x_longer' */
770 char * x_shorter; /* We also transform a substring of 'longer' */
771 Size_t x_len_shorter;
773 /* _mem_collxfrm() is used get the transformation (though here we
774 * are interested only in its length). It is used because it has
775 * the intelligence to handle all cases, but to work, it needs some
776 * values of 'm' and 'b' to get it started. For the purposes of
777 * this calculation we use a very conservative estimate of 'm' and
778 * 'b'. This assumes a weight can be multiple bytes, enough to
779 * hold any UV on the platform, and there are 5 levels, 4 weight
780 * bytes, and a trailing NUL. */
781 PL_collxfrm_base = 5;
782 PL_collxfrm_mult = 5 * sizeof(UV);
784 /* Find out how long the transformation really is */
785 x_longer = _mem_collxfrm(longer,
789 /* We avoid converting to UTF-8 in the
790 * called function by telling it the
791 * string is in UTF-8 if the locale is a
792 * UTF-8 one. Since the string passed
793 * here is invariant under UTF-8, we can
794 * claim it's UTF-8 even though it isn't.
796 PL_in_utf8_COLLATE_locale);
799 /* Find out how long the transformation of a substring of 'longer'
800 * is. Together the lengths of these transformations are
801 * sufficient to calculate 'm' and 'b'. The substring is all of
802 * 'longer' except the first character. This minimizes the chances
803 * of being swayed by outliers */
804 x_shorter = _mem_collxfrm(longer + 1,
807 PL_in_utf8_COLLATE_locale);
810 /* If the results are nonsensical for this simple test, the whole
811 * locale definition is suspect. Mark it so that locale collation
812 * is not active at all for it. XXX Should we warn? */
813 if ( x_len_shorter == 0
815 || x_len_shorter >= x_len_longer)
817 PL_collxfrm_mult = 0;
818 PL_collxfrm_base = 0;
821 SSize_t base; /* Temporary */
823 /* We have both: m * strlen(longer) + b = x_len_longer
824 * m * strlen(shorter) + b = x_len_shorter;
825 * subtracting yields:
826 * m * (strlen(longer) - strlen(shorter))
827 * = x_len_longer - x_len_shorter
828 * But we have set things up so that 'shorter' is 1 byte smaller
829 * than 'longer'. Hence:
830 * m = x_len_longer - x_len_shorter
832 * But if something went wrong, make sure the multiplier is at
835 if (x_len_longer > x_len_shorter) {
836 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
839 PL_collxfrm_mult = 1;
844 * but in case something has gone wrong, make sure it is
846 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
851 /* Add 1 for the trailing NUL */
852 PL_collxfrm_base = base + 1;
857 if (DEBUG_L_TEST || debug_initialization) {
858 PerlIO_printf(Perl_debug_log,
859 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
861 " collate multipler=%zu, collate base=%zu\n",
863 PL_in_utf8_COLLATE_locale,
864 x_len_shorter, x_len_longer,
865 PL_collxfrm_mult, PL_collxfrm_base);
872 #endif /* USE_LOCALE_COLLATE */
879 S_win32_setlocale(pTHX_ int category, const char* locale)
881 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
882 * difference between the two unless the input locale is "", which normally
883 * means on Windows to get the machine default, which is set via the
884 * computer's "Regional and Language Options" (or its current equivalent).
885 * In POSIX, it instead means to find the locale from the user's
886 * environment. This routine changes the Windows behavior to first look in
887 * the environment, and, if anything is found, use that instead of going to
888 * the machine default. If there is no environment override, the machine
889 * default is used, by calling the real setlocale() with "".
891 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
892 * use the particular category's variable if set; otherwise to use the LANG
895 bool override_LC_ALL = FALSE;
899 if (locale && strEQ(locale, "")) {
903 locale = PerlEnv_getenv("LC_ALL");
905 if (category == LC_ALL) {
906 override_LC_ALL = TRUE;
912 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
913 if (category == categories[i]) {
914 locale = PerlEnv_getenv(category_names[i]);
919 locale = PerlEnv_getenv("LANG");
935 result = setlocale(category, locale);
936 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
937 setlocale_debug_string(category, locale, result)));
939 if (! override_LC_ALL) {
943 /* Here the input category was LC_ALL, and we have set it to what is in the
944 * LANG variable or the system default if there is no LANG. But these have
945 * lower priority than the other LC_foo variables, so override it for each
946 * one that is set. (If they are set to "", it means to use the same thing
947 * we just set LC_ALL to, so can skip) */
949 for (i = 0; i < LC_ALL_INDEX; i++) {
950 result = PerlEnv_getenv(category_names[i]);
951 if (result && strNE(result, "")) {
952 setlocale(categories[i], result);
953 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
955 setlocale_debug_string(categories[i], result, "not captured")));
959 result = setlocale(LC_ALL, NULL);
960 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
962 setlocale_debug_string(LC_ALL, NULL, result)));
970 Perl_setlocale(int category, const char * locale)
972 /* This wraps POSIX::setlocale() */
978 #ifdef USE_LOCALE_NUMERIC
980 /* A NULL locale means only query what the current one is. We
981 * have the LC_NUMERIC name saved, because we are normally switched
982 * into the C locale for it. Switch back so an LC_ALL query will yield
983 * the correct results; all other categories don't require special
985 if (locale == NULL) {
986 if (category == LC_NUMERIC) {
987 return savepv(PL_numeric_name);
992 else if (category == LC_ALL) {
993 SET_NUMERIC_UNDERLYING();
1002 /* Save retval since subsequent setlocale() calls may overwrite it. */
1003 retval = savepv(do_setlocale_r(category, locale));
1005 DEBUG_L(PerlIO_printf(Perl_debug_log,
1006 "%s:%d: %s\n", __FILE__, __LINE__,
1007 setlocale_debug_string(category, locale, retval)));
1009 /* Should never happen that a query would return an error, but be
1010 * sure and reset to C locale */
1012 SET_NUMERIC_STANDARD();
1018 /* If locale == NULL, we are just querying the state, but may have switched
1019 * to NUMERIC_UNDERLYING. Switch back before returning. */
1020 if (locale == NULL) {
1021 SET_NUMERIC_STANDARD();
1025 /* Now that have switched locales, we have to update our records to
1030 #ifdef USE_LOCALE_CTYPE
1037 #ifdef USE_LOCALE_COLLATE
1040 new_collate(retval);
1044 #ifdef USE_LOCALE_NUMERIC
1047 new_numeric(retval);
1055 /* LC_ALL updates all the things we care about. The values may not
1056 * be the same as 'retval', as the locale "" may have set things
1059 # ifdef USE_LOCALE_CTYPE
1061 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1062 new_ctype(newlocale);
1064 # endif /* USE_LOCALE_CTYPE */
1065 # ifdef USE_LOCALE_COLLATE
1067 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1068 new_collate(newlocale);
1071 # ifdef USE_LOCALE_NUMERIC
1073 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1074 new_numeric(newlocale);
1076 # endif /* USE_LOCALE_NUMERIC */
1088 PERL_STATIC_INLINE const char *
1089 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1091 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1092 * growing it if necessary */
1094 const Size_t string_size = strlen(string) + offset + 1;
1096 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1098 if (*buf_size == 0) {
1099 Newx(*buf, string_size, char);
1100 *buf_size = string_size;
1102 else if (string_size > *buf_size) {
1103 Renew(*buf, string_size, char);
1104 *buf_size = string_size;
1107 Copy(string, *buf + offset, string_size - offset, char);
1113 =head1 Locale-related functions and macros
1115 =for apidoc Perl_langinfo
1117 This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
1118 taking the same C<item> parameter values, and returning the same information.
1119 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1120 of Perl's locale handling from your code, and can be used on systems that lack
1121 a native C<nl_langinfo>.
1129 It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1130 without you having to write extra code. The reason for the extra code would be
1131 because these are from the C<LC_NUMERIC> locale category, which is normally
1132 kept set to the C locale by Perl, no matter what the underlying locale is
1133 supposed to be, and so to get the expected results, you have to temporarily
1134 toggle into the underlying locale, and later toggle back. (You could use
1135 plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1136 but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1137 keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1138 expecting the radix (decimal point) character to be a dot.)
1142 Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1143 makes your code more portable. Of the fifty-some possible items specified by
1144 the POSIX 2008 standard,
1145 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1146 only two are completely unimplemented. It uses various techniques to recover
1147 the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1148 both of which are specified in C89, so should be always be available. Later
1149 C<strftime()> versions have additional capabilities; C<""> is returned for
1150 those not available on your system.
1152 The details for those items which may differ from what this emulation returns
1153 and what a native C<nl_langinfo()> would return are:
1161 Unimplemented, so returns C<"">.
1167 Only the values for English are returned. Earlier POSIX standards also
1168 specified C<YESSTR> and C<NOSTR>, but these have been removed from POSIX 2008,
1169 and aren't supported by C<Perl_langinfo>.
1173 Always evaluates to C<%x>, the locale's appropriate date representation.
1177 Always evaluates to C<%X>, the locale's appropriate time representation.
1181 Always evaluates to C<%c>, the locale's appropriate date and time
1186 The return may be incorrect for those rare locales where the currency symbol
1187 replaces the radix character.
1188 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1189 to work differently.
1193 Currently this gives the same results as Linux does.
1194 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1195 to work differently.
1201 =item C<ERA_D_T_FMT>
1205 These are derived by using C<strftime()>, and not all versions of that function
1206 know about them. C<""> is returned for these on such systems.
1210 When using C<Perl_langinfo> on systems that don't have a native
1211 C<nl_langinfo()>, you must
1213 #include "perl_langinfo.h"
1215 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1216 C<#include> with this one. (Doing it this way keeps out the symbols that plain
1217 C<langinfo.h> imports into the namespace for code that doesn't need it.)
1219 You also should not use the bare C<langinfo.h> item names, but should preface
1220 them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1221 The C<PERL_I<foo>> versions will also work for this function on systems that do
1222 have a native C<nl_langinfo>.
1226 It is thread-friendly, returning its result in a buffer that won't be
1227 overwritten by another thread, so you don't have to code for that possibility.
1228 The buffer can be overwritten by the next call to C<nl_langinfo> or
1229 C<Perl_langinfo> in the same thread.
1233 ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
1234 *>>, but you are (only by documentation) forbidden to write into the buffer.
1235 By declaring this C<const>, the compiler enforces this restriction. The extra
1236 C<const> is why this isn't an unequivocal drop-in replacement for
1241 The original impetus for C<Perl_langinfo()> was so that code that needs to
1242 find out the current currency symbol, floating point radix character, or digit
1243 grouping separator can use, on all systems, the simpler and more
1244 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1245 pain to make thread-friendly. For other fields returned by C<localeconv>, it
1246 is better to use the methods given in L<perlcall> to call
1247 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1254 #ifdef HAS_NL_LANGINFO
1255 Perl_langinfo(const nl_item item)
1257 Perl_langinfo(const int item)
1260 return my_nl_langinfo(item, TRUE);
1264 #ifdef HAS_NL_LANGINFO
1265 S_my_nl_langinfo(const nl_item item, bool toggle)
1267 S_my_nl_langinfo(const int item, bool toggle)
1272 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1273 #if ! defined(HAS_POSIX_2008_LOCALE)
1275 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
1276 * for those items dependent on it. This must be copied to a buffer before
1277 * switching back, as some systems destroy the buffer when setlocale() is
1283 if (item == PERL_RADIXCHAR || item == PERL_THOUSEP) {
1284 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1291 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1294 do_setlocale_c(LC_NUMERIC, "C");
1299 return PL_langinfo_buf;
1301 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1303 bool do_free = FALSE;
1304 locale_t cur = uselocale((locale_t) 0);
1306 if (cur == LC_GLOBAL_LOCALE) {
1307 cur = duplocale(LC_GLOBAL_LOCALE);
1312 && (item == PERL_RADIXCHAR || item == PERL_THOUSEP))
1314 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1318 save_to_buffer(nl_langinfo_l(item, cur),
1319 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1324 return PL_langinfo_buf;
1327 #else /* Below, emulate nl_langinfo as best we can */
1328 # ifdef HAS_LOCALECONV
1330 const struct lconv* lc;
1333 # ifdef HAS_STRFTIME
1336 bool return_format = FALSE; /* Return the %format, not the value */
1337 const char * format;
1341 /* We copy the results to a per-thread buffer, even if not multi-threaded.
1342 * This is in part to simplify this code, and partly because we need a
1343 * buffer anyway for strftime(), and partly because a call of localeconv()
1344 * could otherwise wipe out the buffer, and the programmer would not be
1345 * expecting this, as this is a nl_langinfo() substitute after all, so s/he
1346 * might be thinking their localeconv() is safe until another localeconv()
1351 const char * retval;
1353 /* These 2 are unimplemented */
1355 case PERL_ERA: /* For use with strftime() %E modifier */
1360 /* We use only an English set, since we don't know any more */
1361 case PERL_YESEXPR: return "^[+1yY]";
1362 case PERL_NOEXPR: return "^[-0nN]";
1364 # ifdef HAS_LOCALECONV
1371 if (! lc || ! lc->currency_symbol || strEQ("", lc->currency_symbol))
1377 /* Leave the first spot empty to be filled in below */
1378 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1379 &PL_langinfo_bufsize, 1);
1380 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1381 { /* khw couldn't figure out how the localedef specifications
1382 would show that the $ should replace the radix; this is
1383 just a guess as to how it might work.*/
1384 *PL_langinfo_buf = '.';
1386 else if (lc->p_cs_precedes) {
1387 *PL_langinfo_buf = '-';
1390 *PL_langinfo_buf = '+';
1396 case PERL_RADIXCHAR:
1402 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1410 retval = (item == PERL_RADIXCHAR)
1412 : lc->thousands_sep;
1418 save_to_buffer(retval, &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1421 do_setlocale_c(LC_NUMERIC, "C");
1429 # ifdef HAS_STRFTIME
1431 /* These are defined by C89, so we assume that strftime supports them,
1432 * and so are returned unconditionally; they may not be what the locale
1433 * actually says, but should give good enough results for someone using
1434 * them as formats (as opposed to trying to parse them to figure out
1435 * what the locale says). The other format items are actually tested to
1436 * verify they work on the platform */
1437 case PERL_D_FMT: return "%x";
1438 case PERL_T_FMT: return "%X";
1439 case PERL_D_T_FMT: return "%c";
1441 /* These formats are only available in later strfmtime's */
1442 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1443 case PERL_T_FMT_AMPM:
1445 /* The rest can be gotten from most versions of strftime(). */
1446 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1447 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1449 case PERL_ALT_DIGITS:
1450 case PERL_AM_STR: case PERL_PM_STR:
1451 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1452 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1453 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1454 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1455 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1456 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1457 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1458 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1459 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11: case PERL_MON_12:
1463 init_tm(&tm); /* Precaution against core dumps */
1467 tm.tm_year = 2017 - 1900;
1473 Perl_croak(aTHX_ "panic: %s: %d: switch case: %d problem",
1474 __FILE__, __LINE__, item);
1475 NOT_REACHED; /* NOTREACHED */
1477 case PERL_PM_STR: tm.tm_hour = 18;
1482 case PERL_ABDAY_7: tm.tm_wday++;
1483 case PERL_ABDAY_6: tm.tm_wday++;
1484 case PERL_ABDAY_5: tm.tm_wday++;
1485 case PERL_ABDAY_4: tm.tm_wday++;
1486 case PERL_ABDAY_3: tm.tm_wday++;
1487 case PERL_ABDAY_2: tm.tm_wday++;
1492 case PERL_DAY_7: tm.tm_wday++;
1493 case PERL_DAY_6: tm.tm_wday++;
1494 case PERL_DAY_5: tm.tm_wday++;
1495 case PERL_DAY_4: tm.tm_wday++;
1496 case PERL_DAY_3: tm.tm_wday++;
1497 case PERL_DAY_2: tm.tm_wday++;
1502 case PERL_ABMON_12: tm.tm_mon++;
1503 case PERL_ABMON_11: tm.tm_mon++;
1504 case PERL_ABMON_10: tm.tm_mon++;
1505 case PERL_ABMON_9: tm.tm_mon++;
1506 case PERL_ABMON_8: tm.tm_mon++;
1507 case PERL_ABMON_7: tm.tm_mon++;
1508 case PERL_ABMON_6: tm.tm_mon++;
1509 case PERL_ABMON_5: tm.tm_mon++;
1510 case PERL_ABMON_4: tm.tm_mon++;
1511 case PERL_ABMON_3: tm.tm_mon++;
1512 case PERL_ABMON_2: tm.tm_mon++;
1517 case PERL_MON_12: tm.tm_mon++;
1518 case PERL_MON_11: tm.tm_mon++;
1519 case PERL_MON_10: tm.tm_mon++;
1520 case PERL_MON_9: tm.tm_mon++;
1521 case PERL_MON_8: tm.tm_mon++;
1522 case PERL_MON_7: tm.tm_mon++;
1523 case PERL_MON_6: tm.tm_mon++;
1524 case PERL_MON_5: tm.tm_mon++;
1525 case PERL_MON_4: tm.tm_mon++;
1526 case PERL_MON_3: tm.tm_mon++;
1527 case PERL_MON_2: tm.tm_mon++;
1532 case PERL_T_FMT_AMPM:
1534 return_format = TRUE;
1537 case PERL_ERA_D_FMT:
1539 return_format = TRUE;
1542 case PERL_ERA_T_FMT:
1544 return_format = TRUE;
1547 case PERL_ERA_D_T_FMT:
1549 return_format = TRUE;
1552 case PERL_ALT_DIGITS:
1554 format = "%Ow"; /* Find the alternate digit for 0 */
1558 /* We can't use my_strftime() because it doesn't look at tm_wday */
1559 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1562 /* A zero return means one of:
1563 * a) there wasn't enough space in PL_langinfo_buf
1564 * b) the format, like a plain %p, returns empty
1565 * c) it was an illegal format, though some implementations of
1566 * strftime will just return the illegal format as a plain
1567 * character sequence.
1569 * To quickly test for case 'b)', try again but precede the
1570 * format with a plain character. If that result is still
1571 * empty, the problem is either 'a)' or 'c)' */
1573 Size_t format_size = strlen(format) + 1;
1574 Size_t mod_size = format_size + 1;
1578 Newx(mod_format, mod_size, char);
1579 Newx(temp_result, PL_langinfo_bufsize, char);
1581 my_strlcpy(mod_format + 1, format, mod_size);
1582 len = strftime(temp_result,
1583 PL_langinfo_bufsize,
1585 Safefree(mod_format);
1586 Safefree(temp_result);
1588 /* If 'len' is non-zero, it means that we had a case like %p
1589 * which means the current locale doesn't use a.m. or p.m., and
1593 /* Here, still didn't work. If we get well beyond a
1594 * reasonable size, bail out to prevent an infinite loop. */
1596 if (PL_langinfo_bufsize > 100 * format_size) {
1597 *PL_langinfo_buf = '\0';
1599 else { /* Double the buffer size to retry; Add 1 in case
1600 original was 0, so we aren't stuck at 0. */
1601 PL_langinfo_bufsize *= 2;
1602 PL_langinfo_bufsize++;
1603 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1611 /* Here, we got a result.
1613 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1614 * alternate format for wday 0. If the value is the same as the
1615 * normal 0, there isn't an alternate, so clear the buffer. */
1616 if ( item == PERL_ALT_DIGITS
1617 && strEQ(PL_langinfo_buf, "0"))
1619 *PL_langinfo_buf = '\0';
1622 /* ALT_DIGITS is problematic. Experiments on it showed that
1623 * strftime() did not always work properly when going from alt-9 to
1624 * alt-10. Only a few locales have this item defined, and in all
1625 * of them on Linux that khw was able to find, nl_langinfo() merely
1626 * returned the alt-0 character, possibly doubled. Most Unicode
1627 * digits are in blocks of 10 consecutive code points, so that is
1628 * sufficient information for those scripts, as we can infer alt-1,
1629 * alt-2, .... But for a Japanese locale, a CJK ideographic 0 is
1630 * returned, and the CJK digits are not in code point order, so you
1631 * can't really infer anything. The localedef for this locale did
1632 * specify the succeeding digits, so that strftime() works properly
1633 * on them, without needing to infer anything. But the
1634 * nl_langinfo() return did not give sufficient information for the
1635 * caller to understand what's going on. So until there is
1636 * evidence that it should work differently, this returns the alt-0
1637 * string for ALT_DIGITS.
1639 * wday was chosen because its range is all a single digit. Things
1640 * like tm_sec have two digits as the minimum: '00' */
1644 /* If to return the format, not the value, overwrite the buffer
1645 * with it. But some strftime()s will keep the original format if
1646 * illegal, so change those to "" */
1647 if (return_format) {
1648 if (strEQ(PL_langinfo_buf, format)) {
1649 *PL_langinfo_buf = '\0';
1652 save_to_buffer(format, &PL_langinfo_buf,
1653 &PL_langinfo_bufsize, 0);
1663 return PL_langinfo_buf;
1670 * Initialize locale awareness.
1673 Perl_init_i18nl10n(pTHX_ int printwarn)
1677 * 0 if not to output warning when setup locale is bad
1678 * 1 if to output warning based on value of PERL_BADLANG
1679 * >1 if to output regardless of PERL_BADLANG
1682 * 1 = set ok or not applicable,
1683 * 0 = fallback to a locale of lower priority
1684 * -1 = fallback to all locales failed, not even to the C locale
1686 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1687 * set, debugging information is output.
1689 * This looks more complicated than it is, mainly due to the #ifdefs.
1691 * We try to set LC_ALL to the value determined by the environment. If
1692 * there is no LC_ALL on this platform, we try the individual categories we
1693 * know about. If this works, we are done.
1695 * But if it doesn't work, we have to do something else. We search the
1696 * environment variables ourselves instead of relying on the system to do
1697 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1698 * think there is one), and the ultimate fallback "C". This is all done in
1699 * the same loop as above to avoid duplicating code, but it makes things
1700 * more complex. The 'trial_locales' array is initialized with just one
1701 * element; it causes the behavior described in the paragraph above this to
1702 * happen. If that fails, we add elements to 'trial_locales', and do extra
1703 * loop iterations to cause the behavior described in this paragraph.
1705 * On Ultrix, the locale MUST come from the environment, so there is
1706 * preliminary code to set it. I (khw) am not sure that it is necessary,
1707 * and that this couldn't be folded into the loop, but barring any real
1708 * platforms to test on, it's staying as-is
1710 * A slight complication is that in embedded Perls, the locale may already
1711 * be set-up, and we don't want to get it from the normal environment
1712 * variables. This is handled by having a special environment variable
1713 * indicate we're in this situation. We simply set setlocale's 2nd
1714 * parameter to be a NULL instead of "". That indicates to setlocale that
1715 * it is not to change anything, but to return the current value,
1716 * effectively initializing perl's db to what the locale already is.
1718 * We play the same trick with NULL if a LC_ALL succeeds. We call
1719 * setlocale() on the individual categores with NULL to get their existing
1720 * values for our db, instead of trying to change them.
1727 PERL_UNUSED_ARG(printwarn);
1729 #else /* USE_LOCALE */
1732 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
1736 /* NULL uses the existing already set up locale */
1737 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1740 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1741 unsigned int trial_locales_count;
1742 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1743 const char * const lang = savepv(PerlEnv_getenv("LANG"));
1744 bool setlocale_failure = FALSE;
1747 /* A later getenv() could zap this, so only use here */
1748 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1750 const bool locwarn = (printwarn > 1
1752 && ( ! bad_lang_use_once
1754 /* disallow with "" or "0" */
1756 && strNE("0", bad_lang_use_once)))));
1758 char * sl_result[NOMINAL_LC_ALL_INDEX + 1]; /* setlocale() return vals;
1759 not copied so must be
1760 looked at immediately */
1761 char * curlocales[NOMINAL_LC_ALL_INDEX + 1]; /* current locale for given
1762 category; should have been
1763 copied so aren't volatile
1765 char * locale_param;
1769 /* In some systems you can find out the system default locale
1770 * and use that as the fallback locale. */
1771 # define SYSTEM_DEFAULT_LOCALE
1773 # ifdef SYSTEM_DEFAULT_LOCALE
1775 const char *system_default_locale = NULL;
1780 # define DEBUG_LOCALE_INIT(a,b,c)
1783 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
1785 # define DEBUG_LOCALE_INIT(category, locale, result) \
1787 if (debug_initialization) { \
1788 PerlIO_printf(Perl_debug_log, \
1790 __FILE__, __LINE__, \
1791 setlocale_debug_string(category, \
1797 /* Make sure the parallel arrays are properly set up */
1798 # ifdef USE_LOCALE_NUMERIC
1799 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
1800 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
1802 # ifdef USE_LOCALE_CTYPE
1803 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
1804 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
1806 # ifdef USE_LOCALE_COLLATE
1807 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
1808 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
1810 # ifdef USE_LOCALE_TIME
1811 assert(categories[LC_TIME_INDEX] == LC_TIME);
1812 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
1814 # ifdef USE_LOCALE_MESSAGES
1815 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
1816 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
1818 # ifdef USE_LOCALE_MONETARY
1819 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
1820 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
1823 assert(categories[LC_ALL_INDEX] == LC_ALL);
1824 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
1825 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
1827 # endif /* DEBUGGING */
1828 # ifndef LOCALE_ENVIRON_REQUIRED
1830 PERL_UNUSED_VAR(done);
1831 PERL_UNUSED_VAR(locale_param);
1836 * Ultrix setlocale(..., "") fails if there are no environment
1837 * variables from which to get a locale name.
1843 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
1844 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
1845 if (sl_result[LC_ALL_INDEX])
1848 setlocale_failure = TRUE;
1850 if (! setlocale_failure) {
1851 for (i = 0; i < LC_ALL_INDEX; i++) {
1852 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
1855 sl_result[i] = do_setlocale_r(categories[i], locale_param);
1856 if (! sl_result[i]) {
1857 setlocale_failure = TRUE;
1859 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
1863 # endif /* LC_ALL */
1864 # endif /* LOCALE_ENVIRON_REQUIRED */
1866 /* We try each locale in the list until we get one that works, or exhaust
1867 * the list. Normally the loop is executed just once. But if setting the
1868 * locale fails, inside the loop we add fallback trials to the array and so
1869 * will execute the loop multiple times */
1870 trial_locales[0] = setlocale_init;
1871 trial_locales_count = 1;
1873 for (i= 0; i < trial_locales_count; i++) {
1874 const char * trial_locale = trial_locales[i];
1878 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
1879 * when i==0, but I (khw) don't think that behavior makes much
1881 setlocale_failure = FALSE;
1883 # ifdef SYSTEM_DEFAULT_LOCALE
1886 /* On Windows machines, an entry of "" after the 0th means to use
1887 * the system default locale, which we now proceed to get. */
1888 if (strEQ(trial_locale, "")) {
1891 /* Note that this may change the locale, but we are going to do
1892 * that anyway just below */
1893 system_default_locale = do_setlocale_c(LC_ALL, "");
1894 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
1896 /* Skip if invalid or if it's already on the list of locales to
1898 if (! system_default_locale) {
1899 goto next_iteration;
1901 for (j = 0; j < trial_locales_count; j++) {
1902 if (strEQ(system_default_locale, trial_locales[j])) {
1903 goto next_iteration;
1907 trial_locale = system_default_locale;
1910 # endif /* SYSTEM_DEFAULT_LOCALE */
1915 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
1916 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
1917 if (! sl_result[LC_ALL_INDEX]) {
1918 setlocale_failure = TRUE;
1921 /* Since LC_ALL succeeded, it should have changed all the other
1922 * categories it can to its value; so we massage things so that the
1923 * setlocales below just return their category's current values.
1924 * This adequately handles the case in NetBSD where LC_COLLATE may
1925 * not be defined for a locale, and setting it individually will
1926 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
1927 * the POSIX locale. */
1928 trial_locale = NULL;
1931 # endif /* LC_ALL */
1933 if (! setlocale_failure) {
1935 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
1937 = savepv(do_setlocale_r(categories[j], trial_locale));
1938 if (! curlocales[j]) {
1939 setlocale_failure = TRUE;
1941 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
1944 if (! setlocale_failure) { /* All succeeded */
1945 break; /* Exit trial_locales loop */
1949 /* Here, something failed; will need to try a fallback. */
1955 if (locwarn) { /* Output failure info only on the first one */
1959 PerlIO_printf(Perl_error_log,
1960 "perl: warning: Setting locale failed.\n");
1962 # else /* !LC_ALL */
1964 PerlIO_printf(Perl_error_log,
1965 "perl: warning: Setting locale failed for the categories:\n\t");
1967 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
1968 if (! curlocales[j]) {
1969 PerlIO_printf(Perl_error_log, category_names[j]);
1972 Safefree(curlocales[j]);
1976 PerlIO_printf(Perl_error_log, "and possibly others\n");
1978 # endif /* LC_ALL */
1980 PerlIO_printf(Perl_error_log,
1981 "perl: warning: Please check that your locale settings:\n");
1985 PerlIO_printf(Perl_error_log,
1986 "\tLANGUAGE = %c%s%c,\n",
1987 language ? '"' : '(',
1988 language ? language : "unset",
1989 language ? '"' : ')');
1992 PerlIO_printf(Perl_error_log,
1993 "\tLC_ALL = %c%s%c,\n",
1995 lc_all ? lc_all : "unset",
1996 lc_all ? '"' : ')');
1998 # if defined(USE_ENVIRON_ARRAY)
2003 /* Look through the environment for any variables of the
2004 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2005 * already handled above. These are assumed to be locale
2006 * settings. Output them and their values. */
2007 for (e = environ; *e; e++) {
2008 const STRLEN prefix_len = sizeof("LC_") - 1;
2011 if ( strBEGINs(*e, "LC_")
2012 && ! strBEGINs(*e, "LC_ALL=")
2013 && (uppers_len = strspn(*e + prefix_len,
2014 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2015 && ((*e)[prefix_len + uppers_len] == '='))
2017 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
2018 (int) (prefix_len + uppers_len), *e,
2019 *e + prefix_len + uppers_len + 1);
2026 PerlIO_printf(Perl_error_log,
2027 "\t(possibly more locale environment variables)\n");
2031 PerlIO_printf(Perl_error_log,
2032 "\tLANG = %c%s%c\n",
2034 lang ? lang : "unset",
2037 PerlIO_printf(Perl_error_log,
2038 " are supported and installed on your system.\n");
2041 /* Calculate what fallback locales to try. We have avoided this
2042 * until we have to, because failure is quite unlikely. This will
2043 * usually change the upper bound of the loop we are in.
2045 * Since the system's default way of setting the locale has not
2046 * found one that works, We use Perl's defined ordering: LC_ALL,
2047 * LANG, and the C locale. We don't try the same locale twice, so
2048 * don't add to the list if already there. (On POSIX systems, the
2049 * LC_ALL element will likely be a repeat of the 0th element "",
2050 * but there's no harm done by doing it explicitly.
2052 * Note that this tries the LC_ALL environment variable even on
2053 * systems which have no LC_ALL locale setting. This may or may
2054 * not have been originally intentional, but there's no real need
2055 * to change the behavior. */
2057 for (j = 0; j < trial_locales_count; j++) {
2058 if (strEQ(lc_all, trial_locales[j])) {
2062 trial_locales[trial_locales_count++] = lc_all;
2067 for (j = 0; j < trial_locales_count; j++) {
2068 if (strEQ(lang, trial_locales[j])) {
2072 trial_locales[trial_locales_count++] = lang;
2076 # if defined(WIN32) && defined(LC_ALL)
2078 /* For Windows, we also try the system default locale before "C".
2079 * (If there exists a Windows without LC_ALL we skip this because
2080 * it gets too complicated. For those, the "C" is the next
2081 * fallback possibility). The "" is the same as the 0th element of
2082 * the array, but the code at the loop above knows to treat it
2083 * differently when not the 0th */
2084 trial_locales[trial_locales_count++] = "";
2088 for (j = 0; j < trial_locales_count; j++) {
2089 if (strEQ("C", trial_locales[j])) {
2093 trial_locales[trial_locales_count++] = "C";
2096 } /* end of first time through the loop */
2104 } /* end of looping through the trial locales */
2106 if (ok < 1) { /* If we tried to fallback */
2108 if (! setlocale_failure) { /* fallback succeeded */
2109 msg = "Falling back to";
2111 else { /* fallback failed */
2114 /* We dropped off the end of the loop, so have to decrement i to
2115 * get back to the value the last time through */
2119 msg = "Failed to fall back to";
2121 /* To continue, we should use whatever values we've got */
2123 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2124 Safefree(curlocales[j]);
2125 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2126 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2131 const char * description;
2132 const char * name = "";
2133 if (strEQ(trial_locales[i], "C")) {
2134 description = "the standard locale";
2138 # ifdef SYSTEM_DEFAULT_LOCALE
2140 else if (strEQ(trial_locales[i], "")) {
2141 description = "the system default locale";
2142 if (system_default_locale) {
2143 name = system_default_locale;
2147 # endif /* SYSTEM_DEFAULT_LOCALE */
2150 description = "a fallback locale";
2151 name = trial_locales[i];
2153 if (name && strNE(name, "")) {
2154 PerlIO_printf(Perl_error_log,
2155 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2158 PerlIO_printf(Perl_error_log,
2159 "perl: warning: %s %s.\n", msg, description);
2162 } /* End of tried to fallback */
2164 /* Done with finding the locales; update our records */
2166 # ifdef USE_LOCALE_CTYPE
2168 new_ctype(curlocales[LC_CTYPE_INDEX]);
2171 # ifdef USE_LOCALE_COLLATE
2173 new_collate(curlocales[LC_COLLATE_INDEX]);
2176 # ifdef USE_LOCALE_NUMERIC
2178 new_numeric(curlocales[LC_NUMERIC_INDEX]);
2183 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2184 Safefree(curlocales[i]);
2187 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2189 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
2190 * locale is UTF-8. If PL_utf8locale and PL_unicode (set by -C or by
2191 * $ENV{PERL_UNICODE}) are true, perl.c:S_parse_body() will turn on the
2192 * PerlIO :utf8 layer on STDIN, STDOUT, STDERR, _and_ the default open
2194 PL_utf8locale = _is_cur_LC_category_utf8(LC_CTYPE);
2196 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
2197 This is an alternative to using the -C command line switch
2198 (the -C if present will override this). */
2200 const char *p = PerlEnv_getenv("PERL_UNICODE");
2201 PL_unicode = p ? parse_unicode_opts(&p) : 0;
2202 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2216 #endif /* USE_LOCALE */
2219 /* So won't continue to output stuff */
2220 DEBUG_INITIALIZATION_set(FALSE);
2227 #ifdef USE_LOCALE_COLLATE
2230 Perl__mem_collxfrm(pTHX_ const char *input_string,
2231 STRLEN len, /* Length of 'input_string' */
2232 STRLEN *xlen, /* Set to length of returned string
2233 (not including the collation index
2235 bool utf8 /* Is the input in UTF-8? */
2239 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2240 * differences. First, it handles embedded NULs. Second, it allocates a bit
2241 * more memory than needed for the transformed data itself. The real
2242 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
2243 * the length of that, and doesn't include the collation index size.
2244 * Please see sv_collxfrm() to see how this is used. */
2246 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2248 char * s = (char *) input_string;
2249 STRLEN s_strlen = strlen(input_string);
2251 STRLEN xAlloc; /* xalloc is a reserved word in VC */
2252 STRLEN length_in_chars;
2253 bool first_time = TRUE; /* Cleared after first loop iteration */
2255 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2257 /* Must be NUL-terminated */
2258 assert(*(input_string + len) == '\0');
2260 /* If this locale has defective collation, skip */
2261 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
2262 DEBUG_L(PerlIO_printf(Perl_debug_log,
2263 "_mem_collxfrm: locale's collation is defective\n"));
2267 /* Replace any embedded NULs with the control that sorts before any others.
2268 * This will give as good as possible results on strings that don't
2269 * otherwise contain that character, but otherwise there may be
2270 * less-than-perfect results with that character and NUL. This is
2271 * unavoidable unless we replace strxfrm with our own implementation. */
2272 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2276 STRLEN sans_nuls_len;
2277 int try_non_controls;
2278 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2279 making sure 2nd byte is NUL.
2281 STRLEN this_replacement_len;
2283 /* If we don't know what non-NUL control character sorts lowest for
2284 * this locale, find it */
2285 if (PL_strxfrm_NUL_replacement == '\0') {
2287 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
2288 includes the collation index
2291 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
2293 /* Unlikely, but it may be that no control will work to replace
2294 * NUL, in which case we instead look for any character. Controls
2295 * are preferred because collation order is, in general, context
2296 * sensitive, with adjoining characters affecting the order, and
2297 * controls are less likely to have such interactions, allowing the
2298 * NUL-replacement to stand on its own. (Another way to look at it
2299 * is to imagine what would happen if the NUL were replaced by a
2300 * combining character; it wouldn't work out all that well.) */
2301 for (try_non_controls = 0;
2302 try_non_controls < 2;
2305 /* Look through all legal code points (NUL isn't) */
2306 for (j = 1; j < 256; j++) {
2307 char * x; /* j's xfrm plus collation index */
2308 STRLEN x_len; /* length of 'x' */
2309 STRLEN trial_len = 1;
2310 char cur_source[] = { '\0', '\0' };
2312 /* Skip non-controls the first time through the loop. The
2313 * controls in a UTF-8 locale are the L1 ones */
2314 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2321 /* Create a 1-char string of the current code point */
2322 cur_source[0] = (char) j;
2324 /* Then transform it */
2325 x = _mem_collxfrm(cur_source, trial_len, &x_len,
2326 0 /* The string is not in UTF-8 */);
2328 /* Ignore any character that didn't successfully transform.
2334 /* If this character's transformation is lower than
2335 * the current lowest, this one becomes the lowest */
2336 if ( cur_min_x == NULL
2337 || strLT(x + COLLXFRM_HDR_LEN,
2338 cur_min_x + COLLXFRM_HDR_LEN))
2340 PL_strxfrm_NUL_replacement = j;
2346 } /* end of loop through all 255 characters */
2348 /* Stop looking if found */
2353 /* Unlikely, but possible, if there aren't any controls that
2354 * work in the locale, repeat the loop, looking for any
2355 * character that works */
2356 DEBUG_L(PerlIO_printf(Perl_debug_log,
2357 "_mem_collxfrm: No control worked. Trying non-controls\n"));
2358 } /* End of loop to try first the controls, then any char */
2361 DEBUG_L(PerlIO_printf(Perl_debug_log,
2362 "_mem_collxfrm: Couldn't find any character to replace"
2363 " embedded NULs in locale %s with", PL_collation_name));
2367 DEBUG_L(PerlIO_printf(Perl_debug_log,
2368 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
2369 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
2371 Safefree(cur_min_x);
2372 } /* End of determining the character that is to replace NULs */
2374 /* If the replacement is variant under UTF-8, it must match the
2375 * UTF8-ness as the original */
2376 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2377 this_replacement_char[0] =
2378 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2379 this_replacement_char[1] =
2380 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
2381 this_replacement_len = 2;
2384 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
2385 /* this_replacement_char[1] = '\0' was done at initialization */
2386 this_replacement_len = 1;
2389 /* The worst case length for the replaced string would be if every
2390 * character in it is NUL. Multiply that by the length of each
2391 * replacement, and allow for a trailing NUL */
2392 sans_nuls_len = (len * this_replacement_len) + 1;
2393 Newx(sans_nuls, sans_nuls_len, char);
2396 /* Replace each NUL with the lowest collating control. Loop until have
2397 * exhausted all the NULs */
2398 while (s + s_strlen < e) {
2399 my_strlcat(sans_nuls, s, sans_nuls_len);
2401 /* Do the actual replacement */
2402 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
2404 /* Move past the input NUL */
2406 s_strlen = strlen(s);
2409 /* And add anything that trails the final NUL */
2410 my_strlcat(sans_nuls, s, sans_nuls_len);
2412 /* Switch so below we transform this modified string */
2415 } /* End of replacing NULs */
2417 /* Make sure the UTF8ness of the string and locale match */
2418 if (utf8 != PL_in_utf8_COLLATE_locale) {
2419 const char * const t = s; /* Temporary so we can later find where the
2422 /* Here they don't match. Change the string's to be what the locale is
2425 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2426 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2429 else { /* locale is not UTF-8; but input is; downgrade the input */
2431 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2433 /* If the downgrade was successful we are done, but if the input
2434 * contains things that require UTF-8 to represent, have to do
2435 * damage control ... */
2436 if (UNLIKELY(utf8)) {
2438 /* What we do is construct a non-UTF-8 string with
2439 * 1) the characters representable by a single byte converted
2440 * to be so (if necessary);
2441 * 2) and the rest converted to collate the same as the
2442 * highest collating representable character. That makes
2443 * them collate at the end. This is similar to how we
2444 * handle embedded NULs, but we use the highest collating
2445 * code point instead of the smallest. Like the NUL case,
2446 * this isn't perfect, but is the best we can reasonably
2447 * do. Every above-255 code point will sort the same as
2448 * the highest-sorting 0-255 code point. If that code
2449 * point can combine in a sequence with some other code
2450 * points for weight calculations, us changing something to
2451 * be it can adversely affect the results. But in most
2452 * cases, it should work reasonably. And note that this is
2453 * really an illegal situation: using code points above 255
2454 * on a locale where only 0-255 are valid. If two strings
2455 * sort entirely equal, then the sort order for the
2456 * above-255 code points will be in code point order. */
2460 /* If we haven't calculated the code point with the maximum
2461 * collating order for this locale, do so now */
2462 if (! PL_strxfrm_max_cp) {
2465 /* The current transformed string that collates the
2466 * highest (except it also includes the prefixed collation
2468 char * cur_max_x = NULL;
2470 /* Look through all legal code points (NUL isn't) */
2471 for (j = 1; j < 256; j++) {
2474 char cur_source[] = { '\0', '\0' };
2476 /* Create a 1-char string of the current code point */
2477 cur_source[0] = (char) j;
2479 /* Then transform it */
2480 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2482 /* If something went wrong (which it shouldn't), just
2483 * ignore this code point */
2488 /* If this character's transformation is higher than
2489 * the current highest, this one becomes the highest */
2490 if ( cur_max_x == NULL
2491 || strGT(x + COLLXFRM_HDR_LEN,
2492 cur_max_x + COLLXFRM_HDR_LEN))
2494 PL_strxfrm_max_cp = j;
2503 DEBUG_L(PerlIO_printf(Perl_debug_log,
2504 "_mem_collxfrm: Couldn't find any character to"
2505 " replace above-Latin1 chars in locale %s with",
2506 PL_collation_name));
2510 DEBUG_L(PerlIO_printf(Perl_debug_log,
2511 "_mem_collxfrm: highest 1-byte collating character"
2512 " in locale %s is 0x%02X\n",
2514 PL_strxfrm_max_cp));
2516 Safefree(cur_max_x);
2519 /* Here we know which legal code point collates the highest.
2520 * We are ready to construct the non-UTF-8 string. The length
2521 * will be at least 1 byte smaller than the input string
2522 * (because we changed at least one 2-byte character into a
2523 * single byte), but that is eaten up by the trailing NUL */
2529 char * e = (char *) t + len;
2531 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2533 if (UTF8_IS_INVARIANT(cur_char)) {
2536 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
2537 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2539 else { /* Replace illegal cp with highest collating
2541 s[d++] = PL_strxfrm_max_cp;
2545 Renew(s, d, char); /* Free up unused space */
2550 /* Here, we have constructed a modified version of the input. It could
2551 * be that we already had a modified copy before we did this version.
2552 * If so, that copy is no longer needed */
2553 if (t != input_string) {
2558 length_in_chars = (utf8)
2559 ? utf8_length((U8 *) s, (U8 *) s + len)
2562 /* The first element in the output is the collation id, used by
2563 * sv_collxfrm(); then comes the space for the transformed string. The
2564 * equation should give us a good estimate as to how much is needed */
2565 xAlloc = COLLXFRM_HDR_LEN
2567 + (PL_collxfrm_mult * length_in_chars);
2568 Newx(xbuf, xAlloc, char);
2569 if (UNLIKELY(! xbuf)) {
2570 DEBUG_L(PerlIO_printf(Perl_debug_log,
2571 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
2575 /* Store the collation id */
2576 *(U32*)xbuf = PL_collation_ix;
2578 /* Then the transformation of the input. We loop until successful, or we
2582 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
2584 /* If the transformed string occupies less space than we told strxfrm()
2585 * was available, it means it successfully transformed the whole
2587 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
2589 /* Some systems include a trailing NUL in the returned length.
2590 * Ignore it, using a loop in case multiple trailing NULs are
2593 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2598 /* If the first try didn't get it, it means our prediction was low.
2599 * Modify the coefficients so that we predict a larger value in any
2600 * future transformations */
2602 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2603 STRLEN computed_guess = PL_collxfrm_base
2604 + (PL_collxfrm_mult * length_in_chars);
2606 /* On zero-length input, just keep current slope instead of
2608 const STRLEN new_m = (length_in_chars != 0)
2609 ? needed / length_in_chars
2612 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2613 "%s: %d: initial size of %zu bytes for a length "
2614 "%zu string was insufficient, %zu needed\n",
2616 computed_guess, length_in_chars, needed));
2618 /* If slope increased, use it, but discard this result for
2619 * length 1 strings, as we can't be sure that it's a real slope
2621 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
2625 STRLEN old_m = PL_collxfrm_mult;
2626 STRLEN old_b = PL_collxfrm_base;
2630 PL_collxfrm_mult = new_m;
2631 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2632 computed_guess = PL_collxfrm_base
2633 + (PL_collxfrm_mult * length_in_chars);
2634 if (computed_guess < needed) {
2635 PL_collxfrm_base += needed - computed_guess;
2638 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2639 "%s: %d: slope is now %zu; was %zu, base "
2640 "is now %zu; was %zu\n",
2642 PL_collxfrm_mult, old_m,
2643 PL_collxfrm_base, old_b));
2645 else { /* Slope didn't change, but 'b' did */
2646 const STRLEN new_b = needed
2649 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2650 "%s: %d: base is now %zu; was %zu\n",
2652 new_b, PL_collxfrm_base));
2653 PL_collxfrm_base = new_b;
2660 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2661 DEBUG_L(PerlIO_printf(Perl_debug_log,
2662 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2663 *xlen, PERL_INT_MAX));
2667 /* A well-behaved strxfrm() returns exactly how much space it needs
2668 * (usually not including the trailing NUL) when it fails due to not
2669 * enough space being provided. Assume that this is the case unless
2670 * it's been proven otherwise */
2671 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
2672 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
2674 else { /* Here, either:
2675 * 1) The strxfrm() has previously shown bad behavior; or
2676 * 2) It isn't the first time through the loop, which means
2677 * that the strxfrm() is now showing bad behavior, because
2678 * we gave it what it said was needed in the previous
2679 * iteration, and it came back saying it needed still more.
2680 * (Many versions of cygwin fit this. When the buffer size
2681 * isn't sufficient, they return the input size instead of
2682 * how much is needed.)
2683 * Increase the buffer size by a fixed percentage and try again.
2685 xAlloc += (xAlloc / 4) + 1;
2686 PL_strxfrm_is_behaved = FALSE;
2690 if (DEBUG_Lv_TEST || debug_initialization) {
2691 PerlIO_printf(Perl_debug_log,
2692 "_mem_collxfrm required more space than previously calculated"
2693 " for locale %s, trying again with new guess=%d+%zu\n",
2694 PL_collation_name, (int) COLLXFRM_HDR_LEN,
2695 xAlloc - COLLXFRM_HDR_LEN);
2702 Renew(xbuf, xAlloc, char);
2703 if (UNLIKELY(! xbuf)) {
2704 DEBUG_L(PerlIO_printf(Perl_debug_log,
2705 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
2715 if (DEBUG_Lv_TEST || debug_initialization) {
2717 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2718 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
2719 PerlIO_printf(Perl_debug_log, "%s\n",
2720 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2726 /* Free up unneeded space; retain ehough for trailing NUL */
2727 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
2729 if (s != input_string) {
2737 if (s != input_string) {
2744 if (DEBUG_Lv_TEST || debug_initialization) {
2745 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
2756 S_print_collxfrm_input_and_return(pTHX_
2757 const char * const s,
2758 const char * const e,
2759 const STRLEN * const xlen,
2763 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
2765 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
2766 (UV)PL_collation_ix);
2768 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
2771 PerlIO_printf(Perl_debug_log, "NULL");
2773 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
2775 print_bytes_for_locale(s, e, is_utf8);
2777 PerlIO_printf(Perl_debug_log, "'\n");
2781 S_print_bytes_for_locale(pTHX_
2782 const char * const s,
2783 const char * const e,
2787 bool prev_was_printable = TRUE;
2788 bool first_time = TRUE;
2790 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
2794 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
2797 if (! prev_was_printable) {
2798 PerlIO_printf(Perl_debug_log, " ");
2800 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
2801 prev_was_printable = TRUE;
2805 PerlIO_printf(Perl_debug_log, " ");
2807 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
2808 prev_was_printable = FALSE;
2810 t += (is_utf8) ? UTF8SKIP(t) : 1;
2815 # endif /* #ifdef DEBUGGING */
2816 #endif /* USE_LOCALE_COLLATE */
2821 Perl__is_cur_LC_category_utf8(pTHX_ int category)
2823 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
2824 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
2825 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
2826 * could give the wrong result. The result will very likely be correct for
2827 * languages that have commonly used non-ASCII characters, but for notably
2828 * English, it comes down to if the locale's name ends in something like
2829 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
2831 char *save_input_locale = NULL;
2836 assert(category != LC_ALL);
2840 /* First dispose of the trivial cases */
2841 save_input_locale = do_setlocale_r(category, NULL);
2842 if (! save_input_locale) {
2843 DEBUG_L(PerlIO_printf(Perl_debug_log,
2844 "Could not find current locale for category %d\n",
2846 return FALSE; /* XXX maybe should croak */
2848 save_input_locale = stdize_locale(savepv(save_input_locale));
2849 if (isNAME_C_OR_POSIX(save_input_locale)) {
2850 DEBUG_L(PerlIO_printf(Perl_debug_log,
2851 "Current locale for category %d is %s\n",
2852 category, save_input_locale));
2853 Safefree(save_input_locale);
2857 # if defined(USE_LOCALE_CTYPE) \
2858 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
2860 { /* Next try nl_langinfo or MB_CUR_MAX if available */
2862 char *save_ctype_locale = NULL;
2865 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
2867 /* Get the current LC_CTYPE locale */
2868 save_ctype_locale = do_setlocale_c(LC_CTYPE, NULL);
2869 if (! save_ctype_locale) {
2870 DEBUG_L(PerlIO_printf(Perl_debug_log,
2871 "Could not find current locale for LC_CTYPE\n"));
2872 goto cant_use_nllanginfo;
2874 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
2876 /* If LC_CTYPE and the desired category use the same locale, this
2877 * means that finding the value for LC_CTYPE is the same as finding
2878 * the value for the desired category. Otherwise, switch LC_CTYPE
2879 * to the desired category's locale */
2880 if (strEQ(save_ctype_locale, save_input_locale)) {
2881 Safefree(save_ctype_locale);
2882 save_ctype_locale = NULL;
2884 else if (! do_setlocale_c(LC_CTYPE, save_input_locale)) {
2885 DEBUG_L(PerlIO_printf(Perl_debug_log,
2886 "Could not change LC_CTYPE locale to %s\n",
2887 save_input_locale));
2888 Safefree(save_ctype_locale);
2889 goto cant_use_nllanginfo;
2893 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
2894 save_input_locale));
2896 /* Here the current LC_CTYPE is set to the locale of the category whose
2897 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
2898 * should give the correct results */
2900 # if defined(HAS_NL_LANGINFO) && defined(CODESET)
2901 /* The task is easiest if has this POSIX 2001 function */
2904 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
2905 /* FALSE => already in dest locale */
2907 DEBUG_L(PerlIO_printf(Perl_debug_log,
2908 "\tnllanginfo returned CODESET '%s'\n", codeset));
2910 if (codeset && strNE(codeset, "")) {
2911 /* If we switched LC_CTYPE, switch back */
2912 if (save_ctype_locale) {
2913 do_setlocale_c(LC_CTYPE, save_ctype_locale);
2914 Safefree(save_ctype_locale);
2917 is_utf8 = ( ( strlen(codeset) == STRLENs("UTF-8")
2918 && foldEQ(codeset, STR_WITH_LEN("UTF-8")))
2919 || ( strlen(codeset) == STRLENs("UTF8")
2920 && foldEQ(codeset, STR_WITH_LEN("UTF8"))));
2922 DEBUG_L(PerlIO_printf(Perl_debug_log,
2923 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
2925 Safefree(save_input_locale);
2933 /* Here, either we don't have nl_langinfo, or it didn't return a
2934 * codeset. Try MB_CUR_MAX */
2936 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
2937 * Unicode code point. Since UTF-8 is the only non-single byte
2938 * encoding we handle, we just say any such encoding is UTF-8, and if
2939 * turns out to be wrong, other things will fail */
2940 is_utf8 = (unsigned) MB_CUR_MAX >= STRLENs(MAX_UNICODE_UTF8);
2942 DEBUG_L(PerlIO_printf(Perl_debug_log,
2943 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
2944 (int) MB_CUR_MAX, is_utf8));
2946 Safefree(save_input_locale);
2950 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
2951 * since they are both in the C99 standard. We can feed a known byte
2952 * string to the latter function, and check that it gives the expected
2958 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
2960 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
2963 if ( len != STRLENs(REPLACEMENT_CHARACTER_UTF8)
2964 || wc != (wchar_t) UNICODE_REPLACEMENT)
2967 DEBUG_L(PerlIO_printf(Perl_debug_log, "\replacement=U+%x\n",
2969 DEBUG_L(PerlIO_printf(Perl_debug_log,
2970 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
2977 /* If we switched LC_CTYPE, switch back */
2978 if (save_ctype_locale) {
2979 do_setlocale_c(LC_CTYPE, save_ctype_locale);
2980 Safefree(save_ctype_locale);
2989 cant_use_nllanginfo:
2991 # else /* nl_langinfo should work if available, so don't bother compiling this
2992 fallback code. The final fallback of looking at the name is
2993 compiled, and will be executed if nl_langinfo fails */
2995 /* nl_langinfo not available or failed somehow. Next try looking at the
2996 * currency symbol to see if it disambiguates things. Often that will be
2997 * in the native script, and if the symbol isn't in UTF-8, we know that the
2998 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
2999 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
3002 # ifdef HAS_LOCALECONV
3003 # ifdef USE_LOCALE_MONETARY
3006 char *save_monetary_locale = NULL;
3007 bool only_ascii = FALSE;
3008 bool is_utf8 = FALSE;
3011 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
3012 * the desired category, if it isn't that locale already */
3014 if (category != LC_MONETARY) {
3016 save_monetary_locale = do_setlocale_c(LC_MONETARY, NULL);
3017 if (! save_monetary_locale) {
3018 DEBUG_L(PerlIO_printf(Perl_debug_log,
3019 "Could not find current locale for LC_MONETARY\n"));
3020 goto cant_use_monetary;
3022 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
3024 if (strEQ(save_monetary_locale, save_input_locale)) {
3025 Safefree(save_monetary_locale);
3026 save_monetary_locale = NULL;
3028 else if (! do_setlocale_c(LC_MONETARY, save_input_locale)) {
3029 DEBUG_L(PerlIO_printf(Perl_debug_log,
3030 "Could not change LC_MONETARY locale to %s\n",
3031 save_input_locale));
3032 Safefree(save_monetary_locale);
3033 goto cant_use_monetary;
3037 /* Here the current LC_MONETARY is set to the locale of the category
3038 * whose information is desired. */
3042 || ! lc->currency_symbol
3043 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
3045 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));
3049 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
3052 /* If we changed it, restore LC_MONETARY to its original locale */
3053 if (save_monetary_locale) {
3054 do_setlocale_c(LC_MONETARY, save_monetary_locale);
3055 Safefree(save_monetary_locale);
3060 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3061 * otherwise assume the locale is UTF-8 if and only if the symbol
3062 * is non-ascii UTF-8. */
3063 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3064 save_input_locale, is_utf8));
3065 Safefree(save_input_locale);
3071 # endif /* USE_LOCALE_MONETARY */
3072 # endif /* HAS_LOCALECONV */
3074 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
3076 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3077 * the names of the months and weekdays, timezone, and am/pm indicator */
3079 char *save_time_locale = NULL;
3081 bool is_dst = FALSE;
3085 char * formatted_time;
3088 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3089 * desired category, if it isn't that locale already */
3091 if (category != LC_TIME) {
3093 save_time_locale = do_setlocale_c(LC_TIME, NULL);
3094 if (! save_time_locale) {
3095 DEBUG_L(PerlIO_printf(Perl_debug_log,
3096 "Could not find current locale for LC_TIME\n"));
3099 save_time_locale = stdize_locale(savepv(save_time_locale));
3101 if (strEQ(save_time_locale, save_input_locale)) {
3102 Safefree(save_time_locale);
3103 save_time_locale = NULL;
3105 else if (! do_setlocale_c(LC_TIME, save_input_locale)) {
3106 DEBUG_L(PerlIO_printf(Perl_debug_log,
3107 "Could not change LC_TIME locale to %s\n",
3108 save_input_locale));
3109 Safefree(save_time_locale);
3114 /* Here the current LC_TIME is set to the locale of the category
3115 * whose information is desired. Look at all the days of the week and
3116 * month names, and the timezone and am/pm indicator for UTF-8 variant
3117 * characters. The first such a one found will tell us if the locale
3118 * is UTF-8 or not */
3120 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3121 formatted_time = my_strftime("%A %B %Z %p",
3122 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
3123 if ( ! formatted_time
3124 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3127 /* Here, we didn't find a non-ASCII. Try the next time through
3128 * with the complemented dst and am/pm, and try with the next
3129 * weekday. After we have gotten all weekdays, try the next
3132 hour = (hour + 12) % 24;
3140 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3141 * false otherwise. But first, restore LC_TIME to its original
3142 * locale if we changed it */
3143 if (save_time_locale) {
3144 do_setlocale_c(LC_TIME, save_time_locale);
3145 Safefree(save_time_locale);
3148 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3150 is_utf8_string((U8 *) formatted_time, 0)));
3151 Safefree(save_input_locale);
3152 return is_utf8_string((U8 *) formatted_time, 0);
3155 /* Falling off the end of the loop indicates all the names were just
3156 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3157 * to its original locale */
3158 if (save_time_locale) {
3159 do_setlocale_c(LC_TIME, save_time_locale);
3160 Safefree(save_time_locale);
3162 DEBUG_L(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));
3168 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
3170 /* This code is ifdefd out because it was found to not be necessary in testing
3171 * on our dromedary test machine, which has over 700 locales. There, this
3172 * added no value to looking at the currency symbol and the time strings. I
3173 * left it in so as to avoid rewriting it if real-world experience indicates
3174 * that dromedary is an outlier. Essentially, instead of returning abpve if we
3175 * haven't found illegal utf8, we continue on and examine all the strerror()
3176 * messages on the platform for utf8ness. If all are ASCII, we still don't
3177 * know the answer; but otherwise we have a pretty good indication of the
3178 * utf8ness. The reason this doesn't help much is that the messages may not
3179 * have been translated into the locale. The currency symbol and time strings
3180 * are much more likely to have been translated. */
3183 bool is_utf8 = FALSE;
3184 bool non_ascii = FALSE;
3185 char *save_messages_locale = NULL;
3186 const char * errmsg = NULL;
3188 /* Like above, we set LC_MESSAGES to the locale of the desired
3189 * category, if it isn't that locale already */
3191 if (category != LC_MESSAGES) {
3193 save_messages_locale = do_setlocale_c(LC_MESSAGES, NULL);
3194 if (! save_messages_locale) {
3195 DEBUG_L(PerlIO_printf(Perl_debug_log,
3196 "Could not find current locale for LC_MESSAGES\n"));
3197 goto cant_use_messages;
3199 save_messages_locale = stdize_locale(savepv(save_messages_locale));
3201 if (strEQ(save_messages_locale, save_input_locale)) {
3202 Safefree(save_messages_locale);
3203 save_messages_locale = NULL;
3205 else if (! do_setlocale_c(LC_MESSAGES, save_input_locale)) {
3206 DEBUG_L(PerlIO_printf(Perl_debug_log,
3207 "Could not change LC_MESSAGES locale to %s\n",
3208 save_input_locale));
3209 Safefree(save_messages_locale);
3210 goto cant_use_messages;
3214 /* Here the current LC_MESSAGES is set to the locale of the category
3215 * whose information is desired. Look through all the messages. We
3216 * can't use Strerror() here because it may expand to code that
3217 * segfaults in miniperl */
3219 for (e = 0; e <= sys_nerr; e++) {
3221 errmsg = sys_errlist[e];
3222 if (errno || !errmsg) {
3225 errmsg = savepv(errmsg);
3226 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
3228 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3234 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3235 if (save_messages_locale) {
3236 do_setlocale_c(LC_MESSAGES, save_messages_locale);
3237 Safefree(save_messages_locale);
3242 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3243 * any non-ascii means it is one; otherwise we assume it isn't */
3244 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3247 Safefree(save_input_locale);
3251 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));
3256 # endif /* the code that is compiled when no nl_langinfo */
3258 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
3261 /* As a last resort, look at the locale name to see if it matches
3262 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3263 * return of setlocale(), is actually defined to be opaque, so we can't
3264 * really rely on the absence of various substrings in the name to indicate
3265 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3266 * be a UTF-8 locale. Similarly for the other common names */
3268 final_pos = strlen(save_input_locale) - 1;
3269 if (final_pos >= 3) {
3270 char *name = save_input_locale;
3272 /* Find next 'U' or 'u' and look from there */
3273 while ((name += strcspn(name, "Uu") + 1)
3274 <= save_input_locale + final_pos - 2)
3276 if ( isALPHA_FOLD_NE(*name, 't')
3277 || isALPHA_FOLD_NE(*(name + 1), 'f'))
3282 if (*(name) == '-') {
3283 if ((name > save_input_locale + final_pos - 1)) {
3288 if (*(name) == '8') {
3289 DEBUG_L(PerlIO_printf(Perl_debug_log,
3290 "Locale %s ends with UTF-8 in name\n",
3291 save_input_locale));
3292 Safefree(save_input_locale);
3296 DEBUG_L(PerlIO_printf(Perl_debug_log,
3297 "Locale %s doesn't end with UTF-8 in name\n",
3298 save_input_locale));
3304 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
3305 if (memENDs(save_input_locale, final_pos, "65001")) {
3306 DEBUG_L(PerlIO_printf(Perl_debug_log,
3307 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
3308 save_input_locale));
3309 Safefree(save_input_locale);
3315 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3316 * since we are about to return FALSE anyway, there is no point in doing
3317 * this extra work */
3320 if (instr(save_input_locale, "8859")) {
3321 DEBUG_L(PerlIO_printf(Perl_debug_log,
3322 "Locale %s has 8859 in name, not UTF-8 locale\n",
3323 save_input_locale));
3324 Safefree(save_input_locale);
3329 DEBUG_L(PerlIO_printf(Perl_debug_log,
3330 "Assuming locale %s is not a UTF-8 locale\n",
3331 save_input_locale));
3332 Safefree(save_input_locale);
3340 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3343 /* Internal function which returns if we are in the scope of a pragma that
3344 * enables the locale category 'category'. 'compiling' should indicate if
3345 * this is during the compilation phase (TRUE) or not (FALSE). */
3347 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3349 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3350 if (! categories || categories == &PL_sv_placeholder) {
3354 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3355 * a valid unsigned */
3356 assert(category >= -1);
3357 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3361 Perl_my_strerror(pTHX_ const int errnum)
3363 /* Returns a mortalized copy of the text of the error message associated
3364 * with 'errnum'. It uses the current locale's text unless the platform
3365 * doesn't have the LC_MESSAGES category or we are not being called from
3366 * within the scope of 'use locale'. In the former case, it uses whatever
3367 * strerror returns; in the latter case it uses the text from the C locale.
3369 * The function just calls strerror(), but temporarily switches, if needed,
3370 * to the C locale */
3375 #ifndef USE_LOCALE_MESSAGES
3377 /* If platform doesn't have messages category, we don't do any switching to
3378 * the C locale; we just use whatever strerror() returns */
3380 errstr = savepv(Strerror(errnum));
3382 #else /* Has locale messages */
3384 const bool within_locale_scope = IN_LC(LC_MESSAGES);
3386 # if defined(HAS_POSIX_2008_LOCALE) && defined(HAS_STRERROR_L)
3388 /* This function is trivial if we don't have to worry about thread safety
3389 * and have strerror_l(), as it handles the switch of locales so we don't
3390 * have to deal with that. We don't have to worry about thread safety if
3391 * this is an unthreaded build, or if strerror_r() is also available. Both
3392 * it and strerror_l() are thread-safe. Plain strerror() isn't thread
3393 * safe. But on threaded builds when strerror_r() is available, the
3394 * apparent call to strerror() below is actually a macro that
3395 * behind-the-scenes calls strerror_r().
3398 # if ! defined(USE_ITHREADS) || defined(HAS_STRERROR_R)
3400 if (within_locale_scope) {
3401 errstr = savepv(strerror(errnum));
3404 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
3409 /* Here we have strerror_l(), but not strerror_r() and we are on a
3410 * threaded-build. We use strerror_l() for everything, constructing a
3411 * locale to pass to it if necessary */
3413 bool do_free = FALSE;
3414 locale_t locale_to_use;
3416 if (within_locale_scope) {
3417 locale_to_use = uselocale((locale_t) 0);
3418 if (locale_to_use == LC_GLOBAL_LOCALE) {
3419 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
3423 else { /* Use C locale if not within 'use locale' scope */
3424 locale_to_use = PL_C_locale_obj;
3427 errstr = savepv(strerror_l(errnum, locale_to_use));
3430 freelocale(locale_to_use);
3434 # else /* Doesn't have strerror_l() */
3436 # ifdef USE_POSIX_2008_LOCALE
3438 locale_t save_locale = NULL;
3442 char * save_locale = NULL;
3443 bool locale_is_C = FALSE;
3445 /* We have a critical section to prevent another thread from changing the
3446 * locale out from under us (or zapping the buffer returned from
3452 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3453 "my_strerror called with errnum %d\n", errnum));
3454 if (! within_locale_scope) {
3457 # ifdef USE_POSIX_2008_LOCALE /* Use the thread-safe locale functions */
3459 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3460 "Not within locale scope, about to call"
3461 " uselocale(0x%p)\n", PL_C_locale_obj));
3462 save_locale = uselocale(PL_C_locale_obj);
3463 if (! save_locale) {
3464 DEBUG_L(PerlIO_printf(Perl_debug_log,
3465 "uselocale failed, errno=%d\n", errno));
3468 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3469 "uselocale returned 0x%p\n", save_locale));
3472 # else /* Not thread-safe build */
3474 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
3475 if (! save_locale) {
3476 DEBUG_L(PerlIO_printf(Perl_debug_log,
3477 "setlocale failed, errno=%d\n", errno));
3480 locale_is_C = isNAME_C_OR_POSIX(save_locale);
3482 /* Switch to the C locale if not already in it */
3483 if (! locale_is_C) {
3485 /* The setlocale() just below likely will zap 'save_locale', so
3487 save_locale = savepv(save_locale);
3488 do_setlocale_c(LC_MESSAGES, "C");
3494 } /* end of ! within_locale_scope */
3496 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
3497 __FILE__, __LINE__));
3500 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3501 "Any locale change has been done; about to call Strerror\n"));
3502 errstr = savepv(Strerror(errnum));
3504 if (! within_locale_scope) {
3507 # ifdef USE_POSIX_2008_LOCALE
3509 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3510 "%s: %d: not within locale scope, restoring the locale\n",
3511 __FILE__, __LINE__));
3512 if (save_locale && ! uselocale(save_locale)) {
3513 DEBUG_L(PerlIO_printf(Perl_debug_log,
3514 "uselocale restore failed, errno=%d\n", errno));
3520 if (save_locale && ! locale_is_C) {
3521 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
3522 DEBUG_L(PerlIO_printf(Perl_debug_log,
3523 "setlocale restore failed, errno=%d\n", errno));
3525 Safefree(save_locale);
3532 # endif /* End of doesn't have strerror_l */
3533 #endif /* End of does have locale messages */
3537 if (DEBUG_Lv_TEST) {
3538 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
3539 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
3540 PerlIO_printf(Perl_debug_log, "'\n");
3551 =for apidoc sync_locale
3553 Changing the program's locale should be avoided by XS code. Nevertheless,
3554 certain non-Perl libraries called from XS, such as C<Gtk> do so. When this
3555 happens, Perl needs to be told that the locale has changed. Use this function
3556 to do so, before returning to Perl.
3562 Perl_sync_locale(pTHX)
3566 #ifdef USE_LOCALE_CTYPE
3568 newlocale = do_setlocale_c(LC_CTYPE, NULL);
3569 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3570 "%s:%d: %s\n", __FILE__, __LINE__,
3571 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
3572 new_ctype(newlocale);
3574 #endif /* USE_LOCALE_CTYPE */
3575 #ifdef USE_LOCALE_COLLATE
3577 newlocale = do_setlocale_c(LC_COLLATE, NULL);
3578 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3579 "%s:%d: %s\n", __FILE__, __LINE__,
3580 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
3581 new_collate(newlocale);
3584 #ifdef USE_LOCALE_NUMERIC
3586 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
3587 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3588 "%s:%d: %s\n", __FILE__, __LINE__,
3589 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
3590 new_numeric(newlocale);
3592 #endif /* USE_LOCALE_NUMERIC */
3596 #if defined(DEBUGGING) && defined(USE_LOCALE)
3599 S_setlocale_debug_string(const int category, /* category number,
3601 const char* const locale, /* locale name */
3603 /* return value from setlocale() when attempting to
3604 * set 'category' to 'locale' */
3605 const char* const retval)
3607 /* Returns a pointer to a NUL-terminated string in static storage with
3608 * added text about the info passed in. This is not thread safe and will
3609 * be overwritten by the next call, so this should be used just to
3610 * formulate a string to immediately print or savepv() on. */
3612 /* initialise to a non-null value to keep it out of BSS and so keep
3613 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
3614 static char ret[128] = "If you can read this, thank your buggy C"
3615 " library strlcpy(), and change your hints file"
3621 const unsigned int highest_index = LC_ALL_INDEX;
3625 const unsigned int highest_index = NOMINAL_LC_ALL_INDEX - 1;
3630 my_strlcpy(ret, "setlocale(", sizeof(ret));
3632 /* Look for category in our list, and if found, add its name */
3633 for (i = 0; i <= highest_index; i++) {
3634 if (category == categories[i]) {
3635 my_strlcat(ret, category_names[i], sizeof(ret));
3636 goto found_category;
3640 /* Unknown category to us */
3641 my_snprintf(ret, sizeof(ret), "%s? %d", ret, category);
3645 my_strlcat(ret, ", ", sizeof(ret));
3648 my_strlcat(ret, "\"", sizeof(ret));
3649 my_strlcat(ret, locale, sizeof(ret));
3650 my_strlcat(ret, "\"", sizeof(ret));
3653 my_strlcat(ret, "NULL", sizeof(ret));
3656 my_strlcat(ret, ") returned ", sizeof(ret));
3659 my_strlcat(ret, "\"", sizeof(ret));
3660 my_strlcat(ret, retval, sizeof(ret));
3661 my_strlcat(ret, "\"", sizeof(ret));
3664 my_strlcat(ret, "NULL", sizeof(ret));
3667 assert(strlen(ret) < sizeof(ret));
3676 * ex: set ts=8 sts=4 sw=4 et: