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
36 * There is more than the typical amount of variation between platforms with
37 * regard to locale handling. At the end of these introductory comments, are
38 * listed various relevent Configuration options, including some that can be
39 * used to pretend to some extent that this is being developed on a different
40 * platform than it actually is. This allows you to make changes and catch
41 * some errors without having access to those other platforms.
43 * This code now has multi-thread-safe locale handling on systems that support
44 * that. This is completely transparent to most XS code. On earlier systems,
45 * it would be possible to emulate thread-safe locales, but this likely would
46 * involve a lot of locale switching, and would require XS code changes.
47 * Macros could be written so that the code wouldn't have to know which type of
48 * system is being used.
50 * Table-driven code is used for simplicity and clarity, as many operations
51 * differ only in which category is being worked on. However the system
52 * categories need not be small contiguous integers, so do not lend themselves
53 * to table lookup. Instead we have created our own equivalent values which
54 * are all small contiguous non-negative integers, and translation functions
55 * between the two sets. For category 'LC_foo', the name of our index is
56 * LC_foo_INDEX_. Various parallel tables, indexed by these, are used for the
57 * translation. The tables are generated at compile-time based on platform
58 * characteristics and Configure options. They hide from the code many of the
59 * vagaries of the different locale implementations out there.
61 * On unthreaded perls, most operations expand out to just the basic
62 * setlocale() calls. That sort of is true on threaded perls on modern Windows
63 * systems where the same API, after set up, is used for thread-safe locale
64 * handling. (But there are complications on Windows due to internal character
65 * set issues.) On other systems, there is a completely different API,
66 * specified in POSIX 2008, to do thread-safe locales. On these systems, our
67 * bool_setlocale_2008_i() function is used to hide the different API from the
68 * outside. This makes it completely transparent to most XS code.
70 * A huge complicating factor is that the LC_NUMERIC category is normally held
71 * in the C locale, except during those relatively rare times when it needs to
72 * be in the underlying locale. There is a bunch of code to accomplish this,
73 * and to allow easy switches from one state to the other.
75 * In addition, the setlocale equivalents have versions for the return context,
76 * 'void' and 'bool', besides the full return value. This can present
77 * opportunities for avoiding work. We don't have to necessarily create a safe
78 * copy to return if no return is desired.
80 * There are 3.5 major implementations here; which one chosen depends on what
81 * the platform has available, and Configuration options.
83 * 1) Raw posix_setlocale(). This implementation is basically the libc
84 * setlocale(), with possibly minor tweaks. This is used for startup, and
85 * always for unthreaded perls, and when the API for safe locale threading
86 * is identical to the unsafe API (Windows, currently).
88 * This implementation is composed of two layers:
89 * a) posix_setlocale() implements the libc setlocale(). In most cases,
90 * it is just an alias for the libc version. But Windows doesn't
91 * fully conform to the POSIX standard, and this is a layer on top of
92 * libc to bring it more into conformance. And in Configurations
93 * where perl is to ignore some locale categories that the libc
94 * setlocale() knows about, there is a layer to cope with that.
95 * b) stdized_setlocale() is a layer above a) that fixes some vagaries in
96 * the return value of the libc setlocale(). On most platforms this
97 * layer is empty; it requires perl to be Configured with a parameter
98 * indicating the platform's defect, in order to be activated. The
99 * current ones are listed at the definition of the macro.
101 * 2) An implementation that adds a minimal layer above implementation 1),
102 * making that implementation uninterruptible and returning a
103 * per-thread/per-category value.
105 * 3a and 3b) An implementation of POSIX 2008 thread-safe locale handling,
106 * hiding from the programmer the completely different API for this.
107 * This automatically makes almost all code thread-safe without need for
108 * changes. This implementation is chosen on threaded perls when the
109 * platform properly supports the POSIX 2008 functions, and when there is no
110 * manual override to the contrary passed to Configure.
112 * 3a) is when the platform has a documented reliable querylocale() function
113 * or equivalent that is selected to be used.
114 * 3b) is when we have to emulate that functionality.
116 * Unfortunately, it seems that some platforms that claim to support these
117 * are buggy, in one way or another. There are workarounds encoded here,
118 * where feasible, for platforms where the bugs are amenable to that
119 * (glibc, for example). But other platforms instead don't use this
122 * z/OS (os390) is an outlier. Locales really don't work under threads when
123 * either the radix character isn't a dot, or attempts are made to change
124 * locales after the first thread is created. The reason is that IBM has made
125 * it thread-safe by refusing to change locales (returning failure if
126 * attempted) any time after an application has called pthread_create() to
127 * create another thread. The expectation is that an application will set up
128 * its locale information before the first fork, and be stable thereafter. But
129 * perl toggles LC_NUMERIC if the locale's radix character isn't a dot, as do
130 * the other toggles, which are less common.
132 * Associated with each implementation are three sets of macros that translate
133 * a consistent API into what that implementation needs. Each set consists of
134 * three macros with the suffixes:
135 * _c Means the argument is a locale category number known at compile time.
136 * An example would be LC_TIME. This token is a compile-time constant
137 * and can be passed to a '_c' macro.
138 * _r Means the argument is a locale category number whose value might not be
139 * known until runtime
140 * _i Means the argument is our internal index of a locale category
142 * The three sets are: ('_X' means one of '_c', '_r', '_i')
143 * 1) bool_setlocale_X()
144 * This calls the appropriate setlocale()-equivalent for the
145 * implementation, with the category and new locale. The input locale is
146 * not necessarily valid, so the return is true or false depending on
147 * whether or not the setlocale() succeeded. This is not used for
148 * querying the locale, so the input locale must not be NULL.
150 * This macro is suitable for toggling the locale back and forth during an
151 * operation. For example, the names of days and months under LC_TIME are
152 * strings that are also subject to LC_CTYPE. If the locales of these two
153 * categories differ, mojibake can result on many platforms. The code
154 * here will toggle LC_CTYPE into the locale of LC_TIME temporarily to
157 * Several categories require extra work when their locale is changed.
158 * LC_CTYPE, for example, requires the calculation of the table of which
159 * characters fold to which others under /i pattern matching or fc(), as
160 * folding is not a concept in POSIX. This table isn't needed when the
161 * LC_CTYPE locale gets toggled during an operation, and will be toggled
162 * back before return to the caller. To save work that would be
163 * discarded, the bool_setlocale_X() implementations don't do this extra
164 * work. Instead, there is a separate function for just this purpose to
165 * be done before control is transferred back to the external caller. All
166 * categories that have such requirements have such a function. The
167 * update_functions[] array contains pointers to them (or NULL for
168 * categories which don't need a function).
170 * Care must be taken to remember to call the separate function before
171 * returning to an external caller, and to not use things it updates
172 * before its call. An alternative approach would be to have
173 * bool_setlocale_X() always call the update, which would return
174 * immediately if a flag wasn't set indicating it was time to actually
177 * 2) void_setlocale_X()
178 * This is like bool_setlocale_X(), but it is used only when it is
179 * expected that the call must succeed, or something is seriously wrong.
180 * A panic is issued if it fails. The caller uses this form when it just
181 * wants to assume things worked.
184 * This returns a string that specifies the current locale for the given
185 * category given by the input argument. The string is safe from other
186 * threads zapping it, and the caller need not worry about freeing it, but
187 * it may be mortalized, so must be copied if you need to preserve it
188 * across calls, or long term. This returns the actual current locale,
189 * not the nominal. These differ, for example, when LC_NUMERIC is
190 * supposed to be a locale whose decimal radix character is a comma. As
191 * mentioned above, Perl actually keeps this category set to C in such
192 * circumstances so that XS code can just assume a dot radix character.
193 * querylocale_X() returns the locale that libc has stored at this moment,
194 * so most of the time will return a locale whose radix character is a
195 * dot. The macro query_nominal_locale_i() can be used to get the nominal
196 * locale that an external caller would expect, for all categories except
197 * LC_ALL. For that, you can use the function
198 * S_calculate_LC_ALL_string(). Or S_native_querylocale_i() will operate
201 * The underlying C API that this implements uses category numbers, hence the
202 * code is structured to use '_r' at the API level to convert to indexes, which
203 * are then used internally with the '_i' forms.
205 * The splitting apart into setting vs querying means that the return value of
206 * the bool macros is not subject to potential clashes with other threads,
207 * eliminating any need for the calling code to worry about that and get it
208 * wrong. Whereas, you do have to think about thread interactions when using a
211 * Additionally, for the implementations where there aren't any complications,
212 * a setlocale_i() is defined that is like plain setlocale(), returning the new
213 * locale. Thus it combines a bool_setlocale_X() with a querylocale_X(). It
214 * is used only for performance on implementations that allow it, such as
215 * non-threaded perls.
217 * There are also a few other macros herein that use this naming convention to
218 * describe their category parameter.
220 * Relevant Configure options
222 * -Accflags=-DNO_LOCALE
223 * This compiles perl to always use the C locale, ignoring any
224 * attempts to change it. This could be useful on platforms with a
225 * crippled locale implementation.
227 * -Accflags=-DNO_THREAD_SAFE_LOCALE
228 * Even if thread-safe operations are available on this platform and
229 * would otherwise be used (because this is a perl with multiplicity),
230 * perl is compiled to not use them. This could be useful on
231 * platforms where the libc is buggy.
233 * -Accflags=-DNO_POSIX_2008_LOCALE
234 * Even if the libc locale operations specified by the Posix 2008
235 * Standard are available on this platform and would otherwise be used
236 * (because this is a perl with multiplicity), perl is compiled to not
237 * use them. This could be useful on platforms where the libc is
238 * buggy. This is like NO_THREAD_SAFE_LOCALE, but has no effect on
239 * platforms that don't have these functions.
241 * -Accflags=-DUSE_POSIX_2008_LOCALE
242 * Normally, setlocale() is used for locale operations on perls
243 * compiled without multiplicity. This option causes the locale
244 * operations defined by the Posix 2008 Standard to always be used
245 * instead. This could be useful on platforms where the libc
246 * setlocale() is buggy.
248 * -Accflags=-DNO_THREAD_SAFE_QUERYLOCALE
249 * This applies only to platforms that have a querylocale() libc
250 * function. perl assumes that that function is thread-safe, unless
251 * overridden by this, typically in a hints file. When overridden,
252 * querylocale() is called only while the locale mutex is locked, and
253 * the result is copied to a per-thread place before unlocking.
255 * -Accflags=-DUSE_NL_LOCALE_NAME
256 * glibc has an undocumented equivalent function to querylocale(). It
257 * currently isn't used by default because it is undocumented. But
258 * testing hasn't found any problems with it. Using this Configure
259 * option enables it on systems that have it (with no effect on
260 * systems lacking it). Enabling this removes the need for perl
261 * to keep its own records, hence is more efficient and guaranteed to
264 * -Accflags=-DNO_LOCALE_CTYPE
265 * -Accflags=-DNO_LOCALE_NUMERIC
268 * If the named category(ies) does(do) not exist on this platform,
269 * these have no effect. Otherwise they cause perl to be compiled to
270 * always keep the named category(ies) in the C locale.
272 * -Accflags=-DHAS_BROKEN_SETLOCALE_QUERY_LC_ALL
273 * This would be set in a hints file to tell perl that doing a libc
274 * setlocale(LC_ALL, NULL)
275 * can give erroneous results, and perl will compensate to get the
276 * correct results. This is known to be a problem in earlier AIX
279 * -Accflags=-DHAS_LF_IN_SETLOCALE_RETURN
280 * This would be set in a hints file to tell perl that a libc
281 * setlocale() can return results containing \n characters that need
282 * to be stripped off. khw believes there aren't any such platforms
283 * still in existence.
285 * -Accflags=USE_FAKE_LC_ALL_POSITIONAL_NOTATION
286 * This is used when developing Perl on a platform that uses
287 * 'name=value;' notation to represent LC_ALL when not all categories
288 * are the same. When so compiled, much of the code gets compiled
289 * and exercised that applies to platforms that instead use positional
290 * notation. This allows for finding many bugs in that portion of the
291 * implementation, without having to access such a platform.
293 * -Accflags=-DWIN32_USE_FAKE_OLD_MINGW_LOCALES
294 * This is used when developing Perl on a non-Windows platform to
295 * compile and exercise much of the locale-related code that instead
296 * applies to MingW platforms that don't use the more modern UCRT
297 * library. This allows for finding many bugs in that portion of the
298 * implementation, without having to access such a platform.
301 /* If the environment says to, we can output debugging information during
302 * initialization. This is done before option parsing, and before any thread
303 * creation, so can be a file-level static. (Must come before #including
307 /* Returns the Unix errno portion; ignoring any others. This is a macro here
308 * instead of putting it into perl.h, because unclear to khw what should be
310 #define GET_ERRNO saved_errno
313 static int debug_initialization = 0;
314 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
315 # define DEBUG_LOCALE_INITIALIZATION_ debug_initialization
317 # ifdef HAS_EXTENDED_OS_ERRNO
318 /* Output the non-zero errno and/or the non-zero extended errno */
319 # define DEBUG_ERRNO \
321 int extended = get_extended_os_errno(); \
322 const char * errno_string; \
323 if (GET_ERRNO == 0) { /* Skip output if both errno types are 0 */ \
324 if (LIKELY(extended == 0)) errno_string = ""; \
325 else errno_string = Perl_form(aTHX_ "; $^E=%d", extended); \
327 else if (LIKELY(extended == GET_ERRNO)) \
328 errno_string = Perl_form(aTHX_ "; $!=%d", GET_ERRNO); \
329 else errno_string = Perl_form(aTHX_ "; $!=%d, $^E=%d", \
330 GET_ERRNO, extended);
332 /* Output the errno, if non-zero */
333 # define DEBUG_ERRNO \
335 const char * errno_string = ""; \
336 if (GET_ERRNO != 0) { \
338 errno_string = Perl_form(aTHX_ "; $!=%d", GET_ERRNO); \
342 /* Automatically include the caller's file, and line number in debugging output;
343 * and the errno (and/or extended errno) if non-zero. On threaded perls add
345 # if defined(USE_ITHREADS) && ! defined(NO_LOCALE_THREADS)
346 # define DEBUG_PRE_STMTS \
348 PerlIO_printf(Perl_debug_log, "\n%s: %" LINE_Tf ": 0x%p%s: ", \
349 __FILE__, (line_t)__LINE__, aTHX_ \
352 # define DEBUG_PRE_STMTS \
354 PerlIO_printf(Perl_debug_log, "\n%s: %" LINE_Tf "%s: ", \
355 __FILE__, (line_t)__LINE__, \
358 # define DEBUG_POST_STMTS RESTORE_ERRNO;
360 # define debug_initialization 0
361 # define DEBUG_INITIALIZATION_set(v)
362 # define DEBUG_PRE_STMTS
363 # define DEBUG_POST_STMTS
367 #define PERL_IN_LOCALE_C
370 #ifdef WIN32_USE_FAKE_OLD_MINGW_LOCALES
372 /* Use -Accflags=-DWIN32_USE_FAKE_OLD_MINGW_LOCALES on a POSIX or *nix box
373 * to get a semblance of pretending the locale handling is that of a MingW
374 * that doesn't use UCRT (hence 'OLD' in the name). This exercizes code
375 * paths that are not compiled on non-Windows boxes, and allows for ASAN
376 * and PERL_MEMLOG. This is thus a way to see if locale.c on Windows is
377 * likely going to compile, without having to use a real Win32 box. And
378 * running the test suite will verify to a large extent our logic and memory
379 * allocation handling for such boxes. Of course the underlying calls are
380 * to the POSIX libc, so any differences in implementation between those and
381 * the Windows versions will not be caught by this. */
384 # undef P_CS_PRECEDES
385 # undef CURRENCY_SYMBOL
387 # undef _configthreadlocale
388 # define _configthreadlocale(arg) NOOP
390 # define MultiByteToWideChar(cp, flags, byte_string, m1, wstring, req_size) \
391 (mbsrtowcs(wstring, &(byte_string), req_size, NULL) + 1)
392 # define WideCharToMultiByte(cp, flags, wstring, m1, byte_string, \
393 req_size, default_char, found_default_char) \
394 (wcsrtombs(byte_string, &(wstring), req_size, NULL) + 1)
398 static const wchar_t * wsetlocale_buf = NULL;
399 static Size_t wsetlocale_buf_size = 0;
400 static PerlInterpreter * wsetlocale_buf_aTHX = NULL;
404 S_wsetlocale(const int category, const wchar_t * wlocale)
406 /* Windows uses a setlocale that takes a wchar_t* locale. Other boxes
407 * don't have this, so this Windows replacement converts the wchar_t input
408 * to plain 'char*', calls plain setlocale(), and converts the result back
411 const char * byte_locale = NULL;
413 byte_locale = Win_wstring_to_byte_string(CP_UTF8, wlocale);
416 const char * byte_result = setlocale(category, byte_locale);
417 Safefree(byte_locale);
418 if (byte_result == NULL) {
422 const wchar_t * wresult = Win_byte_string_to_wstring(CP_UTF8, byte_result);
428 /* Emulate a global static memory return from wsetlocale(). This currently
429 * leaks at process end; would require changing LOCALE_TERM to fix that */
430 Size_t string_size = wcslen(wresult) + 1;
432 if (wsetlocale_buf_size == 0) {
433 Newx(wsetlocale_buf, string_size, wchar_t);
434 wsetlocale_buf_size = string_size;
439 wsetlocale_buf_aTHX = aTHX;
444 else if (string_size > wsetlocale_buf_size) {
445 Renew(wsetlocale_buf, string_size, wchar_t);
446 wsetlocale_buf_size = string_size;
449 Copy(wresult, wsetlocale_buf, string_size, wchar_t);
452 return wsetlocale_buf;
455 # define _wsetlocale(category, wlocale) S_wsetlocale(category, wlocale)
457 #endif /* WIN32_USE_FAKE_OLD_MINGW_LOCALES */
459 /* 'for' loop headers to hide the necessary casts */
460 #define for_all_individual_category_indexes(i) \
461 for (locale_category_index i = (locale_category_index) 0; \
463 i = (locale_category_index) ((int) i + 1))
465 #define for_all_but_0th_individual_category_indexes(i) \
466 for (locale_category_index i = (locale_category_index) 1; \
468 i = (locale_category_index) ((int) i + 1))
470 #define for_all_category_indexes(i) \
471 for (locale_category_index i = (locale_category_index) 0; \
472 i <= LC_ALL_INDEX_; \
473 i = (locale_category_index) ((int) i + 1))
476 # if defined(USE_FAKE_LC_ALL_POSITIONAL_NOTATION) && defined(LC_ALL)
478 /* This simulates an underlying positional notation for LC_ALL when compiled on
479 * a system that uses name=value notation. Use this to develop on Linux and
480 * make a quick check that things have some chance of working on a positional
481 * box. Enable by adding to the Congfigure parameters:
482 * -Accflags=USE_FAKE_LC_ALL_POSITIONAL_NOTATION
484 * NOTE it redefines setlocale() and usequerylocale()
488 S_positional_name_value_xlation(const char * locale, bool direction)
489 { /* direction == 1 is from name=value to positional
490 direction == 0 is from positional to name=value */
494 const char * individ_locales[LC_ALL_INDEX_] = { NULL };
496 /* This parses either notation */
497 switch (parse_LC_ALL_string(locale,
498 (const char **) &individ_locales,
499 no_override, /* Handled by other code */
500 false, /* Return only [0] if suffices */
501 false, /* Don't panic on error */
504 default: /* Some compilers don't realize that below is the complete
505 list of the available enum values */
512 SAVEFREEPV(individ_locales[0]);
513 return individ_locales[0];
516 calc_LC_ALL_format format = (direction)
517 ? EXTERNAL_FORMAT_FOR_SET
519 const char * retval = calculate_LC_ALL_string(individ_locales,
524 for_all_individual_category_indexes(i) {
525 Safefree(individ_locales[i]);
534 S_positional_setlocale(int cat, const char * locale)
536 if (cat != LC_ALL) return setlocale(cat, locale);
538 if (locale && strNE(locale, "")) {
539 locale = S_positional_name_value_xlation(locale, 0);
540 if (! locale) return NULL;
543 locale = setlocale(cat, locale);
544 if (locale == NULL) return NULL;
545 return S_positional_name_value_xlation(locale, 1);
549 # define setlocale(a,b) S_positional_setlocale(a,b)
550 # ifdef USE_POSIX_2008_LOCALE
553 S_positional_newlocale(int mask, const char * locale, locale_t base)
557 if (mask != LC_ALL_MASK) return newlocale(mask, locale, base);
559 if (strNE(locale, "")) locale = S_positional_name_value_xlation(locale, 0);
560 if (locale == NULL) return NULL;
561 return newlocale(LC_ALL_MASK, locale, base);
565 # define newlocale(a,b,c) S_positional_newlocale(a,b,c)
568 #endif /* End of fake positional notation */
579 /* The main errno that gets used is this one, on platforms that support it */
581 # define SET_EINVAL SETERRNO(EINVAL, LIB_INVARG)
586 /* If we have any of these library functions, we can reliably determine is a
587 * locale is a UTF-8 one or not. And if we aren't using locales at all, we act
588 * as if everything is the C locale, so the answer there is always "No, it
589 * isn't UTF-8"; this too is reliably accurate */
590 #if defined(HAS_SOME_LANGINFO) || defined(HAS_MBTOWC) \
591 || defined(HAS_MBRTOWC) || ! defined(USE_LOCALE)
592 # define HAS_RELIABLE_UTF8NESS_DETERMINATION
595 /* This is a starting guess as to when this is true. It definititely isn't
596 * true on *BSD where positional LC_ALL notation is used. Likely this will end
597 * up being defined in hints files. */
598 #ifdef PERL_LC_ALL_USES_NAME_VALUE_PAIRS
599 # define NEWLOCALE_HANDLES_DISPARATE_LC_ALL
602 /* But regardless, we have to look at individual categories if some are
604 #ifdef HAS_IGNORED_LOCALE_CATEGORIES_
605 # undef NEWLOCALE_HANDLES_DISPARATE_LC_ALL
609 /* Not all categories need be set to the same locale. This macro determines if
610 * 'name' which represents LC_ALL is uniform or disparate. There are two
611 * situations: 1) the platform uses unordered name=value pairs; 2) the platform
612 * uses ordered positional values, with a separator string between them */
613 # ifdef PERL_LC_ALL_SEPARATOR /* positional */
614 # define is_disparate_LC_ALL(name) cBOOL(instr(name, PERL_LC_ALL_SEPARATOR))
615 # else /* name=value */
617 /* In the, hopefully never occurring, event that the platform doesn't use
618 * either mechanism for disparate LC_ALL's, assume the name=value pairs
619 * form, rather than taking the extreme step of refusing to compile. Many
620 * programs won't have disparate locales, so will generally work */
621 # define PERL_LC_ALL_SEPARATOR ";"
622 # define is_disparate_LC_ALL(name) cBOOL( strchr(name, ';') \
623 && strchr(name, '='))
626 /* It is possible to compile perl to always keep any individual category in the
627 * C locale. This would be done where the implementation on a platform is
628 * flawed or incomplete. At the time of this writing, for example, OpenBSD has
629 * not implemented LC_COLLATE beyond the C locale. The 'category_available[]'
630 * table is a bool that says whether a category is changeable, or must be kept
631 * in C. This macro substitutes C for the locale appropriately, expanding to
632 * nothing on the more typical case where all possible categories present on
633 * the platform are handled. */
634 # ifdef HAS_IGNORED_LOCALE_CATEGORIES_
635 # define need_to_override_category(i) (! category_available[i])
636 # define override_ignored_category(i, new_locale) \
637 ((need_to_override_category(i)) ? "C" : (new_locale))
639 # define need_to_override_category(i) 0
640 # define override_ignored_category(i, new_locale) (new_locale)
643 PERL_STATIC_INLINE const char *
644 S_mortalized_pv_copy(pTHX_ const char * const pv)
646 PERL_ARGS_ASSERT_MORTALIZED_PV_COPY;
648 /* Copies the input pv, and arranges for it to be freed at an unspecified
655 const char * copy = savepv(pv);
662 /* Default values come from the C locale */
663 #define C_codeset "ANSI_X3.4-1968" /* Only in some Configurations, and usually
664 a single instance, so is a #define */
665 static const char C_decimal_point[] = ".";
667 #if (defined(USE_LOCALE_NUMERIC) && ! defined(TS_W32_BROKEN_LOCALECONV)) \
668 || ! ( defined(USE_LOCALE_NUMERIC) \
669 && (defined(HAS_SOME_LANGINFO) || defined(HAS_LOCALECONV)))
670 static const char C_thousands_sep[] = "";
673 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
674 * return of setlocale(), then this is extremely likely to be the C or POSIX
675 * locale. However, the output of setlocale() is documented to be opaque, but
676 * the odds are extremely small that it would return these two strings for some
677 * other locale. Note that VMS includes many non-ASCII characters in these two
678 * locales as controls and punctuation (below are hex bytes):
680 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
681 * Oddly, none there are listed as alphas, though some represent alphabetics
682 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
683 #define isNAME_C_OR_POSIX(name) \
685 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
686 || strEQ((name), "POSIX")))
688 #if defined(HAS_NL_LANGINFO_L) || defined(HAS_NL_LANGINFO)
689 # define HAS_SOME_LANGINFO
692 #define my_langinfo_c(item, category, locale, retbufp, retbuf_sizep, utf8ness) \
693 my_langinfo_i(item, category##_INDEX_, locale, retbufp, \
694 retbuf_sizep, utf8ness)
697 # define setlocale_debug_string_i(index, locale, result) \
698 my_setlocale_debug_string_i(index, locale, result, __LINE__)
699 # define setlocale_debug_string_c(category, locale, result) \
700 setlocale_debug_string_i(category##_INDEX_, locale, result)
701 # define setlocale_debug_string_r(category, locale, result) \
702 setlocale_debug_string_i(get_category_index(category), \
706 # define toggle_locale_i(index, locale) \
707 S_toggle_locale_i(aTHX_ index, locale, __LINE__)
708 # define toggle_locale_c(cat, locale) toggle_locale_i(cat##_INDEX_, locale)
709 # define restore_toggled_locale_i(index, locale) \
710 S_restore_toggled_locale_i(aTHX_ index, locale, __LINE__)
711 # define restore_toggled_locale_c(cat, locale) \
712 restore_toggled_locale_i(cat##_INDEX_, locale)
714 /* On systems without LC_ALL, pretending it exists anyway simplifies things.
715 * Choose a value for it that is very unlikely to clash with any actual
717 # define FAKE_LC_ALL PERL_INT_MIN
719 /* Below are parallel arrays for locale information indexed by our mapping of
720 * category numbers into small non-negative indexes. locale_table.h contains
721 * an entry like this for each individual category used on this system:
722 * PERL_LOCALE_TABLE_ENTRY(CTYPE, S_new_ctype)
724 * Each array redefines PERL_LOCALE_TABLE_ENTRY to generate the information
725 * needed for that array, and #includes locale_table.h to get the valid
728 * An entry for the conglomerate category LC_ALL is added here, immediately
729 * following the individual categories. (The treatment for it varies, so can't
730 * be in locale_table.h.)
732 * Following this, each array ends with an entry for illegal categories. All
733 * category numbers unknown to perl get mapped to this entry. This is likely
734 * to be a parameter error from the calling program; but it could be that this
735 * platform has a category we don't know about, in which case it needs to be
736 * added, using the paradigm of one of the existing categories. */
738 /* The first array is the locale categories perl uses on this system, used to
739 * map our index back to the system's category number. */
740 STATIC const int categories[] = {
742 # undef PERL_LOCALE_TABLE_ENTRY
743 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) LC_ ## name,
744 # include "locale_table.h"
752 (FAKE_LC_ALL + 1) /* Entry for unknown category; this number is unlikely
753 to clash with a real category */
756 # define GET_NAME_AS_STRING(token) # token
757 # define GET_LC_NAME_AS_STRING(token) GET_NAME_AS_STRING(LC_ ## token)
759 /* The second array is the category names. */
760 STATIC const char * const category_names[] = {
762 # undef PERL_LOCALE_TABLE_ENTRY
763 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) GET_LC_NAME_AS_STRING(name),
764 # include "locale_table.h"
767 # define LC_ALL_STRING "LC_ALL"
769 # define LC_ALL_STRING "If you see this, it is a bug in perl;" \
770 " please report it via perlbug"
775 # define LC_UNKNOWN_STRING "Locale category unknown to Perl; if you see" \
776 " this, it is a bug in perl; please report it" \
781 STATIC const Size_t category_name_lengths[] = {
783 # undef PERL_LOCALE_TABLE_ENTRY
784 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) \
785 STRLENs(GET_LC_NAME_AS_STRING(name)),
786 # include "locale_table.h"
788 STRLENs(LC_ALL_STRING),
789 STRLENs(LC_UNKNOWN_STRING)
792 /* Each entry includes space for the '=' and ';' */
793 # undef PERL_LOCALE_TABLE_ENTRY
794 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) \
795 + STRLENs(GET_LC_NAME_AS_STRING(name)) + 2
797 STATIC const Size_t lc_all_boiler_plate_length = 1 /* space for trailing NUL */
798 # include "locale_table.h"
801 /* A few categories require additional setup when they are changed. This table
802 * points to the functions that do that setup */
803 STATIC void (*update_functions[]) (pTHX_ const char *, bool force) = {
805 # undef PERL_LOCALE_TABLE_ENTRY
806 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) call_back,
807 # include "locale_table.h"
810 NULL, /* No update for unknown category */
813 # if defined(HAS_IGNORED_LOCALE_CATEGORIES_)
815 /* Indicates if each category on this platform is available to use not in
817 STATIC const bool category_available[] = {
819 # undef PERL_LOCALE_TABLE_ENTRY
820 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) LC_ ## name ## _AVAIL_,
821 # include "locale_table.h"
829 false /* LC_UNKNOWN_AVAIL_ */
833 # if defined(USE_POSIX_2008_LOCALE)
835 STATIC const int category_masks[] = {
837 # undef PERL_LOCALE_TABLE_ENTRY
838 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) LC_ ## name ## _MASK,
839 # include "locale_table.h"
841 LC_ALL_MASK, /* Will rightly refuse to compile unless this is defined */
842 0 /* Empty mask for unknown category */
846 # if ! defined(PERL_LC_ALL_USES_NAME_VALUE_PAIRS)
848 /* On platforms that use positional notation for expressing LC_ALL, this maps
849 * the position of each category to our corresponding internal index for it.
850 * This is initialized at run time if needed. LC_ALL_INDEX_ is not legal for
851 * an individual locale, hence marks the elements here as not actually
855 map_LC_ALL_position_to_index[LC_ALL_INDEX_] = { LC_ALL_INDEX_ };
859 #if defined(USE_LOCALE) || defined(DEBUGGING)
862 S_get_displayable_string(pTHX_
863 const char * const s,
864 const char * const e,
867 PERL_ARGS_ASSERT_GET_DISPLAYABLE_STRING;
874 bool prev_was_printable = TRUE;
875 bool first_time = TRUE;
878 /* Worst case scenario: All are non-printable so have a blank between each.
879 * If UTF-8, all are the largest possible code point; otherwise all are a
880 * single byte. '(2 + 1)' is from each byte takes 2 characters to
881 * display, and a blank (or NUL for the final one) after it */
882 const Size_t size = (e - s) * (2 + 1) * ((is_utf8) ? UVSIZE : 1);
883 Newxz(ret, size, char);
888 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
891 if (! prev_was_printable) {
892 my_strlcat(ret, " ", size);
895 /* Escape these to avoid any ambiguity */
896 if (cp == ' ' || cp == '\\') {
897 my_strlcat(ret, "\\", size);
899 my_strlcat(ret, Perl_form(aTHX_ "%c", (U8) cp), size);
900 prev_was_printable = TRUE;
904 my_strlcat(ret, " ", size);
906 my_strlcat(ret, Perl_form(aTHX_ "%02" UVXf, cp), size);
907 prev_was_printable = FALSE;
909 t += (is_utf8) ? UTF8SKIP(t) : 1;
919 # define get_category_index(cat) get_category_index_helper(cat, NULL, __LINE__)
921 STATIC locale_category_index
922 S_get_category_index_helper(pTHX_ const int category, bool * succeeded,
923 const line_t caller_line)
925 PERL_ARGS_ASSERT_GET_CATEGORY_INDEX_HELPER;
927 /* Given a category, return the equivalent internal index we generally use
928 * instead, warn or panic if not found. */
930 locale_category_index i;
932 # undef PERL_LOCALE_TABLE_ENTRY
933 # define PERL_LOCALE_TABLE_ENTRY(name, call_back) \
934 case LC_ ## name: i = LC_ ## name ## _INDEX_; break;
938 # include "locale_table.h"
940 case LC_ALL: i = LC_ALL_INDEX_; break;
943 default: goto unknown_locale;
946 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
947 "index of category %d (%s) is %d;"
948 " called from %" LINE_Tf "\n",
949 category, category_names[i], i, caller_line));
961 return LC_ALL_INDEX_; /* Arbitrary */
964 locale_panic_via_(Perl_form(aTHX_ "Unknown locale category %d", category),
965 __FILE__, caller_line);
966 NOT_REACHED; /* NOTREACHED */
969 #endif /* ifdef USE_LOCALE */
972 Perl_force_locale_unlock(pTHX)
974 /* Remove any locale mutex, in preperation for an inglorious termination,
975 * typically a panic */
977 #if defined(USE_LOCALE_THREADS)
979 /* If recursively locked, clear all at once */
980 if (PL_locale_mutex_depth > 1) {
981 PL_locale_mutex_depth = 1;
984 if (PL_locale_mutex_depth > 0) {
992 #ifdef USE_POSIX_2008_LOCALE
995 S_use_curlocale_scratch(pTHX)
997 /* This function is used to hide from the caller the case where the current
998 * locale_t object in POSIX 2008 is the global one, which is illegal in
999 * many of the P2008 API calls. This checks for that and, if necessary
1000 * creates a proper P2008 object. Any prior object is deleted, as is any
1001 * remaining object during global destruction. */
1003 locale_t cur = uselocale((locale_t) 0);
1005 if (cur != LC_GLOBAL_LOCALE) {
1009 if (PL_scratch_locale_obj) {
1010 freelocale(PL_scratch_locale_obj);
1013 PL_scratch_locale_obj = duplocale(LC_GLOBAL_LOCALE);
1014 return PL_scratch_locale_obj;
1020 Perl_locale_panic(const char * msg,
1021 const line_t immediate_caller_line,
1022 const char * const higher_caller_file,
1023 const line_t higher_caller_line)
1025 PERL_ARGS_ASSERT_LOCALE_PANIC;
1029 force_locale_unlock();
1031 #ifdef USE_C_BACKTRACE
1032 dump_c_backtrace(Perl_debug_log, 20, 1);
1035 const char * called_by = "";
1036 if ( strNE(__FILE__, higher_caller_file)
1037 || immediate_caller_line != higher_caller_line)
1039 called_by = Perl_form(aTHX_ "\nCalled by %s: %" LINE_Tf "\n",
1040 higher_caller_file, higher_caller_line);
1045 const char * errno_text;
1047 #ifdef HAS_EXTENDED_OS_ERRNO
1049 const int extended_errnum = get_extended_os_errno();
1050 if (errno != extended_errnum) {
1051 errno_text = Perl_form(aTHX_ "; errno=%d, $^E=%d",
1052 errno, extended_errnum);
1059 errno_text = Perl_form(aTHX_ "; errno=%d", errno);
1062 /* diag_listed_as: panic: %s */
1063 Perl_croak(aTHX_ "%s: %" LINE_Tf ": panic: %s%s%s\n",
1064 __FILE__, immediate_caller_line,
1065 msg, errno_text, called_by);
1068 /* Macros to report and croak on an unexpected failure to set the locale. The
1069 * via version has more stack trace information */
1070 #define setlocale_failure_panic_i(i, cur, fail, line, higher_line) \
1071 setlocale_failure_panic_via_i(i, cur, fail, __LINE__, line, \
1072 __FILE__, higher_line)
1074 #define setlocale_failure_panic_c(cat, cur, fail, line, higher_line) \
1075 setlocale_failure_panic_i(cat##_INDEX_, cur, fail, line, higher_line)
1077 #if defined(USE_LOCALE)
1079 /* Expands to the code to
1080 * result = savepvn(s, len)
1081 * if the category whose internal index is 'i' doesn't need to be kept in the C
1082 * locale on this system, or if 'action is 'no_override'. Otherwise it expands
1084 * result = savepv("C")
1085 * unless 'action' isn't 'check_that_overridden', in which case if the string
1086 * 's' isn't already "C" it panics */
1087 # ifndef HAS_IGNORED_LOCALE_CATEGORIES_
1088 # define OVERRIDE_AND_SAVEPV(s, len, result, i, action) \
1089 result = savepvn(s, len)
1091 # define OVERRIDE_AND_SAVEPV(s, len, result, i, action) \
1093 if (LIKELY( ! need_to_override_category(i) \
1094 || action == no_override)) { \
1095 result = savepvn(s, len); \
1098 const char * temp = savepvn(s, len); \
1099 result = savepv(override_ignored_category(i, temp)); \
1100 if (action == check_that_overridden && strNE(result, temp)) { \
1101 locale_panic_(Perl_form(aTHX_ \
1102 "%s expected to be '%s', instead is '%s'", \
1103 category_names[i], result, temp)); \
1110 STATIC parse_LC_ALL_string_return
1111 S_parse_LC_ALL_string(pTHX_ const char * string,
1112 const char ** output,
1113 const parse_LC_ALL_STRING_action override,
1114 bool always_use_full_array,
1115 const bool panic_on_error,
1116 const line_t caller_line)
1118 /* This function parses the value of the input 'string' which is expected
1119 * to be the representation of an LC_ALL locale, and splits the result into
1120 * the values for the individual component categories, returning those in
1121 * the 'output' array. Each array value will be a savepv() copy that is
1122 * the responsibility of the caller to make sure gets freed
1124 * The locale for each category is independent of the other categories.
1125 * Often, they are all the same, but certainly not always. Perl, in fact,
1126 * usually keeps LC_NUMERIC in the C locale, regardless of the underlying
1127 * locale. LC_ALL has to be able to represent the case of when not all
1128 * categories have the same locale. Platforms have differing ways of
1129 * representing this. Internally, this file uses the 'name=value;'
1130 * representation found on some platforms, so this function always looks
1131 * for and parses that. Other platforms use a positional notation. On
1132 * those platforms, this function also parses that form. It examines the
1133 * input to see which form is being parsed.
1135 * Often, all categories will have the same locale. This is special cased
1136 * if 'always_use_full_array' is false on input:
1137 * 1) If the input 'string' is a single value, this function doesn't
1138 * store anything into 'output', and returns 'no_array'
1139 * 2) Some platforms will return multiple occurrences of the same
1140 * value rather than coalescing them down to a single one. HP-UX
1141 * is such a one. This function will do that collapsing for you,
1142 * returning 'only_element_0' and saving the single value in
1143 * output[0], which the caller will need to arrange to be freed.
1144 * The rest of output[] is undefined, and does not need to be
1147 * Otherwise, the input 'string' may not be valid. This function looks
1148 * mainly for syntactic errors, and if found, returns 'invalid'. 'output'
1149 * will not be filled in in that case, but the input state of it isn't
1150 * necessarily preserved. Turning on -DL debugging will give details as to
1151 * the error. If 'panic_on_error' is 'true', the function panics instead
1152 * of returning on error, with a message giving the details.
1154 * Otherwise, output[] will be filled with the individual locale names for
1155 * all categories on the system, 'full_array' will be returned, and the
1156 * caller needs to arrange for each to be freed. This means that either at
1157 * least one category differed from the others, or 'always_use_full_array' was
1160 * perl may be configured to ignore changes to a category's locale to
1161 * non-C. The parameter 'override' tells this function what to do when
1162 * encountering such an illegal combination:
1164 * no_override indicates to take no special action
1165 * override_if_ignored, indicates to return 'C' instead of what the
1166 * input string actually says.
1167 * check_that_overridden indicates to panic if the string says the
1168 * category is not 'C'. This is used when
1169 * non-C is very unexpected behavior.
1172 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1173 "Entering parse_LC_ALL_string; called from %" \
1174 LINE_Tf "\nnew='%s'\n", caller_line, string));
1176 # ifdef PERL_LC_ALL_USES_NAME_VALUE_PAIRS
1178 const char separator[] = ";";
1179 const Size_t separator_len = 1;
1180 const bool single_component = (strchr(string, ';') == NULL);
1184 /* It's possible (but quite unlikely) that the separator string is an '='
1185 * or a ';'. Requiring both to be present for using the 'name=value;' form
1186 * properly handles those possibilities */
1187 const bool name_value = strchr(string, '=') && strchr(string, ';');
1188 const char * separator;
1189 Size_t separator_len;
1190 bool single_component;
1194 single_component = false; /* Since has both [;=], must be multi */
1197 separator = PERL_LC_ALL_SEPARATOR;
1198 separator_len = STRLENs(PERL_LC_ALL_SEPARATOR);
1199 single_component = instr(string, separator) == NULL;
1202 Size_t component_number = 0; /* Position in the parsing loop below */
1205 # ifndef HAS_IGNORED_LOCALE_CATEGORIES_
1206 PERL_UNUSED_ARG(override);
1209 /* Any ignored categories are to be set to "C", so if this single-component
1210 * LC_ALL isn't to C, it has both "C" and non-C, so isn't really a single
1211 * component. All the non-ignored categories are set to the input
1212 * component, but the ignored ones are overridden to be C.
1214 * This incidentally handles the case where the string is "". The return
1215 * will be C for each ignored category and "" for the others. Then the
1216 * caller can individually set each category, and get the right answer. */
1217 if (single_component && ! isNAME_C_OR_POSIX(string)) {
1218 for_all_individual_category_indexes(i) {
1219 OVERRIDE_AND_SAVEPV(string, strlen(string), output[i], i, override);
1227 if (single_component) {
1228 if (! always_use_full_array) {
1232 for_all_individual_category_indexes(i) {
1233 output[i] = savepv(string);
1239 /* Here the input is multiple components. Parse through them. (It is
1240 * possible that these components are all the same, so we check, and if so,
1241 * return just the 0th component (unless 'always_use_full_array' is true)
1243 * This enum notes the possible errors findable in parsing */
1248 contains_LC_ALL_element
1251 /* Keep track of the categories we have encountered so far */
1252 bool seen[LC_ALL_INDEX_] = { false };
1254 Size_t index; /* Our internal index for the current category */
1255 const char * s = string;
1256 const char * e = s + strlen(string);
1257 const char * category_end = NULL;
1258 const char * saved_first = NULL;
1260 /* Parse the input locale string */
1263 /* 'separator' has been set up to delimit the components */
1264 const char * next_sep = instr(s, separator);
1265 if (! next_sep) { /* At the end of the input */
1269 # ifndef PERL_LC_ALL_USES_NAME_VALUE_PAIRS
1272 /* Get the index of the category in this position */
1273 index = map_LC_ALL_position_to_index[component_number++];
1279 { /* Get the category part when each component is the
1280 * 'category=locale' form */
1282 category_end = strchr(s, '=');
1284 /* The '=' terminates the category name. If no '=', is improper
1286 if (! category_end) {
1291 /* Find our internal index of the category name; uses a linear
1292 * search. (XXX This could be avoided by various means, but the
1293 * maximum likely search is 6 items, and khw doesn't think the
1294 * added complexity would save very much at all.) */
1295 const unsigned int name_len = (unsigned int) (category_end - s);
1296 for (index = 0; index < C_ARRAY_LENGTH(category_names); index++) {
1297 if ( name_len == category_name_lengths[index]
1298 && memEQ(s, category_names[index], name_len))
1300 goto found_category;
1304 /* Here, the category is not in our list. */
1305 error = unknown_category;
1308 found_category: /* The system knows about this category. */
1310 if (index == LC_ALL_INDEX_) {
1311 error = contains_LC_ALL_element;
1315 /* The locale name starts just beyond the '=' */
1316 s = category_end + 1;
1318 /* Linux (and maybe others) doesn't treat a duplicate category in
1319 * the string as an error. Instead it uses the final occurrence as
1320 * the intended value. So if this is a duplicate, free the former
1321 * value before setting the new one */
1323 Safefree(output[index]);
1330 /* Here, 'index' contains our internal index number for the current
1331 * category, and 's' points to the beginning of the locale name for
1333 OVERRIDE_AND_SAVEPV(s, next_sep - s, output[index], index, override);
1335 if (! always_use_full_array) {
1336 if (! saved_first) {
1337 saved_first = output[index];
1340 if (strNE(saved_first, output[index])) {
1341 always_use_full_array = true;
1346 /* Next time start from the new position */
1347 s = next_sep + separator_len;
1350 /* Finished looping through all the categories
1352 * Check if the input was incomplete. */
1354 # ifndef PERL_LC_ALL_USES_NAME_VALUE_PAIRS
1356 if (! name_value) { /* Positional notation */
1357 if (component_number != LC_ALL_INDEX_) {
1366 { /* Here is the name=value notation */
1367 for_all_individual_category_indexes(i) {
1375 /* In the loop above, we changed 'always_use_full_array' to true iff not all
1376 * categories have the same locale. Hence, if it is still 'false', all of
1377 * them are the same. */
1378 if (always_use_full_array) {
1382 /* Free the dangling ones */
1383 for_all_but_0th_individual_category_indexes(i) {
1384 Safefree(output[i]);
1388 return only_element_0;
1392 /* Don't leave memory dangling that we allocated before the failure */
1393 for_all_individual_category_indexes(i) {
1395 Safefree(output[i]);
1401 const char * display_start = s;
1402 const char * display_end = e;
1406 msg = "doesn't list every locale category";
1407 display_start = string;
1410 msg = "needs an '=' to split name=value";
1412 case unknown_category:
1413 msg = "is an unknown category";
1414 display_end = (category_end && category_end > display_start)
1418 case contains_LC_ALL_element:
1419 msg = "has LC_ALL, which is illegal here";
1423 msg = Perl_form(aTHX_ "'%.*s' %s\n",
1424 (int) (display_end - display_start),
1425 display_start, msg);
1427 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s", msg));
1429 if (panic_on_error) {
1430 locale_panic_via_(msg, __FILE__, caller_line);
1436 # undef OVERRIDE_AND_SAVEPV
1439 /*==========================================================================
1440 * Here starts the code that gives a uniform interface to its callers, hiding
1441 * the differences between platforms.
1443 * base_posix_setlocale_() presents a consistent POSIX-compliant interface to
1444 * setlocale(). Windows requres a customized base-level setlocale(). This
1445 * layer should only be used by the next level up: the plain posix_setlocale
1446 * layer. Any necessary mutex locking needs to be done at a higher level. The
1447 * return may be overwritten by the next call to this function */
1449 # define base_posix_setlocale_(cat, locale) win32_setlocale(cat, locale)
1451 # define base_posix_setlocale_(cat, locale) \
1452 ((const char *) setlocale(cat, locale))
1455 /*==========================================================================
1456 * Here is the main posix layer. It is the same as the base one unless the
1457 * system is lacking LC_ALL, or there are categories that we ignore, but that
1458 * the system libc knows about */
1460 #if ! defined(USE_LOCALE) \
1461 || (defined(LC_ALL) && ! defined(HAS_IGNORED_LOCALE_CATEGORIES_))
1462 # define posix_setlocale(cat, locale) base_posix_setlocale_(cat, locale)
1464 # define posix_setlocale(cat, locale) \
1465 S_posix_setlocale_with_complications(aTHX_ cat, locale, __LINE__)
1468 S_posix_setlocale_with_complications(pTHX_ const int cat,
1469 const char * new_locale,
1470 const line_t caller_line)
1472 /* This implements the posix layer above the base posix layer.
1473 * It is needed to reconcile our internal records that reflect only a
1474 * proper subset of the categories known by the system. */
1476 /* Querying the current locale returns the real value */
1477 if (new_locale == NULL) {
1478 new_locale = base_posix_setlocale_(cat, NULL);
1483 const char * locale_on_entry = NULL;
1485 /* If setting from the environment, actually do the set to get the system's
1486 * idea of what that means; we may have to override later. */
1487 if (strEQ(new_locale, "")) {
1488 locale_on_entry = base_posix_setlocale_(cat, NULL);
1489 assert(locale_on_entry);
1490 new_locale = base_posix_setlocale_(cat, "");
1499 const char * new_locales[LC_ALL_INDEX_] = { NULL };
1501 if (cat == LC_ALL) {
1502 switch (parse_LC_ALL_string(new_locale,
1503 (const char **) &new_locales,
1504 override_if_ignored, /* Override any
1507 false, /* Return only [0] if suffices */
1508 false, /* Don't panic on error */
1518 case only_element_0:
1519 new_locale = new_locales[0];
1520 SAVEFREEPV(new_locale);
1525 /* Turn the array into a string that the libc setlocale() should
1526 * understand. (Another option would be to loop, setting the
1527 * individual locales, and then return base(cat, NULL) */
1528 new_locale = calculate_LC_ALL_string(new_locales,
1529 EXTERNAL_FORMAT_FOR_SET,
1533 for_all_individual_category_indexes(i) {
1534 Safefree(new_locales[i]);
1537 /* And call the libc setlocale. We could avoid this call if
1538 * locale_on_entry is set and eq the new_locale. But that would be
1539 * only for the relatively rare case of the desired locale being
1540 * "", and the time spent in doing the string compare might be more
1541 * than that of just setting it unconditionally */
1542 new_locale = base_posix_setlocale_(cat, new_locale);
1553 /* Here, 'new_locale' is a single value, not an aggregation. Just set it.
1556 base_posix_setlocale_(cat,
1557 override_ignored_category(
1558 get_category_index(cat), new_locale));
1567 /* 'locale_on_entry' being set indicates there has likely been a change in
1568 * locale which needs to be restored */
1569 if (locale_on_entry) {
1570 if (! base_posix_setlocale_(cat, locale_on_entry)) {
1571 setlocale_failure_panic_i(get_category_index(cat),
1572 NULL, locale_on_entry,
1573 __LINE__, caller_line);
1583 /* End of posix layer
1584 *==========================================================================
1586 * The next layer up is to catch vagaries and bugs in the libc setlocale return
1587 * value. The return is not guaranteed to be stable.
1589 * Any necessary mutex locking needs to be done at a higher level.
1591 * On most platforms this layer is empty, expanding to just the layer
1592 * below. To enable it, call Configure with either or both:
1593 * -Accflags=-DHAS_LF_IN_SETLOCALE_RETURN
1594 * to indicate that extraneous \n characters can be returned
1596 * -Accflags=-DHAS_BROKEN_SETLOCALE_QUERY_LC_ALL
1597 * to indicate that setlocale(LC_ALL, NULL) cannot be relied
1601 #define STDIZED_SETLOCALE_LOCK POSIX_SETLOCALE_LOCK
1602 #define STDIZED_SETLOCALE_UNLOCK POSIX_SETLOCALE_UNLOCK
1603 #if ! defined(USE_LOCALE) \
1604 || ! ( defined(HAS_LF_IN_SETLOCALE_RETURN) \
1605 || defined(HAS_BROKEN_SETLOCALE_QUERY_LC_ALL))
1606 # define stdized_setlocale(cat, locale) posix_setlocale(cat, locale)
1607 # define stdize_locale(cat, locale) (locale)
1609 # define stdized_setlocale(cat, locale) \
1610 S_stdize_locale(aTHX_ cat, posix_setlocale(cat, locale), __LINE__)
1613 S_stdize_locale(pTHX_ const int category,
1614 const char *input_locale,
1615 const line_t caller_line)
1617 /* The return value of setlocale() is opaque, but is required to be usable
1618 * as input to a future setlocale() to create the same state.
1619 * Unfortunately not all systems are compliant. This function brings those
1620 * outliers into conformance. It is based on what problems have arisen in
1623 * This has similar constraints as the posix layer. You need to lock
1624 * around it until its return is safely copied or no longer needed. (The
1625 * return may point to a global static buffer or may be mortalized.)
1627 * The current things this corrects are:
1628 * 1) A new-line. This function chops any \n characters
1629 * 2) A broken 'setlocale(LC_ALL, foo)' This constructs a proper returned
1630 * string from the constituent categories
1632 * If no changes were made, the input is returned as-is */
1634 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1635 "Entering stdize_locale(%d, '%s');"
1636 " called from %" LINE_Tf "\n",
1637 category, input_locale, caller_line));
1639 if (input_locale == NULL) {
1644 char * retval = (char *) input_locale;
1646 # if defined(LC_ALL) && defined(HAS_BROKEN_SETLOCALE_QUERY_LC_ALL)
1648 /* If setlocale(LC_ALL, NULL) is broken, compute what the system
1649 * actually thinks it should be from its individual components */
1650 if (category == LC_ALL) {
1651 retval = (char *) calculate_LC_ALL_string(
1652 NULL, /* query each individ locale */
1653 EXTERNAL_FORMAT_FOR_SET,
1659 # ifdef HAS_NL_IN_SETLOCALE_RETURN
1661 char * first_bad = NULL;
1665 PERL_UNUSED_ARG(category);
1666 PERL_UNUSED_ARG(caller_line);
1668 # define INPUT_LOCALE retval
1669 # define MARK_CHANGED
1672 char * individ_locales[LC_ALL_INDEX_] = { NULL };
1673 bool made_changes = false;
1675 if (category != LC_ALL) {
1676 individ_locales[0] = retval;
1681 /* And parse the locale string, splitting into its individual
1683 switch (parse_LC_ALL_string(retval,
1684 (const char **) &individ_locales,
1685 check_that_overridden, /* ignored
1689 false, /* Return only [0] if suffices */
1690 false, /* Don't panic on error */
1697 case full_array: /* Loop below through all the component categories.
1699 upper = LC_ALL_INDEX_ - 1;
1703 /* All categories here are set to the same locale, and the parse
1704 * didn't fill in any of 'individ_locales'. Set the 0th element to
1706 individ_locales[0] = retval;
1709 case only_element_0: /* Element 0 is the only element we need to look
1716 for (unsigned int i = 0; i <= upper; i++)
1718 # define INPUT_LOCALE individ_locales[i]
1719 # define MARK_CHANGED made_changes = true;
1720 # endif /* Has LC_ALL */
1723 first_bad = (char *) strchr(INPUT_LOCALE, '\n');
1725 /* Most likely, there isn't a problem with the input */
1726 if (UNLIKELY(first_bad)) {
1728 /* This element will need to be adjusted. Create a modifiable
1731 retval = savepv(INPUT_LOCALE);
1734 /* Translate the found position into terms of the copy */
1735 first_bad = retval + (first_bad - INPUT_LOCALE);
1737 /* Get rid of the \n and what follows. (Originally, only a
1738 * trailing \n was stripped. Unsure what to do if not trailing) */
1739 *((char *) first_bad) = '\0';
1740 } /* End of needs adjusting */
1741 } /* End of looking for problems */
1745 /* If we had multiple elements, extra work is required */
1748 /* If no changes were made to the input, 'retval' already contains it
1752 /* But if did make changes, need to calculate the new value */
1753 retval = (char *) calculate_LC_ALL_string(
1754 (const char **) &individ_locales,
1755 EXTERNAL_FORMAT_FOR_SET,
1760 /* And free the no-longer needed memory */
1761 for (unsigned int i = 0; i <= upper; i++) {
1762 Safefree(individ_locales[i]);
1767 # undef INPUT_LOCALE
1768 # undef MARK_CHANGED
1769 # endif /* HAS_NL_IN_SETLOCALE_RETURN */
1771 return (const char *) retval;
1774 #endif /* USE_LOCALE */
1776 /* End of stdize_locale layer
1778 * ==========================================================================
1780 * The next many lines form several implementations of a layer above the
1781 * close-to-the-metal 'posix' and 'stdized' macros. They are used to present a
1782 * uniform API to the rest of the code in this file in spite of the disparate
1783 * underlying implementations. Which implementation gets compiled depends on
1784 * the platform capabilities (and some user choice) as determined by Configure.
1786 * As more fully described in the introductory comments in this file, the
1787 * API of each implementation consists of three sets of macros. Each set has
1788 * three variants with suffixes '_c', '_r', and '_i'. In the list below '_X'
1789 * is to be replaced by any of these suffixes.
1791 * 1) bool_setlocale_X attempts to set the given category's locale to the
1792 * given value, returning if it worked or not.
1793 * 2) void_setlocale_X is like the corresponding bool_setlocale, but used when
1794 * success is the only sane outcome, so failure causes it
1796 * 3) querylocale_X to see what the given category's locale is
1798 * 4) setlocale_i() is defined only in those implementations where the bool
1799 * and query forms are essentially the same, and can be
1800 * combined to save CPU time.
1802 * Each implementation below is separated by ==== lines, and includes bool,
1803 * void, and query macros. The query macros are first, followed by any
1804 * functions needed to implement them. Then come the bool, again followed by
1805 * any implementing functions Then are the void macros; next is setlocale_i if
1806 * present on this implementation. Finally are any helper functions. The sets
1807 * in each implementation are separated by ---- lines.
1809 * The returned strings from all the querylocale...() forms in all
1810 * implementations are thread-safe, and the caller should not free them,
1811 * but each may be a mortalized copy. If you need something stable across
1812 * calls, you need to savepv() the result yourself.
1814 *===========================================================================*/
1816 #if (! defined(USE_LOCALE_THREADS) && ! defined(USE_POSIX_2008_LOCALE)) \
1817 || ( defined(WIN32) && defined(USE_THREAD_SAFE_LOCALE))
1819 /* For non-threaded perls, the implementation just expands to the base-level
1820 * functions (except if we are Configured to nonetheless use the POSIX 2008
1821 * interface) This implementation is also used on threaded perls where
1822 * threading is invisible to us. Currently this is only on later Windows
1825 # define querylocale_r(cat) mortalized_pv_copy(stdized_setlocale(cat, NULL))
1826 # define querylocale_c(cat) querylocale_r(cat)
1827 # define querylocale_i(i) querylocale_c(categories[i])
1829 /*---------------------------------------------------------------------------*/
1831 # define bool_setlocale_r(cat, locale) cBOOL(posix_setlocale(cat, locale))
1832 # define bool_setlocale_i(i, locale) \
1833 bool_setlocale_c(categories[i], locale)
1834 # define bool_setlocale_c(cat, locale) bool_setlocale_r(cat, locale)
1836 /*---------------------------------------------------------------------------*/
1838 # define void_setlocale_r_with_caller(cat, locale, file, line) \
1840 if (! bool_setlocale_r(cat, locale)) \
1841 setlocale_failure_panic_via_i(get_category_index(cat), \
1842 NULL, locale, __LINE__, 0, \
1846 # define void_setlocale_c_with_caller(cat, locale, file, line) \
1847 void_setlocale_r_with_caller(cat, locale, file, line)
1849 # define void_setlocale_i_with_caller(i, locale, file, line) \
1850 void_setlocale_r_with_caller(categories[i], locale, file, line)
1852 # define void_setlocale_r(cat, locale) \
1853 void_setlocale_r_with_caller(cat, locale, __FILE__, __LINE__)
1854 # define void_setlocale_c(cat, locale) void_setlocale_r(cat, locale)
1855 # define void_setlocale_i(i, locale) void_setlocale_r(categories[i], locale)
1857 /*---------------------------------------------------------------------------*/
1859 /* setlocale_i is only defined for Configurations where the libc setlocale()
1860 * doesn't need any tweaking. It allows for some shortcuts */
1861 # ifndef USE_LOCALE_THREADS
1862 # define setlocale_i(i, locale) stdized_setlocale(categories[i], locale)
1864 # elif defined(WIN32) && defined(USE_THREAD_SAFE_LOCALE)
1866 /* On Windows, we don't know at compile time if we are in thread-safe mode or
1867 * not. If we are, we can just return the result of the layer below us. If we
1868 * are in unsafe mode, we need to first copy that result to a safe place while
1869 * in a critical section */
1871 # define setlocale_i(i, locale) S_setlocale_i(aTHX_ categories[i], locale)
1874 S_setlocale_i(pTHX_ const int category, const char * locale)
1876 if (LIKELY(_configthreadlocale(0) == _ENABLE_PER_THREAD_LOCALE)) {
1877 return stdized_setlocale(category, locale);
1881 const char * retval = save_to_buffer(stdized_setlocale(category, locale),
1883 &PL_setlocale_bufsize);
1891 /*===========================================================================*/
1892 #elif defined(USE_LOCALE_THREADS) \
1893 && ! defined(USE_THREAD_SAFE_LOCALE)
1895 /* Here, there are threads, and there is no support for thread-safe
1896 * operation. This is a dangerous situation, which perl is documented as
1897 * not supporting, but it arises in practice. We can do a modicum of
1898 * automatic mitigation by making sure there is a per-thread return from
1899 * setlocale(), and that a mutex protects it from races */
1901 # define querylocale_r(cat) \
1902 mortalized_pv_copy(less_dicey_setlocale_r(cat, NULL))
1903 # define querylocale_c(cat) querylocale_r(cat)
1904 # define querylocale_i(i) querylocale_r(categories[i])
1907 S_less_dicey_setlocale_r(pTHX_ const int category, const char * locale)
1909 const char * retval;
1911 PERL_ARGS_ASSERT_LESS_DICEY_SETLOCALE_R;
1913 STDIZED_SETLOCALE_LOCK;
1915 retval = save_to_buffer(stdized_setlocale(category, locale),
1916 &PL_less_dicey_locale_buf,
1917 &PL_less_dicey_locale_bufsize);
1919 STDIZED_SETLOCALE_UNLOCK;
1924 /*---------------------------------------------------------------------------*/
1926 # define bool_setlocale_r(cat, locale) \
1927 less_dicey_bool_setlocale_r(cat, locale)
1928 # define bool_setlocale_i(i, locale) \
1929 bool_setlocale_r(categories[i], locale)
1930 # define bool_setlocale_c(cat, locale) bool_setlocale_r(cat, locale)
1933 S_less_dicey_bool_setlocale_r(pTHX_ const int cat, const char * locale)
1937 PERL_ARGS_ASSERT_LESS_DICEY_BOOL_SETLOCALE_R;
1939 /* Unlikely, but potentially possible that another thread could zap the
1940 * buffer from true to false or vice-versa, so need to lock here */
1941 POSIX_SETLOCALE_LOCK;
1942 retval = cBOOL(posix_setlocale(cat, locale));
1943 POSIX_SETLOCALE_UNLOCK;
1948 /*---------------------------------------------------------------------------*/
1950 # define void_setlocale_r_with_caller(cat, locale, file, line) \
1952 if (! bool_setlocale_r(cat, locale)) \
1953 setlocale_failure_panic_via_i(get_category_index(cat), \
1954 NULL, locale, __LINE__, 0, \
1958 # define void_setlocale_c_with_caller(cat, locale, file, line) \
1959 void_setlocale_r_with_caller(cat, locale, file, line)
1961 # define void_setlocale_i_with_caller(i, locale, file, line) \
1962 void_setlocale_r_with_caller(categories[i], locale, file, line)
1964 # define void_setlocale_r(cat, locale) \
1965 void_setlocale_r_with_caller(cat, locale, __FILE__, __LINE__)
1966 # define void_setlocale_c(cat, locale) void_setlocale_r(cat, locale)
1967 # define void_setlocale_i(i, locale) void_setlocale_r(categories[i], locale)
1969 /*===========================================================================*/
1971 #elif defined(USE_POSIX_2008_LOCALE)
1973 # error This code assumes that LC_ALL is available on a system modern enough to have POSIX 2008
1976 /* Here, there is a completely different API to get thread-safe locales. We
1977 * emulate the setlocale() API with our own function(s). setlocale categories,
1978 * like LC_NUMERIC, are not valid here for the POSIX 2008 API. Instead, there
1979 * are equivalents, like LC_NUMERIC_MASK, which we use instead, which we find
1980 * by table lookup. */
1982 # if defined(__GLIBC__) && defined(USE_LOCALE_MESSAGES)
1983 /* https://sourceware.org/bugzilla/show_bug.cgi?id=24936 */
1984 # define HAS_GLIBC_LC_MESSAGES_BUG
1985 # include <libintl.h>
1988 # define querylocale_i(i) querylocale_2008_i(i, __LINE__)
1989 # define querylocale_c(cat) querylocale_i(cat##_INDEX_)
1990 # define querylocale_r(cat) querylocale_i(get_category_index(cat))
1993 S_querylocale_2008_i(pTHX_ const locale_category_index index,
1994 const line_t caller_line)
1996 PERL_ARGS_ASSERT_QUERYLOCALE_2008_I;
1997 assert(index <= LC_ALL_INDEX_);
1999 /* This function returns the name of the locale category given by the input
2000 * 'index' into our parallel tables of them.
2002 * POSIX 2008, for some sick reason, chose not to provide a method to find
2003 * the category name of a locale, disregarding a basic linguistic tenet
2004 * that for any object, people will create a name for it. (The next
2005 * version of the POSIX standard is proposed to fix this.) Some vendors
2006 * have created a querylocale() function to do this in the meantime. On
2007 * systems without querylocale(), we have to keep track of what the locale
2008 * has been set to, so that we can return its name so as to emulate
2009 * setlocale(). There are potential problems with this:
2011 * 1) We don't know what calling newlocale() with the locale argument ""
2012 * actually does. It gets its values from the program's environment.
2013 * find_locale_from_environment() is used to work around this. But it
2014 * isn't fool-proof. See the comments for that function for details.
2015 * 2) It's possible for C code in some library to change the locale
2016 * without us knowing it, and thus our records become wrong;
2017 * querylocale() would catch this. But as of September 2017, there
2018 * are no occurrences in CPAN of uselocale(). Some libraries do use
2019 * setlocale(), but that changes the global locale, and threads using
2020 * per-thread locales will just ignore those changes.
2021 * 3) Many systems have multiple names for the same locale. Generally,
2022 * there is an underlying base name, with aliases that evaluate to it.
2023 * On some systems, if you set the locale to an alias, and then
2024 * retrieve the name, you get the alias as expected; but on others you
2025 * get the base name, not the alias you used. And sometimes the
2026 * charade is incomplete. See
2027 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=269375.
2029 * The code is structured so that the returned locale name when the
2030 * locale is changed is whatever the result of querylocale() on the
2031 * new locale is. This effectively gives the result the system
2032 * expects. Without querylocale, the name returned is always the
2033 * input name. Theoretically this could cause problems, but khw knows
2034 * of none so far, but mentions it here in case you are trying to
2035 * debug something. (This could be worked around by messing with the
2036 * global locale temporarily, using setlocale() to get the base name;
2037 * but that could cause a race. The comments for
2038 * find_locale_from_environment() give details on the potential race.)
2041 const locale_t cur_obj = uselocale((locale_t) 0);
2042 const char * retval;
2044 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "querylocale_2008_i(%s) on %p;"
2045 " called from %" LINE_Tf "\n",
2046 category_names[index], cur_obj,
2049 if (UNLIKELY(cur_obj == LC_GLOBAL_LOCALE)) {
2051 /* Even on platforms that have querylocale(), it is unclear if they
2052 * work in the global locale, and we have the means to get the correct
2053 * answer anyway. khw is unsure this situation even comes up these
2054 * days, hence the branch prediction */
2055 POSIX_SETLOCALE_LOCK;
2056 retval = mortalized_pv_copy(posix_setlocale(categories[index], NULL));
2057 POSIX_SETLOCALE_UNLOCK;
2060 /* Here we have handled the case of the the current locale being the global
2061 * one. Below is the 'else' case of that. There are two different
2062 * implementations, depending on USE_PL_CURLOCALES */
2064 # ifdef USE_PL_CURLOCALES
2068 /* PL_curlocales[] is kept up-to-date for all categories except LC_ALL,
2069 * which may have been invalidated by setting it to NULL, and if so,
2070 * should now be calculated. (The called function updates that
2072 if (index == LC_ALL_INDEX_ && PL_curlocales[LC_ALL_INDEX_] == NULL) {
2073 calculate_LC_ALL_string((const char **) &PL_curlocales,
2079 if (cur_obj == PL_C_locale_obj) {
2081 /* If the current locale object is the C object, then the answer is
2082 * "C" or POSIX, regardless of the category. Handling this
2083 * reasonably likely case specially shortcuts extra effort, and
2084 * hides some bugs from us in OS's that alias other locales to C,
2085 * but do so incompletely. If our records say it is POSIX, use
2086 * that; otherwise use C. See
2087 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=269375 */
2088 retval = mortalized_pv_copy((strEQ(PL_curlocales[index], "POSIX"))
2093 retval = mortalized_pv_copy(PL_curlocales[index]);
2099 /* Below is the implementation of the 'else' clause which handles the case
2100 * of the current locale not being the global one on platforms where
2101 * USE_PL_CURLOCALES is NOT in effect. That means the system must have
2102 * some form of querylocale. But these have varying characteristics, so
2103 * first create some #defines to make the actual 'else' clause uniform.
2105 * First, glibc has a function that implements querylocale(), but is called
2106 * something else, and takes the category number; the others take the mask.
2108 # if defined(USE_QUERYLOCALE) && ( defined(_NL_LOCALE_NAME) \
2109 && defined(HAS_NL_LANGINFO_L))
2110 # define my_querylocale(index, cur_obj) \
2111 nl_langinfo_l(_NL_LOCALE_NAME(categories[index]), cur_obj)
2113 /* Experience so far shows it is thread-safe, as well as glibc's
2114 * nl_langinfo_l(), so unless overridden, mark it so */
2115 # ifdef NO_THREAD_SAFE_QUERYLOCALE
2116 # undef HAS_THREAD_SAFE_QUERYLOCALE
2118 # define HAS_THREAD_SAFE_QUERYLOCALE
2120 # else /* below, ! glibc */
2122 /* Otherwise, use the system's querylocale(). */
2123 # define my_querylocale(index, cur_obj) \
2124 querylocale(category_masks[index], cur_obj)
2126 /* There is no standard for this function, and khw has never seen
2127 * anything beyond minimal vendor documentation, lacking important
2128 * details. Experience has shown that some implementations have race
2129 * condiions, and their returns may not be thread safe. It would be
2130 * unreliable to test for complete thread safety in Configure. What we
2131 * do instead is to assume that it is thread-safe, unless overriden by,
2132 * say, a hints file specifying
2133 * -Accflags='-DNO_THREAD_SAFE_QUERYLOCALE */
2134 # ifdef NO_THREAD_SAFE_QUERYLOCALE
2135 # undef HAS_THREAD_SAFE_QUERYLOCALE
2137 # define HAS_THREAD_SAFE_QUERYLOCALE
2141 /* Here, we have set up enough information to know if this querylocale()
2142 * is thread-safe, or needs to use a mutex */
2143 # ifdef HAS_THREAD_SAFE_QUERYLOCALE
2144 # define QUERYLOCALE_LOCK
2145 # define QUERYLOCALE_UNLOCK
2147 # define QUERYLOCALE_LOCK gwLOCALE_LOCK
2148 # define QUERYLOCALE_UNLOCK gwLOCALE_UNLOCK
2151 /* Finally, everything is ready, so here is the 'else' clause to implement
2152 * the case of the current locale not being the global one on systems that
2153 * have some form of querylocale(). (POSIX will presumably eventually
2154 * publish their next version in their pipeline, which will define a
2155 * precisely specified querylocale equivalent, and there can be a new
2156 * #ifdef to use it without having to guess at its characteristics) */
2159 /* We don't keep records when there is querylocale(), so as to avoid the
2160 * pitfalls mentioned at the beginning of this function.
2162 * That means LC_ALL has to be calculated from all its constituent
2163 * categories each time, since the querylocale() forms on many (if not
2164 * all) platforms only work on individual categories */
2165 if (index == LC_ALL_INDEX_) {
2166 retval = calculate_LC_ALL_string(NULL, INTERNAL_FORMAT,
2173 retval = savepv(my_querylocale(index, cur_obj));
2176 /* querylocale() may conflate the C locale with something that
2177 * isn't exactly the same. See for example
2178 * https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=269375
2179 * We know that if the locale object is the C one, we
2180 * are in the C locale, which may go by the name POSIX, as both, by
2181 * definition, are equivalent. But we consider any other name
2182 * spurious, so override with "C". As in the PL_CURLOCALES case
2183 * above, this hides those glitches, for the most part, from the
2184 * rest of our code. (The code is ordered this way so that if the
2185 * system distinugishes "C" from "POSIX", we do too.) */
2186 if (cur_obj == PL_C_locale_obj && ! isNAME_C_OR_POSIX(retval)) {
2188 retval = savepv("C");
2195 # undef QUERYLOCALE_LOCK
2196 # undef QUERYLOCALE_UNLOCK
2199 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2200 "querylocale_2008_i(%s) returning '%s'\n",
2201 category_names[index], retval));
2202 assert(strNE(retval, ""));
2206 /*---------------------------------------------------------------------------*/
2208 # define bool_setlocale_i(i, locale) \
2209 bool_setlocale_2008_i(i, locale, __LINE__)
2210 # define bool_setlocale_c(cat, locale) \
2211 bool_setlocale_i(cat##_INDEX_, locale)
2212 # define bool_setlocale_r(cat, locale) \
2213 bool_setlocale_i(get_category_index(cat), locale)
2215 /* If this doesn't exist on this platform, make it a no-op (to save #ifdefs) */
2216 # ifndef update_PL_curlocales_i
2217 # define update_PL_curlocales_i(index, new_locale, caller_line)
2221 S_bool_setlocale_2008_i(pTHX_
2223 /* Our internal index of the 'category' setlocale is called with */
2224 const locale_category_index index,
2225 const char * new_locale, /* The locale to set the category to */
2226 const line_t caller_line /* Called from this line number */
2229 PERL_ARGS_ASSERT_BOOL_SETLOCALE_2008_I;
2230 assert(index <= LC_ALL_INDEX_);
2232 /* This function effectively performs a setlocale() on just the current
2233 * thread; thus it is thread-safe. It does this by using the POSIX 2008
2234 * locale functions to emulate the behavior of setlocale(). Similar to
2235 * regular setlocale(), the return from this function points to memory that
2236 * can be overwritten by other system calls, so needs to be copied
2237 * immediately if you need to retain it. The difference here is that
2238 * system calls besides another setlocale() can overwrite it.
2240 * By doing this, most locale-sensitive functions become thread-safe. The
2241 * exceptions are mostly those that return a pointer to static memory.
2244 int mask = category_masks[index];
2245 const locale_t entry_obj = uselocale((locale_t) 0);
2246 const char * locale_on_entry = querylocale_i(index);
2248 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2249 "bool_setlocale_2008_i: input=%d (%s), mask=0x%x,"
2250 " new locale=\"%s\", current locale=\"%s\","
2251 " index=%d, entry object=%p;"
2252 " called from %" LINE_Tf "\n",
2253 categories[index], category_names[index], mask,
2254 ((new_locale == NULL) ? "(nil)" : new_locale),
2255 locale_on_entry, index, entry_obj, caller_line));
2257 /* Here, trying to change the locale, but it is a no-op if the new boss is
2258 * the same as the old boss. Except this routine is called when converting
2259 * from the global locale, so in that case we will create a per-thread
2260 * locale below (with the current values). It also seemed that newlocale()
2261 * could free up the basis locale memory if we called it with the new and
2262 * old being the same, but khw now thinks that this was due to some other
2263 * bug, since fixed, as there are other places where newlocale() gets
2264 * similarly called without problems. */
2265 if ( entry_obj != LC_GLOBAL_LOCALE
2267 && strEQ(new_locale, locale_on_entry))
2269 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2270 "bool_setlocale_2008_i: no-op to change to"
2271 " what it already was\n"));
2275 # ifndef USE_QUERYLOCALE
2277 /* Without a querylocale() mechanism, we have to figure out ourselves what
2278 * happens with setting a locale to "" */
2280 if (strEQ(new_locale, "")) {
2281 new_locale = find_locale_from_environment(index);
2289 # ifdef NEWLOCALE_HANDLES_DISPARATE_LC_ALL
2291 const bool need_loop = false;
2295 bool need_loop = false;
2296 const char * new_locales[LC_ALL_INDEX_] = { NULL };
2298 /* If we're going to have to parse the LC_ALL string, might as well do it
2299 * now before we have made changes that we would have to back out of if the
2301 if (index == LC_ALL_INDEX_) {
2302 switch (parse_LC_ALL_string(new_locale,
2303 (const char **) &new_locales,
2304 override_if_ignored,
2305 false, /* Return only [0] if suffices */
2306 false, /* Don't panic on error */
2317 case only_element_0:
2318 SAVEFREEPV(new_locales[0]);
2319 new_locale = new_locales[0];
2330 # ifdef HAS_GLIBC_LC_MESSAGES_BUG
2332 /* For this bug, if the LC_MESSAGES locale changes, we have to do an
2333 * expensive workaround. Save the current value so we can later determine
2335 const char * old_messages_locale = NULL;
2336 if ( (index == LC_MESSAGES_INDEX_ || index == LC_ALL_INDEX_)
2337 && LIKELY(PL_phase != PERL_PHASE_CONSTRUCT))
2339 old_messages_locale = querylocale_c(LC_MESSAGES);
2344 assert(PL_C_locale_obj);
2346 /* Now ready to switch to the input 'new_locale' */
2348 /* Switching locales generally entails freeing the current one's space (at
2349 * the C library's discretion), hence we can't be using that locale at the
2350 * time of the switch (this wasn't obvious to khw from the man pages). So
2351 * switch to a known locale object that we don't otherwise mess with. */
2352 if (! uselocale(PL_C_locale_obj)) {
2354 /* Not being able to change to the C locale is severe; don't keep
2356 setlocale_failure_panic_i(index, locale_on_entry, "C",
2357 __LINE__, caller_line);
2358 NOT_REACHED; /* NOTREACHED */
2361 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2362 "bool_setlocale_2008_i: now using C"
2363 " object=%p\n", PL_C_locale_obj));
2365 /* These two objects are special:
2366 * LC_GLOBAL_LOCALE because it is undefined behavior to call
2367 * newlocale() with it as a parameter.
2368 * PL_C_locale_obj because newlocale() generally destroys its locale
2369 * object parameter when it succeeds; and we don't
2370 * want that happening to this immutable object.
2371 * Copies will be made for them to use instead if we get so far as to call
2373 bool entry_obj_is_special = ( entry_obj == LC_GLOBAL_LOCALE
2374 || entry_obj == PL_C_locale_obj);
2377 /* PL_C_locale_obj is LC_ALL set to the C locale. If this call is to
2378 * switch to LC_ALL => C, simply use that object. But in fact, we already
2379 * have switched to it just above, in preparation for the general case.
2380 * Since we're already there, no need to do further switching. */
2381 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(new_locale)) {
2382 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2383 "bool_setlocale_2008_i: will stay in C"
2385 new_obj = PL_C_locale_obj;
2387 /* 'entry_obj' is now dangling, of no further use to anyone (unless it
2388 * is one of the special ones). Free it to avoid a leak */
2389 if (! entry_obj_is_special) {
2390 freelocale(entry_obj);
2393 update_PL_curlocales_i(index, new_locale, caller_line);
2395 else { /* Here is the general case, not to LC_ALL => C */
2397 /* The newlocale() call(s) below take a basis object to build upon to
2398 * create the changed locale, trashing it iff successful.
2400 * For the objects that are not to be modified by this function, we
2401 * create a duplicate that gets trashed instead.
2403 * Also if we will have to loop doing multiple newlocale()s, there is a
2404 * chance we will succeed for the first few, and then fail, having to
2405 * back out. We need to duplicate 'entry_obj' in this case as well, so
2406 * it remains valid as something to back out to. */
2407 locale_t basis_obj = entry_obj;
2409 if (entry_obj_is_special || need_loop) {
2410 basis_obj = duplocale(basis_obj);
2412 locale_panic_via_("duplocale failed", __FILE__, caller_line);
2413 NOT_REACHED; /* NOTREACHED */
2416 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2417 "bool_setlocale_2008_i created %p by"
2418 " duping the input\n", basis_obj));
2421 # define DEBUG_NEW_OBJECT_CREATED(category, locale, new, old, caller_line) \
2422 DEBUG_Lv(PerlIO_printf(Perl_debug_log, \
2423 "bool_setlocale_2008_i(%s, %s): created %p" \
2424 " while freeing %p; called from %" LINE_Tf \
2425 " via %" LINE_Tf "\n", \
2426 category, locale, new, old, \
2427 caller_line, __LINE__))
2428 # define DEBUG_NEW_OBJECT_FAILED(category, locale, basis_obj) \
2429 DEBUG_L(PerlIO_printf(Perl_debug_log, \
2430 "bool_setlocale_2008_i: creating new object" \
2431 " for (%s '%s') from %p failed; called from %" \
2432 LINE_Tf " via %" LINE_Tf "\n", \
2433 category, locale, basis_obj, \
2434 caller_line, __LINE__));
2436 /* Ready to create a new locale by modification of the existing one.
2438 * NOTE: This code may incorrectly show up as a leak under the address
2439 * sanitizer. We do not free this object under normal teardown, however
2440 * you can set PERL_DESTRUCT_LEVEL=2 to cause it to be freed.
2443 # ifdef NEWLOCALE_HANDLES_DISPARATE_LC_ALL
2445 /* Some platforms have a newlocale() that can handle disparate LC_ALL
2446 * input, so on these a single call to newlocale() always works */
2449 /* If a single call to newlocale() will do */
2455 new_obj = newlocale(mask,
2456 override_ignored_category(index, new_locale),
2459 DEBUG_NEW_OBJECT_FAILED(category_names[index], new_locale,
2462 /* Since the call failed, it didn't trash 'basis_obj', which is
2463 * a dup for these objects, and hence would leak if we don't
2464 * free it. XXX However, something is seriously wrong if we
2465 * can't switch to C or the global locale, so maybe should
2467 if (entry_obj_is_special) {
2468 freelocale(basis_obj);
2471 goto must_restore_state;
2474 DEBUG_NEW_OBJECT_CREATED(category_names[index], new_locale,
2475 new_obj, basis_obj, caller_line);
2477 update_PL_curlocales_i(index, new_locale, caller_line);
2480 # ifndef NEWLOCALE_HANDLES_DISPARATE_LC_ALL
2482 else { /* Need multiple newlocale() calls */
2484 /* Loop through the individual categories, setting the locale of
2485 * each to the corresponding name previously populated into
2486 * newlocales[]. Each iteration builds on the previous one, adding
2487 * its category to what's already been calculated, and taking as a
2488 * basis for what's been calculated 'basis_obj', which is updated
2489 * each iteration to be the result of the previous one. Upon
2490 * success, newlocale() trashes the 'basis_obj' parameter to it.
2491 * If any iteration fails, we immediately give up, restore the
2492 * locale to what it was at the time this function was called
2493 * (saved in 'entry_obj'), and return failure. */
2495 /* Loop, using the previous iteration's result as the basis for the
2496 * next one. (The first time we effectively use the locale in
2497 * force upon entry to this function.) */
2498 for_all_individual_category_indexes(i) {
2499 new_obj = newlocale(category_masks[i],
2503 DEBUG_NEW_OBJECT_CREATED(category_names[i],
2507 basis_obj = new_obj;
2511 /* Failed. Likely this is because the proposed new locale
2512 * isn't valid on this system. */
2514 DEBUG_NEW_OBJECT_FAILED(category_names[i],
2518 /* newlocale() didn't trash this, since the function call
2520 freelocale(basis_obj);
2522 for_all_individual_category_indexes(j) {
2523 Safefree(new_locales[j]);
2526 goto must_restore_state;
2529 /* Success for all categories. */
2530 for_all_individual_category_indexes(i) {
2531 update_PL_curlocales_i(i, new_locales[i], caller_line);
2532 Safefree(new_locales[i]);
2535 /* We dup'd entry_obj in case we had to fall back to it. The
2536 * newlocale() above destroyed the dup when it first succeeded, but
2537 * entry_obj itself is left dangling, so free it */
2538 if (! entry_obj_is_special) {
2539 freelocale(entry_obj);
2543 # endif /* End of newlocale can't handle disparate LC_ALL input */
2547 # undef DEBUG_NEW_OBJECT_CREATED
2548 # undef DEBUG_NEW_OBJECT_FAILED
2550 /* Here, successfully created an object representing the desired locale;
2551 * now switch into it */
2552 if (! uselocale(new_obj)) {
2553 freelocale(new_obj);
2554 locale_panic_(Perl_form(aTHX_ "(called from %" LINE_Tf "):"
2555 " bool_setlocale_2008_i: switching"
2556 " into new locale failed",
2560 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2561 "bool_setlocale_2008_i: now using %p\n", new_obj));
2563 # ifdef MULTIPLICITY /* Unlikely, but POSIX 2008 functions could be
2564 Configured to be used on unthreaded perls, in which
2565 case this object doesn't exist */
2567 if (DEBUG_Lv_TEST) {
2568 if (PL_cur_locale_obj != new_obj) {
2569 PerlIO_printf(Perl_debug_log,
2570 "bool_setlocale_2008_i: PL_cur_locale_obj"
2571 " was %p, now is %p\n",
2572 PL_cur_locale_obj, new_obj);
2576 /* Update the current object */
2577 PL_cur_locale_obj = new_obj;
2580 # ifdef HAS_GLIBC_LC_MESSAGES_BUG
2582 /* Invalidate the glibc cache of loaded translations if the locale has
2583 * changed, see [perl #134264] and
2584 * https://sourceware.org/bugzilla/show_bug.cgi?id=24936 */
2585 if (old_messages_locale) {
2586 if (strNE(old_messages_locale, querylocale_c(LC_MESSAGES))) {
2587 textdomain(textdomain(NULL));
2597 /* We earlier switched to the LC_ALL => C locale in anticipation of it
2598 * succeeding, Now have to switch back to the state upon entry. */
2599 if (! uselocale(entry_obj)) {
2600 setlocale_failure_panic_i(index, "switching back to",
2601 locale_on_entry, __LINE__, caller_line);
2607 /*---------------------------------------------------------------------------*/
2609 # define void_setlocale_i_with_caller(i, locale, file, line) \
2611 if (! bool_setlocale_i(i, locale)) \
2612 setlocale_failure_panic_via_i(i, NULL, locale, __LINE__, 0, \
2616 # define void_setlocale_r_with_caller(cat, locale, file, line) \
2617 void_setlocale_i_with_caller(get_category_index(cat), locale, \
2620 # define void_setlocale_c_with_caller(cat, locale, file, line) \
2621 void_setlocale_i_with_caller(cat##_INDEX_, locale, file, line)
2623 # define void_setlocale_i(i, locale) \
2624 void_setlocale_i_with_caller(i, locale, __FILE__, __LINE__)
2625 # define void_setlocale_c(cat, locale) \
2626 void_setlocale_i(cat##_INDEX_, locale)
2627 # define void_setlocale_r(cat, locale) \
2628 void_setlocale_i(get_category_index(cat), locale)
2630 /*===========================================================================*/
2633 # error Unexpected Configuration
2634 #endif /* End of the various implementations of the setlocale and
2635 querylocale macros used in the remainder of this program */
2637 /* query_nominal_locale_i() is used when the caller needs the locale that an
2638 * external caller would be expecting, and not what we're secretly using
2639 * behind the scenes. It deliberately doesn't handle LC_ALL; use
2640 * calculate_LC_ALL_string() for that. */
2641 #ifdef USE_LOCALE_NUMERIC
2642 # define query_nominal_locale_i(i) \
2643 (__ASSERT_(i != LC_ALL_INDEX_) \
2644 ((i == LC_NUMERIC_INDEX_) ? PL_numeric_name : querylocale_i(i)))
2646 # define query_nominal_locale_i(i) \
2647 (__ASSERT_(i != LC_ALL_INDEX_) querylocale_i(i))
2650 #ifdef USE_PL_CURLOCALES
2653 S_update_PL_curlocales_i(pTHX_
2654 const locale_category_index index,
2655 const char * new_locale,
2656 const line_t caller_line)
2658 /* Update PL_curlocales[], which is parallel to the other ones indexed by
2659 * our mapping of libc category number to our internal equivalents. */
2661 PERL_ARGS_ASSERT_UPDATE_PL_CURLOCALES_I;
2662 assert(index <= LC_ALL_INDEX_);
2664 if (index == LC_ALL_INDEX_) {
2666 /* For LC_ALL, we change all individual categories to correspond,
2667 * including the LC_ALL element */
2668 for (unsigned int i = 0; i <= LC_ALL_INDEX_; i++) {
2669 Safefree(PL_curlocales[i]);
2670 PL_curlocales[i] = NULL;
2673 switch (parse_LC_ALL_string(new_locale,
2674 (const char **) &PL_curlocales,
2675 check_that_overridden, /* things should
2679 true, /* Always fill array */
2680 true, /* Panic if fails, as to get here
2681 it earlier had to have succeeded
2687 case only_element_0:
2688 locale_panic_via_("Unexpected return from parse_LC_ALL_string",
2689 __FILE__, caller_line);
2692 /* parse_LC_ALL_string() has already filled PL_curlocales properly,
2693 * except for the LC_ALL element, which should be set to
2695 PL_curlocales[LC_ALL_INDEX_] = savepv(new_locale);
2698 else { /* Not LC_ALL */
2700 /* Update the single category's record */
2701 Safefree(PL_curlocales[index]);
2702 PL_curlocales[index] = savepv(new_locale);
2704 /* Invalidate LC_ALL */
2705 Safefree(PL_curlocales[LC_ALL_INDEX_]);
2706 PL_curlocales[LC_ALL_INDEX_] = NULL;
2710 # endif /* Need PL_curlocales[] */
2712 /*===========================================================================*/
2714 #if defined(USE_LOCALE)
2716 /* This paradigm is needed in several places in the function below. We have to
2717 * substitute the nominal locale for LC_NUMERIC when returning a value for
2718 * external consumption */
2719 # ifndef USE_LOCALE_NUMERIC
2720 # define ENTRY(i, array, format) array[i]
2722 # define ENTRY(i, array, format) \
2723 (UNLIKELY( format == EXTERNAL_FORMAT_FOR_QUERY \
2724 && i == LC_NUMERIC_INDEX_) \
2731 S_calculate_LC_ALL_string(pTHX_ const char ** category_locales_list,
2732 const calc_LC_ALL_format format,
2733 const calc_LC_ALL_return returning,
2734 const line_t caller_line)
2736 PERL_ARGS_ASSERT_CALCULATE_LC_ALL_STRING;
2738 /* NOTE: On Configurations that have PL_curlocales[], this function has the
2739 * side effect of updating the LC_ALL_INDEX_ element with its result.
2741 * This function calculates a string that defines the locale(s) LC_ALL is
2742 * set to, in either:
2743 * 1) Our internal format if 'format' is set to INTERNAL_FORMAT.
2744 * 2) The external format returned by Perl_setlocale() if 'format' is set
2745 * to EXTERNAL_FORMAT_FOR_QUERY or EXTERNAL_FORMAT_FOR_SET.
2747 * These two are distinguished by:
2748 * a) EXTERNAL_FORMAT_FOR_SET returns the actual locale currently in
2750 * b) EXTERNAL_FORMAT_FOR_QUERY returns the nominal locale.
2751 * Currently this can differ only from the actual locale in the
2752 * LC_NUMERIC category when it is set to a locale whose radix is
2753 * not a dot. (The actual locale is kept as a dot to accommodate
2754 * the large corpus of XS code that expects it to be that;
2755 * switched to a non-dot temporarily during certain operations
2756 * that require the actual radix.)
2758 * In both 1) and 2), LC_ALL's values are passed to this function by
2759 * 'category_locales_list' which is either:
2760 * 1) a pointer to an array of strings with up-to-date values of all the
2761 * individual categories; or
2762 * 2) NULL, to indicate to use querylocale_i() to get each individual
2765 * The caller sets 'returning' to
2766 * WANT_TEMP_PV the function returns the calculated string
2767 * as a mortalized temporary, so the caller
2768 * doesn't have to worry about it being
2769 * per-thread, nor needs to arrange for its
2771 * WANT_PL_setlocale_buf the function stores the calculated string
2772 * into the per-thread buffer PL_setlocale_buf
2773 * and returns a pointer to that. The buffer
2774 * is cleaned up automatically in process
2775 * destruction. This return method avoids
2776 * extra copies in some circumstances.
2777 * WANT_VOID NULL is returned. This is used when the
2778 * function is being called only for its side
2779 * effect of updating
2780 * PL_curlocales[LC_ALL_INDEX_]
2782 * querylocale(), on systems that have it, doesn't tend to work for LC_ALL.
2783 * So we have to construct the answer ourselves based on the passed in
2786 * If all individual categories are the same locale, we can just set LC_ALL
2787 * to that locale. But if not, we have to create an aggregation of all the
2788 * categories on the system. Platforms differ as to the syntax they use
2789 * for these non-uniform locales for LC_ALL. Some, like glibc and Windows,
2790 * use an unordered series of name=value pairs, like
2791 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
2792 * to specify LC_ALL; others, like *BSD, use a positional notation with a
2793 * delimitter, typically a single '/' character:
2796 * When the external format is desired, this function returns whatever the
2797 * system expects. The internal format is always name=value pairs.
2799 * For systems that have categories we don't know about, the algorithm
2800 * below won't know about those missing categories, leading to potential
2801 * bugs for code that looks at them. If there is an environment variable
2802 * that sets that category, we won't know to look for it, and so our use of
2803 * LANG or "C" improperly overrides it. On the other hand, if we don't do
2804 * what is done here, and there is no environment variable, the category's
2805 * locale should be set to LANG or "C". So there is no good solution. khw
2806 * thinks the best is to make sure we have a complete list of possible
2807 * categories, adding new ones as they show up on obscure platforms.
2810 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
2811 "Entering calculate_LC_ALL_string(%s);"
2812 " called from %" LINE_Tf "\n",
2813 ((format == EXTERNAL_FORMAT_FOR_QUERY)
2814 ? "EXTERNAL_FORMAT_FOR_QUERY"
2815 : ((format == EXTERNAL_FORMAT_FOR_SET)
2816 ? "EXTERNAL_FORMAT_FOR_SET"
2817 : "INTERNAL_FORMAT")),
2820 bool input_list_was_NULL = (category_locales_list == NULL);
2822 /* If there was no input category list, construct a temporary one
2824 const char * my_category_locales_list[LC_ALL_INDEX_];
2825 const char ** locales_list = category_locales_list;
2826 if (locales_list == NULL) {
2827 locales_list = my_category_locales_list;
2829 if (format == EXTERNAL_FORMAT_FOR_QUERY) {
2830 for_all_individual_category_indexes(i) {
2831 locales_list[i] = query_nominal_locale_i(i);
2835 for_all_individual_category_indexes(i) {
2836 locales_list[i] = querylocale_i(i);
2841 /* While we are calculating LC_ALL, we see if every category's locale is
2842 * the same as every other's or not. */
2843 # ifndef HAS_IGNORED_LOCALE_CATEGORIES_
2845 /* When we pay attention to all categories, we assume they are all the same
2846 * until proven different */
2847 bool disparate = false;
2851 /* But if there are ignored categories, those will be set to "C", so try an
2852 * arbitrary category, and if it isn't C, we know immediately that the
2853 * locales are disparate. (The #if conditionals are to handle the case
2854 * where LC_NUMERIC_INDEX_ is 0. We don't want to use LC_NUMERIC to
2855 * compare, as that may be different between external and internal forms.)
2857 # if ! defined(USE_LOCALE_NUMERIC)
2859 bool disparate = ! isNAME_C_OR_POSIX(locales_list[0]);
2861 # elif LC_NUMERIC_INDEX_ != 0
2863 bool disparate = ! isNAME_C_OR_POSIX(locales_list[0]);
2867 /* Would need revision to handle the very unlikely case where only a single
2868 * category, LC_NUMERIC, is defined */
2869 assert(LOCALE_CATEGORIES_COUNT_ > 0);
2871 bool disparate = ! isNAME_C_OR_POSIX(locales_list[1]);
2876 /* Calculate the needed size for the string listing the individual locales.
2877 * Initialize with values known at compile time. */
2879 const char *separator;
2881 # ifdef PERL_LC_ALL_USES_NAME_VALUE_PAIRS /* Positional formatted LC_ALL */
2882 PERL_UNUSED_ARG(format);
2885 if (format != INTERNAL_FORMAT) {
2887 /* Here, we will be using positional notation. it includes n-1
2889 total_len = ( LOCALE_CATEGORIES_COUNT_ - 1)
2890 * STRLENs(PERL_LC_ALL_SEPARATOR)
2891 + 1; /* And a trailing NUL */
2892 separator = PERL_LC_ALL_SEPARATOR;
2899 /* name=value output is always used in internal format, and when
2900 * positional isn't available on the platform. */
2901 total_len = lc_all_boiler_plate_length;
2905 /* The total length then is just the sum of the above boiler-plate plus the
2906 * total strlen()s of the locale name of each individual category. */
2907 for_all_individual_category_indexes(i) {
2908 const char * entry = ENTRY(i, locales_list, format);
2910 total_len += strlen(entry);
2911 if (! disparate && strNE(entry, locales_list[0])) {
2916 bool free_if_void_return = false;
2917 const char * retval;
2919 /* If all categories have the same locale, we already know the answer */
2921 if (returning == WANT_PL_setlocale_buf) {
2922 save_to_buffer(locales_list[0],
2924 &PL_setlocale_bufsize);
2925 retval = PL_setlocale_buf;
2929 retval = locales_list[0];
2931 /* If a temporary is wanted for the return, and we had to create
2932 * the input list ourselves, we created it into such a temporary,
2933 * so no further work is needed; but otherwise, make a mortal copy
2934 * of this passed-in list element */
2935 if (returning == WANT_TEMP_PV && ! input_list_was_NULL) {
2936 retval = savepv(retval);
2940 /* In all cases here, there's nothing we create that needs to be
2941 * freed, so leave 'free_if_void_return' set to the default
2945 else { /* Here, not all categories have the same locale */
2949 /* If returning to PL_setlocale_buf, set up to write directly to it,
2950 * being sure it is resized to be large enough */
2951 if (returning == WANT_PL_setlocale_buf) {
2952 set_save_buffer_min_size(total_len,
2954 &PL_setlocale_bufsize);
2955 constructed = PL_setlocale_buf;
2957 else { /* Otherwise we need new memory to hold the calculated value. */
2959 Newx(constructed, total_len, char);
2961 /* If returning the new memory, it must be set up to be freed
2962 * later; otherwise at the end of this function */
2963 if (returning == WANT_TEMP_PV) {
2964 SAVEFREEPV(constructed);
2967 free_if_void_return = true;
2971 constructed[0] = '\0';
2973 /* Loop through all the categories */
2974 for_all_individual_category_indexes(j) {
2976 /* Add a separator, except before the first one */
2978 my_strlcat(constructed, separator, total_len);
2985 # ifndef PERL_LC_ALL_USES_NAME_VALUE_PAIRS
2987 if (UNLIKELY(format != INTERNAL_FORMAT)) {
2989 /* In positional notation 'j' means the position, and we have
2990 * to convert to the index 'i' */
2991 i = map_LC_ALL_position_to_index[j];
2993 entry = ENTRY(i, locales_list, format);
2994 needed_len = my_strlcat(constructed, entry, total_len);
3000 /* Below, we are to use name=value notation, either because
3001 * that's what the platform uses, or because this is the
3002 * internal format, which uses that notation regardless of the
3005 entry = ENTRY(i, locales_list, format);
3007 /* "name=locale;" */
3008 my_strlcat(constructed, category_names[i], total_len);
3009 my_strlcat(constructed, "=", total_len);
3010 needed_len = my_strlcat(constructed, entry, total_len);
3013 if (LIKELY(needed_len <= total_len)) {
3017 /* If would have overflowed, panic */
3018 locale_panic_via_(Perl_form(aTHX_
3019 "Internal length calculation wrong.\n"
3020 "\"%s\" was not entirely added to"
3021 " \"%.*s\"; needed=%zu, had=%zu",
3022 entry, (int) total_len,
3024 needed_len, total_len),
3027 } /* End of loop through the categories */
3029 retval = constructed;
3030 } /* End of the categories' locales are displarate */
3032 # if defined(USE_PL_CURLOCALES) && defined(LC_ALL)
3034 if (format == INTERNAL_FORMAT) {
3036 /* PL_curlocales[LC_ALL_INDEX_] is updated as a side-effect of this
3037 * function for internal format. */
3038 Safefree(PL_curlocales[LC_ALL_INDEX_]);
3039 PL_curlocales[LC_ALL_INDEX_] = savepv(retval);
3044 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3045 "calculate_LC_ALL_string calculated '%s'\n",
3048 if (returning == WANT_VOID) {
3049 if (free_if_void_return) {
3059 # if defined(WIN32) || ( defined(USE_POSIX_2008_LOCALE) \
3060 && ! defined(USE_QUERYLOCALE))
3063 S_find_locale_from_environment(pTHX_ const locale_category_index index)
3065 /* NB: This function may actually change the locale on Windows. It
3066 * currently is designed to be called only from setting the locale on
3067 * Windows, and POSIX 2008
3069 * This function returns the locale specified by the program's environment
3070 * for the category specified by our internal index number 'index'. It
3071 * therefore simulates:
3072 * setlocale(cat, "")
3073 * but, except for some cases in Windows, doesn't actually change the
3074 * locale; merely returns it.
3076 * The return need not be freed by the caller. This
3077 * promise relies on PerlEnv_getenv() returning a mortalized copy to us.
3079 * The simulation is needed only on certain platforms; otherwise, libc is
3080 * called with "" to get the actual value(s). The simulation is needed
3083 * 1) On Windows systems, the concept of the POSIX ordering of
3084 * environment variables is missing. To increase portability of
3085 * programs across platforms, the POSIX ordering is emulated on
3088 * 2) On POSIX 2008 systems without querylocale(), it is problematic
3089 * getting the results of the POSIX 2008 equivalent of
3091 * setlocale(category, "")
3093 * To ensure that we know exactly what those values are, we do the
3094 * setting ourselves, using the documented algorithm specified by the
3095 * POSIX standard (assuming the platform follows the Standard) rather
3096 * than use "" as the locale. This will lead to results that differ
3097 * from native behavior if the native behavior differs from the
3098 * Standard's documented value, but khw believes it is better to know
3099 * what's going on, even if different from native, than to just guess.
3101 * glibc systems differ from this standard in having a LANGUAGE
3102 * environment variable used for just LC_MESSAGES. This function does
3105 * Another option for the POSIX 2008 case would be, in a critical
3106 * section, to save the global locale's current value, and do a
3107 * straight setlocale(LC_ALL, ""). That would return our desired
3108 * values, destroying the global locale's, which we would then
3109 * restore. But that could cause races with any other thread that is
3110 * using the global locale and isn't using the mutex. And, the only
3111 * reason someone would have done that is because they are calling a
3112 * library function, like in gtk, that calls setlocale(), and which
3113 * can't be changed to use the mutex. That wouldn't be a problem if
3114 * this were to be done before any threads had switched, say during
3115 * perl construction time. But this code would still be needed for
3118 * The Windows and POSIX 2008 differ in that the ultimate fallback is "C"
3119 * in POSIX, and is the system default locale in Windows. To get that
3120 * system default value, we actually have to call setlocale() on Windows.
3123 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3124 const char * locale_names[LC_ALL_INDEX_] = { NULL };
3126 /* Use any "LC_ALL" environment variable, as it overrides everything else.
3128 if (lc_all && strNE(lc_all, "")) {
3132 /* Here, no usable LC_ALL environment variable. We have to handle each
3133 * category separately. If all categories are desired, we loop through
3134 * them all. If only an individual category is desired, to avoid
3135 * duplicating logic, we use the same loop, but set up the limits so it is
3136 * only executed once, for that particular category. */
3137 locale_category_index lower, upper, offset;
3138 if (index == LC_ALL_INDEX_) {
3139 lower = (locale_category_index) 0;
3140 upper = (locale_category_index) ((int) LC_ALL_INDEX_ - 1);
3141 offset = (locale_category_index) 0;
3147 /* 'offset' is used so that the result of the single loop iteration is
3148 * stored into output[0] */
3152 /* When no LC_ALL environment variable, LANG is used as a default, but
3153 * overridden for individual categories that have corresponding environment
3154 * variables. If no LANG exists, the default is "C" on POSIX 2008, or the
3155 * system default for the category on Windows. */
3156 const char * env_lang = NULL;
3158 /* For each desired category, use any corresponding environment variable;
3159 * or the default if none such exists. */
3160 bool is_disparate = false; /* Assume is uniform until proven otherwise */
3161 for (unsigned i = lower; i <= upper; i++) {
3162 const char * const env_override = PerlEnv_getenv(category_names[i]);
3163 unsigned int j = i - offset;
3165 if (env_override && strNE(env_override, "")) {
3166 locale_names[j] = env_override;
3168 else { /* Here, no corresponding environment variable, see if LANG
3169 exists and is usable. Done this way to avoid fetching LANG
3170 unless it is actually needed */
3171 if (env_lang == NULL) {
3172 env_lang = PerlEnv_getenv("LANG");
3174 /* If not usable, set it to a non-NULL illegal value so won't
3175 * try to use it below */
3176 if (env_lang == NULL || strEQ(env_lang, "")) {
3177 env_lang = (const char *) 1;
3181 /* If a usable LANG exists, use it. */
3182 if (env_lang != NULL && env_lang != (const char *) 1) {
3183 locale_names[j] = env_lang;
3188 /* If no LANG, use the system default on Windows. */
3189 locale_names[j] = wrap_wsetlocale(categories[i], ".ACP");
3190 if (locale_names[j]) {
3191 SAVEFREEPV(locale_names[j]);
3195 { /* If nothing was found or worked, use C */
3196 locale_names[j] = "C";
3201 if (j > 0 && ! is_disparate && strNE(locale_names[0], locale_names[j]))
3203 is_disparate = true;
3206 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3207 "find_locale_from_environment i=%u, j=%u, name=%s,"
3208 " locale=%s, locale of 0th category=%s, disparate=%d\n",
3209 i, j, category_names[i],
3210 locale_names[j], locale_names[0], is_disparate));
3213 if (! is_disparate) {
3214 return locale_names[0];
3217 return calculate_LC_ALL_string(locale_names, INTERNAL_FORMAT,
3223 # if defined(DEBUGGING) || defined(USE_PERL_SWITCH_LOCALE_CONTEXT)
3226 S_get_LC_ALL_display(pTHX)
3228 return calculate_LC_ALL_string(NULL, INTERNAL_FORMAT,
3236 S_setlocale_failure_panic_via_i(pTHX_
3237 const locale_category_index cat_index,
3238 const char * current,
3239 const char * failed,
3240 const line_t proxy_caller_line,
3241 const line_t immediate_caller_line,
3242 const char * const higher_caller_file,
3243 const line_t higher_caller_line)
3245 PERL_ARGS_ASSERT_SETLOCALE_FAILURE_PANIC_VIA_I;
3247 /* Called to panic when a setlocale form unexpectedly failed for the
3248 * category determined by 'cat_index', and the locale that was in effect
3249 * (and likely still is) is 'current'. 'current' may be NULL, which causes
3250 * this function to query what it is.
3252 * The extra caller information is used for when a function acts as a
3253 * stand-in for another function, which a typical reader would more likely
3254 * think would be the caller
3256 * If a line number is 0, its stack (sort-of) frame is omitted; same if
3257 * it's the same line number as the next higher caller. */
3259 const int cat = categories[cat_index];
3260 const char * name = category_names[cat_index];
3264 if (current == NULL) {
3265 current = querylocale_i(cat_index);
3268 const char * proxy_text = "";
3269 if (proxy_caller_line != 0 && proxy_caller_line != immediate_caller_line)
3271 proxy_text = Perl_form(aTHX_ "\nCalled via %s: %" LINE_Tf,
3272 __FILE__, proxy_caller_line);
3274 if ( strNE(__FILE__, higher_caller_file)
3275 || ( immediate_caller_line != 0
3276 && immediate_caller_line != higher_caller_line))
3278 proxy_text = Perl_form(aTHX_ "%s\nCalled via %s: %" LINE_Tf,
3279 proxy_text, __FILE__,
3280 immediate_caller_line);
3283 /* 'false' in the get_displayable_string() calls makes it not think the
3284 * locale is UTF-8, so just dumps bytes. Actually figuring it out can be
3285 * too complicated for a panic situation. */
3286 const char * msg = Perl_form(aTHX_
3287 "Can't change locale for %s (%d) from '%s' to '%s'"
3290 get_displayable_string(current,
3291 current + strlen(current),
3293 get_displayable_string(failed,
3294 failed + strlen(failed),
3299 Perl_locale_panic(msg, __LINE__, higher_caller_file, higher_caller_line);
3300 NOT_REACHED; /* NOTREACHED */
3303 /* Any of these will allow us to find the RADIX */
3304 # if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_SOME_LANGINFO) \
3305 || defined(HAS_LOCALECONV) \
3306 || defined(HAS_SNPRINTF))
3307 # define CAN_CALCULATE_RADIX
3309 # ifdef USE_LOCALE_NUMERIC
3312 S_new_numeric(pTHX_ const char *newnum, bool force)
3314 PERL_ARGS_ASSERT_NEW_NUMERIC;
3316 /* Called after each libc setlocale() or uselocale() call affecting
3317 * LC_NUMERIC, to tell core Perl this and that 'newnum' is the name of the
3318 * new locale, and we are switched into it. It installs this locale as the
3319 * current underlying default, and then switches to the C locale, if
3320 * necessary, so that the code that has traditionally expected the radix
3321 * character to be a dot may continue to do so.
3323 * The default locale and the C locale can be toggled between by use of the
3324 * set_numeric_underlying() and set_numeric_standard() functions, which
3325 * should probably not be called directly, but only via macros like
3326 * SET_NUMERIC_STANDARD() in perl.h.
3328 * The toggling is necessary mainly so that a non-dot radix decimal point
3329 * character can be input and output, while allowing internal calculations
3332 * This sets several interpreter-level variables:
3333 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
3334 * PL_numeric_underlying A boolean indicating if the toggled state is
3335 * such that the current locale is the program's
3337 * PL_numeric_standard An int indicating if the toggled state is such
3338 * that the current locale is the C locale or
3339 * indistinguishable from the C locale. If non-zero, it
3340 * is in C; if > 1, it means it may not be toggled away
3342 * PL_numeric_underlying_is_standard A bool kept by this function
3343 * indicating that the underlying locale and the standard
3344 * C locale are indistinguishable for the purposes of
3345 * LC_NUMERIC. This happens when both of the above two
3346 * variables are true at the same time. (Toggling is a
3347 * no-op under these circumstances.) This variable is
3348 * used to avoid having to recalculate.
3349 * PL_numeric_radix_sv Contains the string that code should use for the
3350 * decimal point. It is set to either a dot or the
3351 * program's underlying locale's radix character string,
3352 * depending on the situation.
3353 * PL_underlying_radix_sv Contains the program's underlying locale's
3354 * radix character string. This is copied into
3355 * PL_numeric_radix_sv when the situation warrants. It
3356 * exists to avoid having to recalculate it when toggling.
3359 DEBUG_L( PerlIO_printf(Perl_debug_log,
3360 "Called new_numeric with %s, PL_numeric_name=%s\n",
3361 newnum, PL_numeric_name));
3363 /* If not forcing this procedure, and there isn't actually a change from
3364 * our records, do nothing. (Our records can be wrong when sync'ing to the
3365 * locale set up by an external library, hence the 'force' parameter) */
3366 if (! force && strEQ(PL_numeric_name, newnum)) {
3370 Safefree(PL_numeric_name);
3371 PL_numeric_name = savepv(newnum);
3373 /* Handle the trivial case. Since this is called at process
3374 * initialization, be aware that this bit can't rely on much being
3376 if (isNAME_C_OR_POSIX(PL_numeric_name)) {
3377 PL_numeric_standard = TRUE;
3378 PL_numeric_underlying_is_standard = TRUE;
3379 PL_numeric_underlying = TRUE;
3380 sv_setpv(PL_numeric_radix_sv, C_decimal_point);
3381 SvUTF8_off(PL_numeric_radix_sv);
3382 sv_setpv(PL_underlying_radix_sv, C_decimal_point);
3383 SvUTF8_off(PL_underlying_radix_sv);
3387 /* We are in the underlying locale until changed at the end of this
3389 PL_numeric_underlying = TRUE;
3391 char * radix = NULL;
3392 utf8ness_t utf8ness = UTF8NESS_IMMATERIAL;
3394 /* Find and save this locale's radix character. */
3395 my_langinfo_c(RADIXCHAR, LC_NUMERIC, PL_numeric_name,
3396 &radix, NULL, &utf8ness);
3397 sv_setpv(PL_underlying_radix_sv, radix);
3399 if (utf8ness == UTF8NESS_YES) {
3400 SvUTF8_on(PL_underlying_radix_sv);
3403 SvUTF8_off(PL_underlying_radix_sv);
3406 DEBUG_L(PerlIO_printf(Perl_debug_log,
3407 "Locale radix is '%s', ?UTF-8=%d\n",
3408 SvPVX(PL_underlying_radix_sv),
3409 cBOOL(SvUTF8(PL_underlying_radix_sv))));
3411 /* This locale is indistinguishable from C (for numeric purposes) if both
3412 * the radix character and the thousands separator are the same as C's.
3413 * Start with the radix. */
3414 PL_numeric_underlying_is_standard = strEQ(C_decimal_point, radix);
3417 # ifndef TS_W32_BROKEN_LOCALECONV
3419 /* If the radix isn't the same as C's, we know it is distinguishable from
3420 * C; otherwise check the thousands separator too. Only if both are the
3421 * same as C's is the locale indistinguishable from C.
3423 * But on earlier Windows versions, there is a potential race. This code
3424 * knows that localeconv() (elsewhere in this file) will be used to extract
3425 * the needed value, and localeconv() was buggy for quite a while, and that
3426 * code in this file hence uses a workaround. And that workaround may have
3427 * an (unlikely) race. Gathering the radix uses a different workaround on
3428 * Windows that doesn't involve a race. It might be possible to do the
3429 * same for this (patches welcome).
3431 * Until then khw doesn't think it's worth even the small risk of a race to
3432 * get this value, which doesn't appear to be used in any of the Microsoft
3433 * library routines anyway. */
3435 char * scratch_buffer = NULL;
3436 if (PL_numeric_underlying_is_standard) {
3437 PL_numeric_underlying_is_standard = strEQ(C_thousands_sep,
3438 my_langinfo_c(THOUSEP, LC_NUMERIC,
3443 Safefree(scratch_buffer);
3447 PL_numeric_standard = PL_numeric_underlying_is_standard;
3449 /* Keep LC_NUMERIC so that it has the C locale radix and thousands
3450 * separator. This is for XS modules, so they don't have to worry about
3451 * the radix being a non-dot. (Core operations that need the underlying
3452 * locale change to it temporarily). */
3453 if (! PL_numeric_standard) {
3454 set_numeric_standard(__FILE__, __LINE__);
3461 Perl_set_numeric_standard(pTHX_ const char * const file, const line_t line)
3463 PERL_ARGS_ASSERT_SET_NUMERIC_STANDARD;
3464 PERL_UNUSED_ARG(file); /* Some Configurations ignore these */
3465 PERL_UNUSED_ARG(line);
3467 # ifdef USE_LOCALE_NUMERIC
3469 /* Unconditionally toggle the LC_NUMERIC locale to the C locale
3471 * Most code should use the macro SET_NUMERIC_STANDARD() in perl.h
3472 * instead of calling this directly. The macro avoids calling this routine
3473 * if toggling isn't necessary according to our records (which could be
3474 * wrong if some XS code has changed the locale behind our back) */
3476 DEBUG_L(PerlIO_printf(Perl_debug_log, "Setting LC_NUMERIC locale to"
3477 " standard C; called from %s: %"
3478 LINE_Tf "\n", file, line));
3480 void_setlocale_c_with_caller(LC_NUMERIC, "C", file, line);
3481 PL_numeric_standard = TRUE;
3482 sv_setpv(PL_numeric_radix_sv, C_decimal_point);
3483 SvUTF8_off(PL_numeric_radix_sv);
3485 PL_numeric_underlying = PL_numeric_underlying_is_standard;
3487 # endif /* USE_LOCALE_NUMERIC */
3492 Perl_set_numeric_underlying(pTHX_ const char * const file, const line_t line)
3494 PERL_ARGS_ASSERT_SET_NUMERIC_UNDERLYING;
3495 PERL_UNUSED_ARG(file); /* Some Configurations ignore these */
3496 PERL_UNUSED_ARG(line);
3498 # ifdef USE_LOCALE_NUMERIC
3500 /* Unconditionally toggle the LC_NUMERIC locale to the current underlying
3503 * Most code should use the macro SET_NUMERIC_UNDERLYING() in perl.h
3504 * instead of calling this directly. The macro avoids calling this routine
3505 * if toggling isn't necessary according to our records (which could be
3506 * wrong if some XS code has changed the locale behind our back) */
3508 DEBUG_L(PerlIO_printf(Perl_debug_log, "Setting LC_NUMERIC locale to %s;"
3509 " called from %s: %" LINE_Tf "\n",
3510 PL_numeric_name, file, line));
3511 /* Maybe not in init? assert(PL_locale_mutex_depth > 0);*/
3513 void_setlocale_c_with_caller(LC_NUMERIC, PL_numeric_name, file, line);
3514 PL_numeric_underlying = TRUE;
3515 sv_setsv_nomg(PL_numeric_radix_sv, PL_underlying_radix_sv);
3517 PL_numeric_standard = PL_numeric_underlying_is_standard;
3519 # endif /* USE_LOCALE_NUMERIC */
3523 # ifdef USE_LOCALE_CTYPE
3526 S_new_ctype(pTHX_ const char *newctype, bool force)
3528 PERL_ARGS_ASSERT_NEW_CTYPE;
3529 PERL_UNUSED_ARG(force);
3531 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
3532 * core Perl this and that 'newctype' is the name of the new locale.
3534 * This function sets up the folding arrays for all 256 bytes, assuming
3535 * that tofold() is tolc() since fold case is not a concept in POSIX,
3538 DEBUG_L(PerlIO_printf(Perl_debug_log, "Entering new_ctype(%s)\n",
3541 /* No change means no-op */
3542 if (strEQ(PL_ctype_name, newctype)) {
3546 /* We will replace any bad locale warning with
3547 * 1) nothing if the new one is ok; or
3548 * 2) a new warning for the bad new locale */
3549 if (PL_warn_locale) {
3550 SvREFCNT_dec_NN(PL_warn_locale);
3551 PL_warn_locale = NULL;
3555 Safefree(PL_ctype_name);
3558 PL_in_utf8_turkic_locale = FALSE;
3560 /* For the C locale, just use the standard folds, and we know there are no
3561 * glitches possible, so return early. Since this is called at process
3562 * initialization, be aware that this bit can't rely on much being
3564 if (isNAME_C_OR_POSIX(newctype)) {
3565 Copy(PL_fold, PL_fold_locale, 256, U8);
3566 PL_ctype_name = savepv(newctype);
3567 PL_in_utf8_CTYPE_locale = FALSE;
3571 /* The cache being cleared signals the called function to compute a new
3573 PL_in_utf8_CTYPE_locale = is_locale_utf8(newctype);
3575 PL_ctype_name = savepv(newctype);
3576 bool maybe_utf8_turkic = FALSE;
3578 /* Don't check for problems if we are suppressing the warnings */
3579 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
3581 if (PL_in_utf8_CTYPE_locale) {
3583 /* A UTF-8 locale gets standard rules. But note that code still has to
3584 * handle this specially because of the three problematic code points
3586 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
3588 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
3589 * Turkic. Make sure these two are the only anomalies. (We don't
3590 * require towupper and towlower because they aren't in C89.) */
3592 # if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
3594 if (towupper('i') == 0x130 && towlower('I') == 0x131)
3598 if (toU8_UPPER_LC('i') == 'i' && toU8_LOWER_LC('I') == 'I')
3603 /* This is how we determine it really is Turkic */
3604 check_for_problems = TRUE;
3605 maybe_utf8_turkic = TRUE;
3608 else { /* Not a canned locale we know the values for. Compute them */
3612 bool has_non_ascii_fold = FALSE;
3613 bool found_unexpected = FALSE;
3615 /* Under -DLv, see if there are any folds outside the ASCII range.
3616 * This factoid is used below */
3617 if (DEBUG_Lv_TEST) {
3618 for (unsigned i = 128; i < 256; i++) {
3619 int j = LATIN1_TO_NATIVE(i);
3620 if (toU8_LOWER_LC(j) != j || toU8_UPPER_LC(j) != j) {
3621 has_non_ascii_fold = TRUE;
3629 for (unsigned i = 0; i < 256; i++) {
3630 if (isU8_UPPER_LC(i))
3631 PL_fold_locale[i] = (U8) toU8_LOWER_LC(i);
3632 else if (isU8_LOWER_LC(i))
3633 PL_fold_locale[i] = (U8) toU8_UPPER_LC(i);
3635 PL_fold_locale[i] = (U8) i;
3639 /* Most locales these days are supersets of ASCII. When debugging
3640 * with -DLv, it is helpful to know what the exceptions to that are
3642 if (DEBUG_Lv_TEST) {
3643 bool unexpected = FALSE;
3645 if (isUPPER_L1(i)) {
3647 if (PL_fold_locale[i] != toLOWER_A(i)) {
3651 else if (has_non_ascii_fold) {
3652 if (PL_fold_locale[i] != toLOWER_L1(i)) {
3656 else if (PL_fold_locale[i] != i) {
3660 else if ( isLOWER_L1(i)
3661 && i != LATIN_SMALL_LETTER_SHARP_S
3665 if (PL_fold_locale[i] != toUPPER_A(i)) {
3669 else if (has_non_ascii_fold) {
3670 if (PL_fold_locale[i] != toUPPER_LATIN1_MOD(i)) {
3674 else if (PL_fold_locale[i] != i) {
3678 else if (PL_fold_locale[i] != i) {
3683 found_unexpected = TRUE;
3684 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3685 "For %s, fold of %02x is %02x\n",
3686 newctype, i, PL_fold_locale[i]));
3691 if (found_unexpected) {
3692 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3693 "All bytes not mentioned above either fold to"
3694 " themselves or are the expected ASCII or"
3698 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3699 "No nonstandard folds were found\n"));
3707 /* We only handle single-byte locales (outside of UTF-8 ones); so if this
3708 * locale requires more than one byte, there are going to be BIG problems.
3711 const int mb_cur_max = MB_CUR_MAX;
3713 if (mb_cur_max > 1 && ! PL_in_utf8_CTYPE_locale
3715 /* Some platforms return MB_CUR_MAX > 1 for even the "C" locale.
3716 * Just assume that the implementation for them (plus for POSIX) is
3717 * correct and the > 1 value is spurious. (Since these are
3718 * specially handled to never be considered UTF-8 locales, as long
3719 * as this is the only problem, everything should work fine */
3720 && ! isNAME_C_OR_POSIX(newctype))
3722 DEBUG_L(PerlIO_printf(Perl_debug_log,
3723 "Unsupported, MB_CUR_MAX=%d\n", mb_cur_max));
3725 Perl_ck_warner_d(aTHX_ packWARN(WARN_LOCALE),
3726 "Locale '%s' is unsupported, and may crash the"
3733 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "check_for_problems=%d\n",
3734 check_for_problems));
3736 /* We don't populate the other lists if a UTF-8 locale, but do check that
3737 * everything works as expected, unless checking turned off */
3738 if (check_for_problems) {
3739 /* Assume enough space for every character being bad. 4 spaces each
3740 * for the 94 printable characters that are output like "'x' "; and 5
3741 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
3743 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
3744 unsigned int bad_count = 0; /* Count of bad characters */
3746 for (unsigned i = 0; i < 256; i++) {
3748 /* If checking for locale problems, see if the native ASCII-range
3749 * printables plus \n and \t are in their expected categories in
3750 * the new locale. If not, this could mean big trouble, upending
3751 * Perl's and most programs' assumptions, like having a
3752 * metacharacter with special meaning become a \w. Fortunately,
3753 * it's very rare to find locales that aren't supersets of ASCII
3754 * nowadays. It isn't a problem for most controls to be changed
3755 * into something else; we check only \n and \t, though perhaps \r
3756 * could be an issue as well. */
3757 if (isGRAPH_A(i) || isBLANK_A(i) || i == '\n') {
3758 bool is_bad = FALSE;
3759 char name[4] = { '\0' };
3761 /* Convert the name into a string */
3766 else if (i == '\n') {
3767 my_strlcpy(name, "\\n", sizeof(name));
3769 else if (i == '\t') {
3770 my_strlcpy(name, "\\t", sizeof(name));
3774 my_strlcpy(name, "' '", sizeof(name));
3777 /* Check each possibe class */
3778 if (UNLIKELY(cBOOL(isU8_ALPHANUMERIC_LC(i)) !=
3779 cBOOL(isALPHANUMERIC_A(i))))
3782 DEBUG_L(PerlIO_printf(Perl_debug_log,
3783 "isalnum('%s') unexpectedly is %x\n",
3784 name, cBOOL(isU8_ALPHANUMERIC_LC(i))));
3786 if (UNLIKELY(cBOOL(isU8_ALPHA_LC(i)) != cBOOL(isALPHA_A(i)))) {
3788 DEBUG_L(PerlIO_printf(Perl_debug_log,
3789 "isalpha('%s') unexpectedly is %x\n",
3790 name, cBOOL(isU8_ALPHA_LC(i))));
3792 if (UNLIKELY(cBOOL(isU8_DIGIT_LC(i)) != cBOOL(isDIGIT_A(i)))) {
3794 DEBUG_L(PerlIO_printf(Perl_debug_log,
3795 "isdigit('%s') unexpectedly is %x\n",
3796 name, cBOOL(isU8_DIGIT_LC(i))));
3798 if (UNLIKELY(cBOOL(isU8_GRAPH_LC(i)) != cBOOL(isGRAPH_A(i)))) {
3800 DEBUG_L(PerlIO_printf(Perl_debug_log,
3801 "isgraph('%s') unexpectedly is %x\n",
3802 name, cBOOL(isU8_GRAPH_LC(i))));
3804 if (UNLIKELY(cBOOL(isU8_LOWER_LC(i)) != cBOOL(isLOWER_A(i)))) {
3806 DEBUG_L(PerlIO_printf(Perl_debug_log,
3807 "islower('%s') unexpectedly is %x\n",
3808 name, cBOOL(isU8_LOWER_LC(i))));
3810 if (UNLIKELY(cBOOL(isU8_PRINT_LC(i)) != cBOOL(isPRINT_A(i)))) {
3812 DEBUG_L(PerlIO_printf(Perl_debug_log,
3813 "isprint('%s') unexpectedly is %x\n",
3814 name, cBOOL(isU8_PRINT_LC(i))));
3816 if (UNLIKELY(cBOOL(isU8_PUNCT_LC(i)) != cBOOL(isPUNCT_A(i)))) {
3818 DEBUG_L(PerlIO_printf(Perl_debug_log,
3819 "ispunct('%s') unexpectedly is %x\n",
3820 name, cBOOL(isU8_PUNCT_LC(i))));
3822 if (UNLIKELY(cBOOL(isU8_SPACE_LC(i)) != cBOOL(isSPACE_A(i)))) {
3824 DEBUG_L(PerlIO_printf(Perl_debug_log,
3825 "isspace('%s') unexpectedly is %x\n",
3826 name, cBOOL(isU8_SPACE_LC(i))));
3828 if (UNLIKELY(cBOOL(isU8_UPPER_LC(i)) != cBOOL(isUPPER_A(i)))) {
3830 DEBUG_L(PerlIO_printf(Perl_debug_log,
3831 "isupper('%s') unexpectedly is %x\n",
3832 name, cBOOL(isU8_UPPER_LC(i))));
3834 if (UNLIKELY(cBOOL(isU8_XDIGIT_LC(i))!= cBOOL(isXDIGIT_A(i)))) {
3836 DEBUG_L(PerlIO_printf(Perl_debug_log,
3837 "isxdigit('%s') unexpectedly is %x\n",
3838 name, cBOOL(isU8_XDIGIT_LC(i))));
3840 if (UNLIKELY(toU8_LOWER_LC(i) != (int) toLOWER_A(i))) {
3842 DEBUG_L(PerlIO_printf(Perl_debug_log,
3843 "tolower('%s')=0x%x instead of the expected 0x%x\n",
3844 name, toU8_LOWER_LC(i), (int) toLOWER_A(i)));
3846 if (UNLIKELY(toU8_UPPER_LC(i) != (int) toUPPER_A(i))) {
3848 DEBUG_L(PerlIO_printf(Perl_debug_log,
3849 "toupper('%s')=0x%x instead of the expected 0x%x\n",
3850 name, toU8_UPPER_LC(i), (int) toUPPER_A(i)));
3852 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
3854 DEBUG_L(PerlIO_printf(Perl_debug_log,
3855 "'\\n' (=%02X) is not a control\n", (int) i));
3858 /* Add to the list; Separate multiple entries with a blank */
3861 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
3863 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
3869 if (bad_count == 2 && maybe_utf8_turkic) {
3871 *bad_chars_list = '\0';
3873 /* The casts are because otherwise some compilers warn:
3874 gcc.gnu.org/bugzilla/show_bug.cgi?id=99950
3875 gcc.gnu.org/bugzilla/show_bug.cgi?id=94182
3877 PL_fold_locale[ (U8) 'I' ] = 'I';
3878 PL_fold_locale[ (U8) 'i' ] = 'i';
3879 PL_in_utf8_turkic_locale = TRUE;
3880 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s is turkic\n", newctype));
3883 /* If we found problems and we want them output, do so */
3884 if ( (UNLIKELY(bad_count))
3885 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
3887 /* WARNING. If you change the wording of these; be sure to update
3888 * t/loc_tools.pl correspondingly */
3890 if (PL_in_utf8_CTYPE_locale) {
3891 PL_warn_locale = Perl_newSVpvf(aTHX_
3892 "Locale '%s' contains (at least) the following characters"
3893 " which have\nunexpected meanings: %s\nThe Perl program"
3894 " will use the expected meanings",
3895 newctype, bad_chars_list);
3900 "\nThe following characters (and maybe"
3901 " others) may not have the same meaning as"
3902 " the Perl program expects: %s\n",
3907 # if defined(HAS_SOME_LANGINFO) || defined(WIN32)
3909 char * scratch_buffer = NULL;
3910 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
3911 my_langinfo_c(CODESET, LC_CTYPE,
3913 &scratch_buffer, NULL,
3915 Safefree(scratch_buffer);
3919 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
3921 /* If we are actually in the scope of the locale or are debugging,
3922 * output the message now. If not in that scope, we save the
3923 * message to be output at the first operation using this locale,
3924 * if that actually happens. Most programs don't use locales, so
3925 * they are immune to bad ones. */
3926 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
3928 /* The '0' below suppresses a bogus gcc compiler warning */
3929 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale),
3931 if (IN_LC(LC_CTYPE)) {
3932 SvREFCNT_dec_NN(PL_warn_locale);
3933 PL_warn_locale = NULL;
3941 Perl_warn_problematic_locale()
3945 /* Core-only function that outputs the message in PL_warn_locale,
3946 * and then NULLS it. Should be called only through the macro
3947 * CHECK_AND_WARN_PROBLEMATIC_LOCALE_ */
3949 if (PL_warn_locale) {
3950 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3951 SvPVX(PL_warn_locale),
3952 0 /* dummy to avoid compiler warning */ );
3953 SvREFCNT_dec_NN(PL_warn_locale);
3954 PL_warn_locale = NULL;
3958 # endif /* USE_LOCALE_CTYPE */
3961 S_new_LC_ALL(pTHX_ const char *lc_all, bool force)
3963 PERL_ARGS_ASSERT_NEW_LC_ALL;
3965 /* new_LC_ALL() updates all the things we care about. Note that this is
3966 * called just after a change, so uses the actual underlying locale just
3967 * set, and not the nominal one (should they differ, as they may in
3970 const char * individ_locales[LC_ALL_INDEX_] = { NULL };
3972 switch (parse_LC_ALL_string(lc_all,
3974 override_if_ignored, /* Override any ignored
3976 true, /* Always fill array */
3977 true, /* Panic if fails, as to get here it
3978 earlier had to have succeeded */
3983 case only_element_0:
3984 locale_panic_("Unexpected return from parse_LC_ALL_string");
3990 for_all_individual_category_indexes(i) {
3991 if (update_functions[i]) {
3992 const char * this_locale = individ_locales[i];
3993 update_functions[i](aTHX_ this_locale, force);
3996 Safefree(individ_locales[i]);
4000 # ifdef USE_LOCALE_COLLATE
4003 S_new_collate(pTHX_ const char *newcoll, bool force)
4005 PERL_ARGS_ASSERT_NEW_COLLATE;
4006 PERL_UNUSED_ARG(force);
4008 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
4009 * core Perl this and that 'newcoll' is the name of the new locale.
4011 * The design of locale collation is that every locale change is given an
4012 * index 'PL_collation_ix'. The first time a string participates in an
4013 * operation that requires collation while locale collation is active, it
4014 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
4015 * magic includes the collation index, and the transformation of the string
4016 * by strxfrm(), q.v. That transformation is used when doing comparisons,
4017 * instead of the string itself. If a string changes, the magic is
4018 * cleared. The next time the locale changes, the index is incremented,
4019 * and so we know during a comparison that the transformation is not
4020 * necessarily still valid, and so is recomputed. Note that if the locale
4021 * changes enough times, the index could wrap, and it is possible that a
4022 * transformation would improperly be considered valid, leading to an
4023 * unlikely bug. The value is declared to the widest possible type on this
4026 /* Return if the locale isn't changing */
4027 if (strEQ(PL_collation_name, newcoll)) {
4031 Safefree(PL_collation_name);
4032 PL_collation_name = savepv(newcoll);
4035 /* Set the new one up if trivial. Since this is called at process
4036 * initialization, be aware that this bit can't rely on much being
4038 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
4039 if (PL_collation_standard) {
4040 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4041 "Setting PL_collation name='%s'\n",
4042 PL_collation_name));
4043 PL_collxfrm_base = 0;
4044 PL_collxfrm_mult = 2;
4045 PL_in_utf8_COLLATE_locale = FALSE;
4046 PL_strxfrm_NUL_replacement = '\0';
4047 PL_strxfrm_max_cp = 0;
4051 /* Flag that the remainder of the set up is being deferred until first
4053 PL_collxfrm_mult = 0;
4054 PL_collxfrm_base = 0;
4058 # endif /* USE_LOCALE_COLLATE */
4063 S_Win_byte_string_to_wstring(const UINT code_page, const char * byte_string)
4065 /* Caller must arrange to free the returned string */
4067 int req_size = MultiByteToWideChar(code_page, 0, byte_string, -1, NULL, 0);
4074 Newx(wstring, req_size, wchar_t);
4076 if (! MultiByteToWideChar(code_page, 0, byte_string, -1, wstring, req_size))
4086 # define Win_utf8_string_to_wstring(s) \
4087 Win_byte_string_to_wstring(CP_UTF8, (s))
4090 S_Win_wstring_to_byte_string(const UINT code_page, const wchar_t * wstring)
4092 /* Caller must arrange to free the returned string */
4095 WideCharToMultiByte(code_page, 0, wstring, -1, NULL, 0, NULL, NULL);
4098 Newx(byte_string, req_size, char);
4100 if (! WideCharToMultiByte(code_page, 0, wstring, -1, byte_string,
4101 req_size, NULL, NULL))
4103 Safefree(byte_string);
4111 # define Win_wstring_to_utf8_string(ws) \
4112 Win_wstring_to_byte_string(CP_UTF8, (ws))
4115 S_wrap_wsetlocale(pTHX_ const int category, const char *locale)
4117 PERL_ARGS_ASSERT_WRAP_WSETLOCALE;
4119 /* Calls _wsetlocale(), converting the parameters/return to/from
4120 * Perl-expected forms as if plain setlocale() were being called instead.
4122 * Caller must arrange for the returned PV to be freed.
4125 const wchar_t * wlocale = NULL;
4128 wlocale = Win_utf8_string_to_wstring(locale);
4135 const wchar_t * wresult = _wsetlocale(category, wlocale);
4143 const char * result = Win_wstring_to_utf8_string(wresult);
4151 S_win32_setlocale(pTHX_ int category, const char* locale)
4153 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
4154 * difference between the two unless the input locale is "", which normally
4155 * means on Windows to get the machine default, which is set via the
4156 * computer's "Regional and Language Options" (or its current equivalent).
4157 * In POSIX, it instead means to find the locale from the user's
4158 * environment. This routine changes the Windows behavior to try the POSIX
4159 * behavior first. Further details are in the called function
4160 * find_locale_from_environment().
4163 if (locale != NULL && strEQ(locale, "")) {
4164 /* Note this function may change the locale, but that's ok because we
4165 * are about to change it anyway */
4166 locale = find_locale_from_environment(get_category_index(category));
4167 if (locale == NULL) {
4173 const char * result = wrap_wsetlocale(category, locale);
4174 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s\n",
4175 setlocale_debug_string_r(category, locale, result)));
4182 save_to_buffer(result, &PL_setlocale_buf, &PL_setlocale_bufsize);
4184 # ifndef USE_PL_CUR_LC_ALL
4190 /* Here, we need to keep track of LC_ALL, so store the new value. but if
4191 * the input locale is NULL, we were just querying, so the original value
4193 if (locale == NULL) {
4198 /* If we set LC_ALL directly above, we already know its new value; but
4199 * if we changed just an individual category, find the new LC_ALL */
4200 if (category != LC_ALL) {
4202 result = wrap_wsetlocale(LC_ALL, NULL);
4205 Safefree(PL_cur_LC_ALL);
4206 PL_cur_LC_ALL = result;
4209 DEBUG_L(PerlIO_printf(Perl_debug_log, "new PL_cur_LC_ALL=%s\n",
4213 return PL_setlocale_buf;
4219 S_native_querylocale_i(pTHX_ const locale_category_index cat_index)
4221 /* Determine the current locale and return it in the form the platform's
4222 * native locale handling understands. This is different only from our
4223 * internal form for the LC_ALL category, as platforms differ in how they
4226 * This is only called from Perl_setlocale(). As such it returns in
4227 * PL_setlocale_buf */
4229 # ifdef USE_LOCALE_NUMERIC
4231 /* We have the LC_NUMERIC name saved, because we are normally switched into
4232 * the C locale (or equivalent) for it. */
4233 if (cat_index == LC_NUMERIC_INDEX_) {
4235 /* We don't have to copy this return value, as it is a per-thread
4236 * variable, and won't change until a future setlocale */
4237 return PL_numeric_name;
4243 if (cat_index != LC_ALL_INDEX_)
4248 /* Here, not LC_ALL, and not LC_NUMERIC: the actual and native values
4251 # ifdef setlocale_i /* Can shortcut if this is defined */
4253 return setlocale_i(cat_index, NULL);
4257 return save_to_buffer(querylocale_i(cat_index),
4258 &PL_setlocale_buf, &PL_setlocale_bufsize);
4263 /* Below, querying LC_ALL */
4266 # ifdef USE_PL_CURLOCALES
4267 # define LC_ALL_ARG PL_curlocales
4269 # define LC_ALL_ARG NULL /* Causes calculate_LC_ALL_string() to find the
4270 locale using a querylocale function */
4273 return calculate_LC_ALL_string(LC_ALL_ARG, EXTERNAL_FORMAT_FOR_QUERY,
4274 WANT_PL_setlocale_buf,
4277 # endif /* has LC_ALL */
4281 #endif /* USE_LOCALE */
4284 =for apidoc Perl_setlocale
4286 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
4287 taking the same parameters, and returning the same information, except that it
4288 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
4289 instead return C<C> if the underlying locale has a non-dot decimal point
4290 character, or a non-empty thousands separator for displaying floating point
4291 numbers. This is because perl keeps that locale category such that it has a
4292 dot and empty separator, changing the locale briefly during the operations
4293 where the underlying one is required. C<Perl_setlocale> knows about this, and
4294 compensates; regular C<setlocale> doesn't.
4296 Another reason it isn't completely a drop-in replacement is that it is
4297 declared to return S<C<const char *>>, whereas the system setlocale omits the
4298 C<const> (presumably because its API was specified long ago, and can't be
4299 updated; it is illegal to change the information C<setlocale> returns; doing
4300 so leads to segfaults.)
4302 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
4303 C<setlocale> can be completely ineffective on some platforms under some
4306 Changing the locale is not a good idea when more than one thread is running,
4307 except on systems where the predefined variable C<${^SAFE_LOCALES}> is
4308 non-zero. This is because on such systems the locale is global to the whole
4309 process and not local to just the thread calling the function. So changing it
4310 in one thread instantaneously changes it in all. On some such systems, the
4311 system C<setlocale()> is ineffective, returning the wrong information, and
4312 failing to actually change the locale. z/OS refuses to try to change the
4313 locale once a second thread is created. C<Perl_setlocale>, should give you
4314 accurate results of what actually happened on these problematic platforms,
4315 returning NULL if the system forbade the locale change.
4317 The return points to a per-thread static buffer, which is overwritten the next
4318 time C<Perl_setlocale> is called from the same thread.
4325 Perl_setlocale(const int category, const char * locale)
4327 /* This wraps POSIX::setlocale() */
4331 PERL_UNUSED_ARG(category);
4332 PERL_UNUSED_ARG(locale);
4340 DEBUG_L(PerlIO_printf(Perl_debug_log,
4341 "Entering Perl_setlocale(%d, \"%s\")\n",
4344 bool valid_category;
4345 locale_category_index cat_index = get_category_index_helper(category,
4348 if (! valid_category) {
4349 if (ckWARN(WARN_LOCALE)) {
4350 const char * conditional_warn_text;
4351 if (locale == NULL) {
4352 conditional_warn_text = "";
4356 conditional_warn_text = "; can't set it to ";
4359 /* diag_listed_as: Unknown locale category %d; can't set it to %s */
4361 packWARN(WARN_LOCALE),
4362 "Unknown locale category %d%s%s",
4363 category, conditional_warn_text, locale);
4372 /* setlocale_i() gets defined only on Configurations that use setlocale()
4373 * in a simple manner that adequately handles all cases. If this category
4374 * doesn't have any perl complications, just do that. */
4375 if (! update_functions[cat_index]) {
4376 return setlocale_i(cat_index, locale);
4381 /* Get current locale */
4382 const char * current_locale = native_querylocale_i(cat_index);
4384 /* A NULL locale means only query what the current one is. */
4385 if (locale == NULL) {
4386 return current_locale;
4389 if (strEQ(current_locale, locale)) {
4390 DEBUG_L(PerlIO_printf(Perl_debug_log,
4391 "Already in requested locale: no action taken\n"));
4392 return current_locale;
4395 /* Here, an actual change is being requested. Do it */
4396 if (! bool_setlocale_i(cat_index, locale)) {
4397 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s\n",
4398 setlocale_debug_string_i(cat_index, locale, "NULL")));
4402 /* At this point, the locale has been changed based on the requested value,
4403 * and the querylocale_i() will return the actual new value that the system
4404 * has for the category. That may not be the same as the input, as libc
4405 * may have returned a synonymous locale name instead of the input one; or,
4406 * if there are locale categories that we are compiled to ignore, any
4407 * attempt to change them away from "C" is overruled */
4408 current_locale = querylocale_i(cat_index);
4410 /* But certain categories need further work. For example we may need to
4411 * calculate new folding or collation rules. And for LC_NUMERIC, we have
4412 * to switch into a locale that has a dot radix. */
4413 if (update_functions[cat_index]) {
4414 update_functions[cat_index](aTHX_ current_locale,
4415 /* No need to force recalculation, as
4416 * aren't coming from a situation
4417 * where Perl hasn't been controlling
4418 * the locale, so has accurate
4423 /* Make sure the result is in a stable buffer for the caller's use, and is
4424 * in the expected format */
4425 current_locale = native_querylocale_i(cat_index);
4427 DEBUG_L(PerlIO_printf(Perl_debug_log, "returning '%s'\n", current_locale));
4429 return current_locale;
4435 #if defined(USE_LOCALE) || defined(HAS_SOME_LANGINFO) || defined(HAS_LOCALECONV)
4438 S_get_locale_string_utf8ness_i(pTHX_ const char * string,
4439 const locale_utf8ness_t known_utf8,
4440 const char * locale,
4441 const locale_category_index cat_index)
4443 PERL_ARGS_ASSERT_GET_LOCALE_STRING_UTF8NESS_I;
4448 PERL_UNUSED_ARG(string);
4449 PERL_UNUSED_ARG(known_utf8);
4450 PERL_UNUSED_ARG(locale);
4451 PERL_UNUSED_ARG(cat_index);
4455 assert(cat_index <= LC_ALL_INDEX_);
4457 /* Return to indicate if 'string' in the locale given by the input
4458 * arguments should be considered UTF-8 or not.
4460 * If the input 'locale' is not NULL, use that for the locale; otherwise
4461 * use the current locale for the category specified by 'cat_index'.
4464 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4465 "Entering get_locale_string_utf8ness_i; locale=%s,"
4466 " index=%u(%s), string=%s, known_utf8=%d\n",
4467 locale, cat_index, category_names[cat_index],
4469 ? _byte_dump_string((U8 *) string,
4474 if (string == NULL) {
4475 return UTF8NESS_IMMATERIAL;
4478 if (IN_BYTES) { /* respect 'use bytes' */
4482 Size_t len = strlen(string);
4484 /* UTF8ness is immaterial if the representation doesn't vary */
4485 const U8 * first_variant = NULL;
4486 if (is_utf8_invariant_string_loc((U8 *) string, len, &first_variant)) {
4487 return UTF8NESS_IMMATERIAL;
4490 /* Can't be UTF-8 if invalid */
4491 if (! is_utf8_string((U8 *) first_variant,
4492 len - ((char *) first_variant - string)))
4497 /* Here and below, we know the string is legal UTF-8, containing at least
4498 * one character requiring a sequence of two or more bytes. It is quite
4499 * likely to be UTF-8. But it pays to be paranoid and do further checking.
4501 * If we already know the UTF-8ness of the locale, then we immediately know
4502 * what the string is */
4503 if (UNLIKELY(known_utf8 != LOCALE_UTF8NESS_UNKNOWN)) {
4504 return (known_utf8 == LOCALE_IS_UTF8) ? UTF8NESS_YES : UTF8NESS_NO;
4507 # ifdef HAS_RELIABLE_UTF8NESS_DETERMINATION
4509 /* Here, we have available the libc functions that can be used to
4510 * accurately determine the UTF8ness of the underlying locale. If it is a
4511 * UTF-8 locale, the string is UTF-8; otherwise it was coincidental that
4512 * the string is legal UTF-8
4514 * However, if the perl is compiled to not pay attention to the category
4515 * being passed in, you might think that that locale is essentially always
4516 * the C locale, so it would make sense to say it isn't UTF-8. But to get
4517 * here, the string has to contain characters unknown in the C locale. And
4518 * in fact, Windows boxes are compiled without LC_MESSAGES, as their
4519 * message catalog isn't really a part of the locale system. But those
4520 * messages really could be UTF-8, and given that the odds are rather small
4521 * of something not being UTF-8 but being syntactically valid UTF-8, khw
4522 * has decided to call such strings as UTF-8. */
4524 if (locale == NULL) {
4525 locale = querylocale_i(cat_index);
4528 return (is_locale_utf8(locale)) ? UTF8NESS_YES : UTF8NESS_NO;
4532 /* Here, we have a valid UTF-8 string containing non-ASCII characters, and
4533 * don't have access to functions to check if the locale is UTF-8 or not.
4534 * Assume that it is. khw tried adding a check that the string is entirely
4535 * in a single Unicode script, but discovered the strftime() timezone is
4536 * user-settable through the environment, which may be in a different
4537 * script than the locale-expected value. */
4538 PERL_UNUSED_ARG(locale);
4539 PERL_UNUSED_ARG(cat_index);
4541 return UTF8NESS_YES;
4549 S_is_locale_utf8(pTHX_ const char * locale)
4551 /* Returns TRUE if the locale 'locale' is UTF-8; FALSE otherwise. It uses
4552 * my_langinfo(), which employs various methods to get this information
4553 * if nl_langinfo() isn't available, using heuristics as a last resort, in
4554 * which case, the result will very likely be correct for locales for
4555 * languages that have commonly used non-ASCII characters, but for notably
4556 * English, it comes down to if the locale's name ends in something like
4557 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4559 * Systems conforming to C99 should have the needed libc calls to give us a
4560 * completely reliable result. */
4562 # if ! defined(USE_LOCALE) \
4563 || ! defined(USE_LOCALE_CTYPE) \
4564 || defined(EBCDIC) /* There aren't any real UTF-8 locales at this time */
4566 PERL_UNUSED_ARG(locale);
4572 char * scratch_buffer = NULL;
4573 const char * codeset;
4576 PERL_ARGS_ASSERT_IS_LOCALE_UTF8;
4578 if (strEQ(locale, PL_ctype_name)) {
4579 return PL_in_utf8_CTYPE_locale;
4582 codeset = my_langinfo_c(CODESET, LC_CTYPE, locale,
4583 &scratch_buffer, NULL, NULL);
4584 retval = is_codeset_name_UTF8(codeset);
4586 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4587 "found codeset=%s, is_utf8=%d\n", codeset, retval));
4589 Safefree(scratch_buffer);
4591 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "is_locale_utf8(%s) returning %d\n",
4603 S_set_save_buffer_min_size(pTHX_ Size_t min_len,
4605 Size_t * buf_cursize)
4607 /* Make sure the buffer pointed to by *buf is at least as large 'min_len';
4608 * *buf_cursize is the size of 'buf' upon entry; it will be updated to the
4609 * new size on exit. 'buf_cursize' being NULL is to be used when this is a
4610 * single use buffer, which will shortly be freed by the caller. */
4612 if (buf_cursize == NULL) {
4613 Newx(*buf, min_len, char);
4615 else if (*buf_cursize == 0) {
4616 Newx(*buf, min_len, char);
4617 *buf_cursize = min_len;
4619 else if (min_len > *buf_cursize) {
4620 Renew(*buf, min_len, char);
4621 *buf_cursize = min_len;
4626 S_save_to_buffer(pTHX_ const char * string, char **buf, Size_t *buf_size)
4628 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
4630 /* Copy the NUL-terminated 'string' to a buffer whose address before this
4631 * call began at *buf, and whose available length before this call was
4634 * If the length of 'string' is greater than the space available, the
4635 * buffer is grown accordingly, which may mean that it gets relocated.
4636 * *buf and *buf_size will be updated to reflect this.
4638 * Regardless, the function returns a pointer to where 'string' is now
4641 * 'string' may be NULL, which means no action gets taken, and NULL is
4644 * 'buf_size' being NULL is to be used when this is a single use buffer,
4645 * which will shortly be freed by the caller.
4647 * If *buf or 'buf_size' are NULL or *buf_size is 0, the buffer is assumed
4648 * empty, and memory is malloc'd.
4655 /* No-op to copy over oneself */
4656 if (string == *buf) {
4660 Size_t string_size = strlen(string) + 1;
4661 set_save_buffer_min_size(string_size, buf, buf_size);
4665 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4666 "Copying '%s' to %p\n",
4667 ((is_utf8_string((U8 *) string, 0))
4669 :_byte_dump_string((U8 *) string, strlen(string), 0)),
4672 /* Catch glitches. Usually this is because LC_CTYPE needs to be the same
4673 * locale as whatever is being worked on */
4674 if (UNLIKELY(instr(string, REPLACEMENT_CHARACTER_UTF8))) {
4675 locale_panic_(Perl_form(aTHX_
4676 "Unexpected REPLACEMENT_CHARACTER in '%s'\n%s",
4677 string, get_LC_ALL_display()));
4682 Copy(string, *buf, string_size, char);
4689 Perl_get_win32_message_utf8ness(pTHX_ const char * string)
4691 /* This is because Windows doesn't have LC_MESSAGES. */
4693 # ifdef USE_LOCALE_CTYPE
4695 return get_locale_string_utf8ness_i(string, LOCALE_IS_UTF8,
4696 NULL, LC_CTYPE_INDEX_);
4706 #endif /* USE_LOCALE */
4709 Perl_mbtowc_(pTHX_ const wchar_t * pwc, const char * s, const Size_t len)
4712 #if ! defined(HAS_MBRTOWC) && ! defined(HAS_MBTOWC)
4714 PERL_UNUSED_ARG(pwc);
4716 PERL_UNUSED_ARG(len);
4719 #else /* Below we have some form of mbtowc() */
4720 # if defined(HAS_MBRTOWC) \
4721 && (defined(USE_LOCALE_THREADS) || ! defined(HAS_MBTOWC))
4722 # define USE_MBRTOWC
4729 if (s == NULL) { /* Initialize the shift state to all zeros in
4732 # if defined(USE_MBRTOWC)
4734 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
4741 retval = mbtowc(NULL, NULL, 0);
4749 # if defined(USE_MBRTOWC)
4753 retval = (SSize_t) mbrtowc((wchar_t *) pwc, s, len, &PL_mbrtowc_ps);
4758 /* Locking prevents races, but locales can be switched out without locking,
4759 * so this isn't a cure all */
4762 retval = mbtowc((wchar_t *) pwc, s, len);
4774 =for apidoc Perl_localeconv
4776 This is a thread-safe version of the libc L<localeconv(3)>. It is the same as
4777 L<POSIX::localeconv|POSIX/localeconv> (returning a hash of the C<localeconv()>
4778 fields), but directly callable from XS code.
4784 Perl_localeconv(pTHX)
4787 #if ! defined(HAS_LOCALECONV)
4793 return my_localeconv(0);
4799 #if defined(HAS_LOCALECONV)
4802 S_my_localeconv(pTHX_ const int item)
4804 PERL_ARGS_ASSERT_MY_LOCALECONV;
4806 /* This returns a mortalized hash containing all or certain elements
4807 * returned by localeconv(). It is used by Perl_localeconv() and
4808 * POSIX::localeconv() and is thread-safe.
4810 * There are two use cases:
4811 * 1) Called from POSIX::locale_conv(). This returns the lconv structure
4812 * copied to a hash, based on the current underlying locales for
4813 * LC_NUMERIC and LC_MONETARY. An input item==0 signifies this case, or
4814 * on many platforms it is the only use case compiled.
4815 * 2) Certain items that nl_langinfo() provides are also derivable from
4816 * the return of localeconv(). Windows notably doesn't have
4817 * nl_langinfo(), so on that, and actually any platform lacking it,
4818 * my_localeconv() is used also to emulate it for those particular
4819 * items. The code to do this is compiled only on such platforms.
4820 * Rather than going to the expense of creating a full hash when only
4821 * one item is needed, the returned hash has just the desired item in
4824 * To access all the localeconv() struct lconv fields, there is a data
4825 * structure that contains every commonly documented field in it. (Maybe
4826 * some minority platforms have extra fields. Those could be added here
4827 * without harm; they would just be ignored on platforms lacking them.)
4829 * Our structure is compiled to make looping through the fields easier by
4830 * pointing each name to its value's offset within lconv, e.g.,
4831 { "thousands_sep", STRUCT_OFFSET(struct lconv, thousands_sep) }
4833 # define LCONV_ENTRY(name) \
4834 {STRINGIFY(name), STRUCT_OFFSET(struct lconv, name)}
4836 /* These synonyms are just for clarity, and to make it easier in case
4837 * something needs to change in the future */
4838 # define LCONV_NUMERIC_ENTRY(name) LCONV_ENTRY(name)
4839 # define LCONV_MONETARY_ENTRY(name) LCONV_ENTRY(name)
4841 /* There are just a few fields for NUMERIC strings */
4842 const lconv_offset_t lconv_numeric_strings[] = {
4843 # ifndef NO_LOCALECONV_GROUPING
4844 LCONV_NUMERIC_ENTRY(grouping),
4846 LCONV_NUMERIC_ENTRY(thousands_sep),
4847 LCONV_NUMERIC_ENTRY(decimal_point),
4851 /* When used to implement nl_langinfo(), we save time by only populating
4852 * the hash with the field(s) needed. Thus we would need a data structure
4854 * LCONV_NUMERIC_ENTRY(decimal_point),
4857 * By placing the decimal_point field last in the full structure, we can
4858 * use just the tail for this bit of it, saving space. This macro yields
4859 * the address of the sub structure. */
4860 # define DECIMAL_POINT_ADDRESS \
4861 &lconv_numeric_strings[(C_ARRAY_LENGTH(lconv_numeric_strings) - 2)]
4863 /* And the MONETARY string fields */
4864 const lconv_offset_t lconv_monetary_strings[] = {
4865 LCONV_MONETARY_ENTRY(int_curr_symbol),
4866 LCONV_MONETARY_ENTRY(mon_decimal_point),
4867 # ifndef NO_LOCALECONV_MON_THOUSANDS_SEP
4868 LCONV_MONETARY_ENTRY(mon_thousands_sep),
4870 # ifndef NO_LOCALECONV_MON_GROUPING
4871 LCONV_MONETARY_ENTRY(mon_grouping),
4873 LCONV_MONETARY_ENTRY(positive_sign),
4874 LCONV_MONETARY_ENTRY(negative_sign),
4875 LCONV_MONETARY_ENTRY(currency_symbol),
4879 /* Like above, this field being last can be used as a sub structure */
4880 # define CURRENCY_SYMBOL_ADDRESS \
4881 &lconv_monetary_strings[(C_ARRAY_LENGTH(lconv_monetary_strings) - 2)]
4883 /* Finally there are integer fields, all are for monetary purposes */
4884 const lconv_offset_t lconv_integers[] = {
4885 LCONV_ENTRY(int_frac_digits),
4886 LCONV_ENTRY(frac_digits),
4887 LCONV_ENTRY(p_sep_by_space),
4888 LCONV_ENTRY(n_cs_precedes),
4889 LCONV_ENTRY(n_sep_by_space),
4890 LCONV_ENTRY(p_sign_posn),
4891 LCONV_ENTRY(n_sign_posn),
4892 # ifdef HAS_LC_MONETARY_2008
4893 LCONV_ENTRY(int_p_cs_precedes),
4894 LCONV_ENTRY(int_p_sep_by_space),
4895 LCONV_ENTRY(int_n_cs_precedes),
4896 LCONV_ENTRY(int_n_sep_by_space),
4897 LCONV_ENTRY(int_p_sign_posn),
4898 LCONV_ENTRY(int_n_sign_posn),
4900 LCONV_ENTRY(p_cs_precedes),
4904 /* Like above, this field being last can be used as a sub structure */
4905 # define P_CS_PRECEDES_ADDRESS \
4906 &lconv_integers[(C_ARRAY_LENGTH(lconv_integers) - 2)]
4908 /* If we aren't paying attention to a given category, use LC_CTYPE instead;
4909 * If not paying attention to that either, the code below should end up not
4910 * using this. Make sure that things blow up if that avoidance gets lost,
4911 * by setting the category to an out-of-bounds value */
4912 locale_category_index numeric_index;
4913 locale_category_index monetary_index;
4915 # ifdef USE_LOCALE_NUMERIC
4916 numeric_index = LC_NUMERIC_INDEX_;
4917 # elif defined(USE_LOCALE_CTYPE)
4918 numeric_index = LC_CTYPE_INDEX_;
4920 numeric_index = LC_ALL_INDEX_; /* Out-of-bounds */
4922 # ifdef USE_LOCALE_MONETARY
4923 monetary_index = LC_MONETARY_INDEX_;
4924 # elif defined(USE_LOCALE_CTYPE)
4925 monetary_index = LC_CTYPE_INDEX_;
4927 monetary_index = LC_ALL_INDEX_; /* Out-of-bounds */
4930 /* Some platforms, for correct non-mojibake results, require LC_CTYPE's
4931 * locale to match LC_NUMERIC's for the numeric fields, and LC_MONETARY's
4932 * for the monetary ones. What happens if LC_NUMERIC and LC_MONETARY
4933 * aren't compatible? Wrong results. To avoid that, we call localeconv()
4934 * twice, once for each locale, setting LC_CTYPE to match the category.
4935 * But if the locales of both categories are the same, there is no need for
4936 * a second call. Assume this is the case unless overridden below */
4937 bool requires_2nd_localeconv = false;
4939 /* The actual hash populating is done by S_populate_hash_from_localeconv().
4940 * It gets passed an array of length two containing the data structure it
4941 * is supposed to use to get the key names to fill the hash with. One
4942 * element is always for the NUMERIC strings (or NULL if none to use), and
4943 * the other element similarly for the MONETARY ones. */
4944 # define NUMERIC_STRING_OFFSET 0
4945 # define MONETARY_STRING_OFFSET 1
4946 const lconv_offset_t * strings[2] = { NULL, NULL };
4948 /* This is a mask, with one bit to tell S_populate_hash_from_localeconv to
4949 * populate the NUMERIC items; another bit for the MONETARY ones. This way
4950 * it can choose which (or both) to populate from */
4953 /* This converts from a locale index to its bit position in the above mask.
4955 # define INDEX_TO_BIT(i) (1 << (i))
4957 /* The two categories can have disparate locales. Initialize them to C and
4958 * override later whichever one(s) we pay attention to */
4959 const char * numeric_locale = "C";
4960 const char * monetary_locale = "C";
4962 /* This will be either 'numeric_locale' or 'monetary_locale' depending on
4963 * what we are working on at the moment */
4964 const char * locale;
4966 /* The LC_MONETARY category also has some integer-valued fields, whose
4967 * information is kept in a separate list */
4968 const lconv_offset_t * integers;
4970 # ifdef HAS_SOME_LANGINFO
4972 /* If the only use-case for this is the full localeconv(), the 'item'
4973 * parameter is ignored. */
4974 PERL_UNUSED_ARG(item);
4978 /* This only gets compiled for the use-case of using localeconv() to
4979 * emulate an nl_langinfo() missing from the platform. */
4981 # ifdef USE_LOCALE_NUMERIC
4983 /* We need this substructure to only return this field for the THOUSEP
4984 * item. The other items also need substructures, but they were handled
4985 * above by placing the substructure's item at the end of the full one, so
4986 * the data structure could do double duty. However, both this and
4987 * RADIXCHAR would need to be in the final position of the same full
4988 * structure; an impossibility. So make this into a separate structure */
4989 const lconv_offset_t thousands_sep_string[] = {
4990 LCONV_NUMERIC_ENTRY(thousands_sep),
4996 /* End of all the initialization of data structures. Now for actual code.
4998 * Without nl_langinfo(), the call to my_localeconv() could be for just one
4999 * of the following 3 items to emulate nl_langinfo(). This is compiled
5000 * only when using perl_langinfo.h, which we control, and it has been
5001 * constructed so that no item is numbered 0.
5003 * For each, set up the appropriate parameters for the call below to
5004 * S_populate_hash_from_localeconv() */
5005 if (item != 0) switch (item) {
5007 locale_panic_(Perl_form(aTHX_
5008 "Unexpected item passed to my_localeconv: %d", item));
5011 # ifdef USE_LOCALE_NUMERIC
5014 locale = numeric_locale = PL_numeric_name;
5015 index_bits = INDEX_TO_BIT(LC_NUMERIC_INDEX_);
5016 strings[NUMERIC_STRING_OFFSET] = DECIMAL_POINT_ADDRESS;
5021 index_bits = INDEX_TO_BIT(LC_NUMERIC_INDEX_);
5022 locale = numeric_locale = PL_numeric_name;
5023 strings[NUMERIC_STRING_OFFSET] = thousands_sep_string;
5028 # ifdef USE_LOCALE_MONETARY
5031 index_bits = INDEX_TO_BIT(LC_MONETARY_INDEX_);
5032 locale = monetary_locale = querylocale_i(LC_MONETARY_INDEX_);
5034 /* This item needs the values for both the currency symbol, and another
5035 * one used to construct the nl_langino()-compatible return */
5036 strings[MONETARY_STRING_OFFSET] = CURRENCY_SYMBOL_ADDRESS;
5037 integers = P_CS_PRECEDES_ADDRESS;
5042 } /* End of switch() */
5044 else /* End of for just one item to emulate nl_langinfo() */
5048 { /* Here, the call is for all of localeconv(). It has a bunch of
5049 * items. As in the individual item case, set up the parameters for
5050 * S_populate_hash_from_localeconv(); */
5052 # ifdef USE_LOCALE_NUMERIC
5053 numeric_locale = PL_numeric_name;
5054 # elif defined(USE_LOCALE_CTYPE)
5055 numeric_locale = querylocale_i(numeric_index);
5057 # if defined(USE_LOCALE_MONETARY) || defined(USE_LOCALE_CTYPE)
5058 monetary_locale = querylocale_i(monetary_index);
5061 /* The first call to S_populate_hash_from_localeconv() will be for the
5062 * MONETARY values */
5063 index_bits = INDEX_TO_BIT(monetary_index);
5064 locale = monetary_locale;
5066 /* And if the locales for the two categories are the same, we can also
5067 * do the NUMERIC values in the same call */
5068 if (strEQ(numeric_locale, monetary_locale)) {
5069 index_bits |= INDEX_TO_BIT(numeric_index);
5072 requires_2nd_localeconv = true;
5075 /* We always pass both sets of strings. 'index_bits' tells
5076 * S_populate_hash_from_localeconv which to actually look at */
5077 strings[NUMERIC_STRING_OFFSET] = lconv_numeric_strings;
5078 strings[MONETARY_STRING_OFFSET] = lconv_monetary_strings;
5080 /* And pass the integer values to populate; again 'index_bits' will
5081 * say to use them or not */
5082 integers = lconv_integers;
5084 } /* End of call is for localeconv() */
5086 /* The code above has determined the parameters to
5087 S_populate_hash_from_localeconv() for both cases of an individual item
5088 and for the entire structure. Below is code common to both */
5090 HV * hv = newHV(); /* The returned hash, initially empty */
5091 sv_2mortal((SV*)hv);
5093 /* Call localeconv() and copy its results into the hash. All the
5094 * parameters have been initialized above */
5095 populate_hash_from_localeconv(hv,
5102 /* The above call may have done all the hash fields, but not always, as
5103 * already explained. If we need a second call it is always for the
5105 if (requires_2nd_localeconv) {
5106 populate_hash_from_localeconv(hv,
5108 INDEX_TO_BIT(numeric_index),
5110 NULL /* There are no NUMERIC integer
5115 /* Here, the hash has been completely populated.
5117 * Now go through all the items and:
5118 * a) For string items, see if they should be marked as UTF-8 or not.
5119 * This would have been more convenient and faster to do while
5120 * populating the hash in the first place, but that operation has to be
5121 * done within a critical section, keeping other threads from
5122 * executing, so only the minimal amount of work necessary is done at
5124 * b) For integer items, convert the C CHAR_MAX value into -1. Again,
5125 * this could have been done in the critical section, but was deferred
5126 * to here to keep to the bare minimum amount the time spent owning the
5127 * processor. CHAR_MAX is a C concept for an 8-bit character type.
5128 * Perl has no such type; the closest fit is a -1.
5130 * XXX On unthreaded perls, this code could be #ifdef'd out, and the
5131 * corrections determined at hash population time, at an extra maintenance
5132 * cost which khw doesn't think is worth it
5134 for (unsigned int i = 0; i < 2; i++) { /* Try both types of strings */
5135 if (! strings[i]) { /* Skip if no strings of this type */
5139 locale = (i == NUMERIC_STRING_OFFSET)
5143 locale_utf8ness_t locale_is_utf8 = LOCALE_UTF8NESS_UNKNOWN;
5145 # ifdef HAS_RELIABLE_UTF8NESS_DETERMINATION
5147 /* It saves time in the loop below to have predetermined the UTF8ness
5148 * of the locale. But only do so if the platform reliably has this
5149 * information; otherwise it's better to do it only it should become
5150 * necessary, which happens on a per-element basis in the loop. */
5152 locale_is_utf8 = (is_locale_utf8(locale))
5156 if (locale_is_utf8 == LOCALE_NOT_UTF8) {
5157 continue; /* No string can be UTF-8 if the locale isn't */
5162 /* Examine each string */
5163 for (const lconv_offset_t *strp = strings[i]; strp->name; strp++) {
5164 const char * name = strp->name;
5166 /* 'value' will contain the string that may need to be marked as
5168 SV ** value = hv_fetch(hv, name, strlen(name), true);
5169 if (! value || ! SvPOK(*value)) {
5173 /* Determine if the string should be marked as UTF-8. */
5174 if (UTF8NESS_YES == (get_locale_string_utf8ness_i(SvPVX(*value),
5177 (locale_category_index) 0)))
5182 } /* End of fixing up UTF8ness */
5185 /* Examine each integer */
5186 for (; integers; integers++) {
5187 const char * name = integers->name;
5189 if (! name) { /* Reached the end */
5193 SV ** value = hv_fetch(hv, name, strlen(name), true);
5198 /* Change CHAR_MAX to -1 */
5199 if (SvIV(*value) == CHAR_MAX) {
5200 sv_setiv(*value, -1);
5208 S_populate_hash_from_localeconv(pTHX_ HV * hv,
5210 /* Switch to this locale to run
5211 * localeconv() from */
5212 const char * locale,
5214 /* bit mask of which categories to
5216 const U32 which_mask,
5218 /* strings[0] points to the numeric
5219 * string fields; [1] to the monetary */
5220 const lconv_offset_t * strings[2],
5222 /* And to the monetary integer fields */
5223 const lconv_offset_t * integers)
5225 PERL_ARGS_ASSERT_POPULATE_HASH_FROM_LOCALECONV;
5226 PERL_UNUSED_ARG(which_mask); /* Some configurations don't use this;
5227 complicated to figure out which */
5229 PERL_UNUSED_ARG(locale);
5232 /* Run localeconv() and copy some or all of its results to the input 'hv'
5233 * hash. Most localeconv() implementations return the values in a global
5234 * static buffer, so the operation must be performed in a critical section,
5235 * ending only after the copy is completed. There are so many locks
5236 * because localeconv() deals with two categories, and returns in a single
5237 * global static buffer. Some locks might be no-ops on this platform, but
5238 * not others. We need to lock if any one isn't a no-op. */
5240 # ifdef USE_LOCALE_CTYPE
5242 /* Some platforms require LC_CTYPE to be congruent with the category we are
5244 const char * orig_CTYPE_locale = toggle_locale_c(LC_CTYPE, locale);
5247 # ifdef USE_LOCALE_NUMERIC
5249 /* We need to toggle to the underlying NUMERIC locale if we are getting
5250 * NUMERIC strings */
5251 const char * orig_NUMERIC_locale = NULL;
5252 if (which_mask & INDEX_TO_BIT(LC_NUMERIC_INDEX_)) {
5257 /* There is a bug in Windows in which setting LC_CTYPE after the others
5258 * doesn't actually take effect for localeconv(). See commit
5259 * 418efacd1950763f74ed3cc22f8cf9206661b892 for details. Thus we have
5260 * to make sure that the locale we want is set after LC_CTYPE. We
5261 * unconditionally toggle away from and back to the current locale
5262 * prior to calling localeconv().
5264 * This code will have no effect if we already are in C, but khw
5265 * hasn't seen any cases where this causes problems when we are in the
5267 orig_NUMERIC_locale = toggle_locale_i(LC_NUMERIC_INDEX_, "C");
5268 toggle_locale_i(LC_NUMERIC_INDEX_, locale);
5272 /* No need for the extra toggle when not on Windows */
5273 orig_NUMERIC_locale = toggle_locale_i(LC_NUMERIC_INDEX_, locale);
5280 # if defined(USE_LOCALE_MONETARY) && defined(WIN32)
5282 /* Same Windows bug as described just above for NUMERIC. Otherwise, no
5283 * need to toggle LC_MONETARY, as it is kept in the underlying locale */
5284 const char * orig_MONETARY_locale = NULL;
5285 if (which_mask & INDEX_TO_BIT(LC_MONETARY_INDEX_)) {
5286 orig_MONETARY_locale = toggle_locale_i(LC_MONETARY_INDEX_, "C");
5287 toggle_locale_i(LC_MONETARY_INDEX_, locale);
5292 /* Finally ready to do the actual localeconv(). Lock to prevent other
5293 * accesses until we have made a copy of its returned static buffer */
5296 # if defined(TS_W32_BROKEN_LOCALECONV) && defined(USE_THREAD_SAFE_LOCALE)
5298 /* This is a workaround for another bug in Windows. localeconv() was
5299 * broken with thread-safe locales prior to VS 15. It looks at the global
5300 * locale instead of the thread one. As a work-around, we toggle to the
5301 * global locale; populate the return; then toggle back. We have to use
5302 * LC_ALL instead of the individual categories because of yet another bug
5303 * in Windows. And this all has to be done in a critical section.
5305 * This introduces a potential race with any other thread that has also
5306 * converted to use the global locale, and doesn't protect its locale calls
5307 * with mutexes. khw can't think of any reason for a thread to do so on
5308 * Windows, as the locale API is the same regardless of thread-safety,
5309 * except if the code is ported from working on another platform where
5310 * there might be some reason to do this. But this is typically due to
5311 * some alien-to-Perl library that thinks it owns locale setting. Such a
5312 * library isn't likely to exist on Windows, so such an application is
5313 * unlikely to be run on Windows
5315 bool restore_per_thread = FALSE;
5317 /* Save the per-thread locale state */
5318 const char * save_thread = querylocale_c(LC_ALL);
5320 /* Change to the global locale, and note if we already were there */
5321 int config_return = _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5322 if (config_return != _DISABLE_PER_THREAD_LOCALE) {
5323 if (config_return == -1) {
5324 locale_panic_("_configthreadlocale returned an error");
5327 restore_per_thread = TRUE;
5330 /* Save the state of the global locale; then convert to our desired
5332 const char * save_global = querylocale_c(LC_ALL);
5333 void_setlocale_c(LC_ALL, save_thread);
5335 # endif /* TS_W32_BROKEN_LOCALECONV */
5337 /* Finally, do the actual localeconv */
5338 const char *lcbuf_as_string = (const char *) localeconv();
5340 /* Fill in the string fields of the HV* */
5341 for (unsigned int i = 0; i < 2; i++) {
5343 /* One iteration is only for the numeric string fields. Skip these
5344 * unless we are compiled to care about those fields and the input
5345 * parameters indicate we want their values */
5346 if ( i == NUMERIC_STRING_OFFSET
5348 # ifdef USE_LOCALE_NUMERIC
5350 && (which_mask & INDEX_TO_BIT(LC_NUMERIC_INDEX_)) == 0
5358 /* The other iteration is only for the monetary string fields. Again
5359 * skip it unless we want those values */
5360 if ( i == MONETARY_STRING_OFFSET
5362 # ifdef USE_LOCALE_MONETARY
5364 && (which_mask & INDEX_TO_BIT(LC_MONETARY_INDEX_)) == 0
5372 /* For each field for the given category ... */
5373 const lconv_offset_t * category_strings = strings[i];
5375 const char * name = category_strings->name;
5376 if (! name) { /* Quit at the end */
5380 /* we have set things up so that we know where in the returned
5381 * structure, when viewed as a string, the corresponding value is.
5383 const char *value = *((const char **)( lcbuf_as_string
5384 + category_strings->offset));
5386 /* Set to get next string on next iteration */
5389 /* Skip if this platform doesn't have this field. */
5394 /* Copy to the hash */
5397 newSVpv(value, strlen(value)),
5401 /* Add any int fields to the HV* */
5402 if (i == MONETARY_STRING_OFFSET && integers) {
5403 while (integers->name) {
5404 const char value = *((const char *)( lcbuf_as_string
5405 + integers->offset));
5406 (void) hv_store(hv, integers->name,
5407 strlen(integers->name), newSViv(value), 0);
5411 } /* End of loop through the fields */
5413 /* Done with copying to the hash. Can unwind the critical section locks */
5415 # if defined(TS_W32_BROKEN_LOCALECONV) && defined(USE_THREAD_SAFE_LOCALE)
5417 /* Restore the global locale's prior state */
5418 void_setlocale_c(LC_ALL, save_global);
5420 /* And back to per-thread locales */
5421 if (restore_per_thread) {
5422 if (_configthreadlocale(_ENABLE_PER_THREAD_LOCALE) == -1) {
5423 locale_panic_("_configthreadlocale returned an error");
5427 /* Restore the per-thread locale state */
5428 void_setlocale_c(LC_ALL, save_thread);
5430 # endif /* TS_W32_BROKEN_LOCALECONV */
5432 gwLOCALE_UNLOCK; /* Finished with the critical section of a
5433 globally-accessible buffer */
5435 # if defined(USE_LOCALE_MONETARY) && defined(WIN32)
5437 restore_toggled_locale_i(LC_MONETARY_INDEX_, orig_MONETARY_locale);
5440 # ifdef USE_LOCALE_NUMERIC
5442 restore_toggled_locale_i(LC_NUMERIC_INDEX_, orig_NUMERIC_locale);
5443 if (which_mask & INDEX_TO_BIT(LC_NUMERIC_INDEX_)) {
5448 # ifdef USE_LOCALE_CTYPE
5450 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_locale);
5456 #endif /* defined(HAS_LOCALECONV) */
5457 #ifndef HAS_SOME_LANGINFO
5459 typedef int nl_item; /* Substitute 'int' for emulated nl_langinfo() */
5465 =for apidoc Perl_langinfo
5466 =for apidoc_item Perl_langinfo8
5468 C<Perl_langinfo> is an (almost) drop-in replacement for the system
5469 C<L<nl_langinfo(3)>>, taking the same C<item> parameter values, and returning
5470 the same information. But it is more thread-safe than regular
5471 C<nl_langinfo()>, and hides the quirks of Perl's locale handling from your
5472 code, and can be used on systems that lack a native C<nl_langinfo>.
5474 However, you should instead use the improved version of this:
5475 L</Perl_langinfo8>, which behaves identically except for an additional
5476 parameter, a pointer to a variable declared as L</C<utf8ness_t>>, into which it
5477 returns to you how you should treat the returned string with regards to it
5478 being encoded in UTF-8 or not.
5480 Concerning the differences between these and plain C<nl_langinfo()>:
5486 C<Perl_langinfo8> has an extra parameter, described above. Besides this, the
5487 other reason they aren't quite a drop-in replacement is actually an advantage.
5488 The C<const>ness of the return allows the compiler to catch attempts to write
5489 into the returned buffer, which is illegal and could cause run-time crashes.
5493 They deliver the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
5494 without you having to write extra code. The reason for the extra code would be
5495 because these are from the C<LC_NUMERIC> locale category, which is normally
5496 kept set by Perl so that the radix is a dot, and the separator is the empty
5497 string, no matter what the underlying locale is supposed to be, and so to get
5498 the expected results, you have to temporarily toggle into the underlying
5499 locale, and later toggle back. (You could use plain C<nl_langinfo> and
5500 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
5501 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
5502 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
5503 (decimal point) character to be a dot.)
5507 The system function they replace can have its static return buffer trashed,
5508 not only by a subsequent call to that function, but by a C<freelocale>,
5509 C<setlocale>, or other locale change. The returned buffer of these functions
5510 is not changed until the next call to one or the other, so the buffer is never
5515 The return buffer is per-thread, so it also is never overwritten by a call to
5516 these functions from another thread; unlike the function it replaces.
5520 But most importantly, they work on systems that don't have C<nl_langinfo>, such
5521 as Windows, hence making your code more portable. Of the fifty-some possible
5522 items specified by the POSIX 2008 standard,
5523 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
5524 only one is completely unimplemented, though on non-Windows platforms, another
5525 significant one is not fully implemented). They use various techniques to
5526 recover the other items, including calling C<L<localeconv(3)>>, and
5527 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
5528 available. Later C<strftime()> versions have additional capabilities.
5529 If an item is not available on your system, this returns either the value
5530 associated with the C locale, or simply C<"">, whichever is more appropriate.
5532 It is important to note that, when called with an item that is recovered by
5533 using C<localeconv>, the buffer from any previous explicit call to
5534 C<L<localeconv(3)>> will be overwritten. But you shouldn't be using
5535 C<localeconv> anyway because it is is very much not thread-safe, and suffers
5536 from the same problems outlined in item 'b.' above for the fields it returns
5537 that are controlled by the LC_NUMERIC locale category. Instead, avoid all of
5538 those problems by calling L</Perl_localeconv>, which is thread-safe; or by
5539 using the methods given in L<perlcall> to call
5540 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is also thread-safe.
5544 The details for those items which may deviate from what this emulation returns
5545 and what a native C<nl_langinfo()> would return are specified in
5548 When using C<Perl_langinfo8> (or plain C<Perl_langinfo>) on systems that don't
5549 have a native C<nl_langinfo()>, you must
5551 #include "perl_langinfo.h"
5553 before the C<perl.h> C<#include>. You can replace your F<langinfo.h>
5554 C<#include> with this one. (Doing it this way keeps out the symbols that plain
5555 F<langinfo.h> would try to import into the namespace for code that doesn't need
5563 Perl_langinfo(const nl_item item)
5565 return Perl_langinfo8(item, NULL);
5569 Perl_langinfo8(const nl_item item, utf8ness_t * utf8ness)
5572 locale_category_index cat_index;
5574 PERL_ARGS_ASSERT_PERL_LANGINFO8;
5576 if (utf8ness) { /* Assume for now */
5577 *utf8ness = UTF8NESS_IMMATERIAL;
5580 /* Find the locale category that controls the input 'item'. If we are not
5581 * paying attention to that category, instead return a default value. Also
5582 * return the default value if there is no way for us to figure out the
5583 * correct value. If we have some form of nl_langinfo(), we can always
5584 * figure it out, but lacking that, there may be alternative methods that
5585 * can be used to recover most of the possible items. Some of those
5586 * methods need libc functions, which may or may not be available. If
5587 * unavailable, we can't compute the correct value, so must here return the
5593 #ifdef USE_LOCALE_CTYPE
5595 cat_index = LC_CTYPE_INDEX_;
5601 #if defined(USE_LOCALE_MESSAGES) && defined(HAS_SOME_LANGINFO)
5603 case YESEXPR: case YESSTR: case NOEXPR: case NOSTR:
5604 cat_index = LC_MESSAGES_INDEX_;
5607 case YESEXPR: return "^[+1yY]";
5608 case YESSTR: return "yes";
5609 case NOEXPR: return "^[-0nN]";
5610 case NOSTR: return "no";
5615 #if defined(USE_LOCALE_MONETARY) \
5616 && (defined(HAS_SOME_LANGINFO) || defined(HAS_LOCALECONV))
5618 cat_index = LC_MONETARY_INDEX_;
5626 #ifdef CAN_CALCULATE_RADIX
5628 cat_index = LC_NUMERIC_INDEX_;
5631 return C_decimal_point;
5636 #if defined(USE_LOCALE_NUMERIC) \
5637 && (defined(HAS_SOME_LANGINFO) || defined(HAS_LOCALECONV))
5639 cat_index = LC_NUMERIC_INDEX_;
5642 return C_thousands_sep;
5645 /* The other possible items are all in LC_TIME. */
5646 #ifdef USE_LOCALE_TIME
5649 cat_index = LC_TIME_INDEX_;
5653 #if ! defined(USE_LOCALE_TIME) || ! defined(HAS_SOME_LANGINFO)
5655 /* If not using LC_TIME, hard code the rest. Or, if there is no
5656 * nl_langinfo(), we use strftime() as an alternative, and it is missing
5657 * functionality to get every single one, so hard-code those */
5659 case ERA: return ""; /* Unimplemented; for use with strftime() %E
5662 /* These formats are defined by C89, so we assume that strftime supports
5663 * them, and so are returned unconditionally; they may not be what the
5664 * locale actually says, but should give good enough results for someone
5665 * using them as formats (as opposed to trying to parse them to figure
5666 * out what the locale says). The other format items are actually tested
5667 * to verify they work on the platform */
5668 case D_FMT: return "%x";
5669 case T_FMT: return "%X";
5670 case D_T_FMT: return "%c";
5672 # if defined(WIN32) || ! defined(USE_LOCALE_TIME)
5674 /* strftime() on Windows doesn't have the POSIX (beyond C89) extensions
5675 * that would allow it to recover these */
5676 case ERA_D_FMT: return "%x";
5677 case ERA_T_FMT: return "%X";
5678 case ERA_D_T_FMT: return "%c";
5679 case ALT_DIGITS: return "0";
5682 # ifndef USE_LOCALE_TIME
5684 case T_FMT_AMPM: return "%r";
5685 case ABDAY_1: return "Sun";
5686 case ABDAY_2: return "Mon";
5687 case ABDAY_3: return "Tue";
5688 case ABDAY_4: return "Wed";
5689 case ABDAY_5: return "Thu";
5690 case ABDAY_6: return "Fri";
5691 case ABDAY_7: return "Sat";
5692 case AM_STR: return "AM";
5693 case PM_STR: return "PM";
5694 case ABMON_1: return "Jan";
5695 case ABMON_2: return "Feb";
5696 case ABMON_3: return "Mar";
5697 case ABMON_4: return "Apr";
5698 case ABMON_5: return "May";
5699 case ABMON_6: return "Jun";
5700 case ABMON_7: return "Jul";
5701 case ABMON_8: return "Aug";
5702 case ABMON_9: return "Sep";
5703 case ABMON_10: return "Oct";
5704 case ABMON_11: return "Nov";
5705 case ABMON_12: return "Dec";
5706 case DAY_1: return "Sunday";
5707 case DAY_2: return "Monday";
5708 case DAY_3: return "Tuesday";
5709 case DAY_4: return "Wednesday";
5710 case DAY_5: return "Thursday";
5711 case DAY_6: return "Friday";
5712 case DAY_7: return "Saturday";
5713 case MON_1: return "January";
5714 case MON_2: return "February";
5715 case MON_3: return "March";
5716 case MON_4: return "April";
5717 case MON_5: return "May";
5718 case MON_6: return "June";
5719 case MON_7: return "July";
5720 case MON_8: return "August";
5721 case MON_9: return "September";
5722 case MON_10: return "October";
5723 case MON_11: return "November";
5724 case MON_12: return "December";
5729 } /* End of switch on item */
5733 Perl_croak_nocontext("panic: Unexpected nl_langinfo() item %d", item);
5734 NOT_REACHED; /* NOTREACHED */
5735 PERL_UNUSED_VAR(cat_index);
5739 return my_langinfo_i(item, cat_index, query_nominal_locale_i(cat_index),
5740 &PL_langinfo_buf, &PL_langinfo_bufsize, utf8ness);
5746 Perl_my_strftime(pTHX_ const char *fmt, int sec, int min, int hour,
5747 int mday, int mon, int year, int wday, int yday,
5749 { /* Documented above */
5750 PERL_ARGS_ASSERT_MY_STRFTIME;
5753 ints_to_tm(&mytm, sec, min, hour, mday, mon, year, wday, yday, isdst);
5754 char * ret = strftime_tm(fmt, &mytm);
5759 Perl_sv_strftime_tm(pTHX_ SV * fmt, const struct tm * mytm)
5760 { /* Documented above */
5761 PERL_ARGS_ASSERT_SV_STRFTIME_TM;
5763 utf8ness_t fmt_utf8ness = (SvUTF8(fmt) && LIKELY(! IN_BYTES))
5767 utf8ness_t result_utf8ness;
5768 char * retval = strftime8(SvPV_nolen(fmt),
5772 true /* calling from sv_strftime */
5776 STRLEN len = strlen(retval);
5778 sv_usepvn_flags(sv, retval, len, SV_HAS_TRAILING_NUL);
5780 if (result_utf8ness == UTF8NESS_YES) {
5789 Perl_sv_strftime_ints(pTHX_ SV * fmt, int sec, int min, int hour,
5790 int mday, int mon, int year, int wday,
5791 int yday, int isdst)
5792 { /* Documented above */
5793 PERL_ARGS_ASSERT_SV_STRFTIME_INTS;
5796 ints_to_tm(&mytm, sec, min, hour, mday, mon, year, wday, yday, isdst);
5797 SV * ret = sv_strftime_tm(fmt, &mytm);
5803 /* There are several implementations of my_langinfo, depending on the
5804 * Configuration. They all share the same beginning of the function */
5806 S_my_langinfo_i(pTHX_
5807 const nl_item item, /* The item to look up */
5808 const locale_category_index cat_index, /* The locale category
5810 /* The locale to look up 'item' in. */
5811 const char * locale,
5813 /* Where to store the result, and where the size of that buffer
5814 * is stored, updated on exit. retbuf_sizep may be NULL for an
5815 * empty-on-entry, single use buffer whose size we don't need
5816 * to keep track of */
5818 Size_t * retbuf_sizep,
5820 /* If not NULL, the location to store the UTF8-ness of 'item's
5821 * value, as documented */
5822 utf8ness_t * utf8ness)
5824 const char * retval = NULL;
5826 PERL_ARGS_ASSERT_MY_LANGINFO_I;
5827 assert(cat_index < LC_ALL_INDEX_);
5829 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5830 "Entering my_langinfo item=%ld, using locale %s\n",
5831 (long) item, locale));
5832 /*--------------------------------------------------------------------------*/
5833 /* Above is the common beginning to all the implementations of my_langinfo().
5834 * Below are the various completions.
5836 * Some platforms don't deal well with non-ASCII strings in locale X when
5837 * LC_CTYPE is not in X. (Actually it is probably when X is UTF-8 and LC_CTYPE
5838 * isn't, or vice versa). There is explicit code to bring the categories into
5839 * sync. This doesn't seem to be a problem with nl_langinfo(), so that
5840 * implementation doesn't currently worry about it. But it is a problem on
5841 * Windows boxes, which don't have nl_langinfo(). */
5843 /*--------------------------------------------------------------------------*/
5844 # if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
5845 # ifdef USE_LOCALE_CTYPE
5847 /* This function sorts out if things actually have to be switched or not,
5848 * for both save and restore. */
5849 const char * orig_CTYPE_locale = toggle_locale_c(LC_CTYPE, locale);
5853 const char * orig_switched_locale = toggle_locale_i(cat_index, locale);
5856 retval = save_to_buffer(nl_langinfo(item), retbufp, retbuf_sizep);
5860 *utf8ness = get_locale_string_utf8ness_i(retval,
5861 LOCALE_UTF8NESS_UNKNOWN,
5865 restore_toggled_locale_i(cat_index, orig_switched_locale);
5867 # ifdef USE_LOCALE_CTYPE
5869 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_locale);
5874 /*--------------------------------------------------------------------------*/
5875 # else /* Below, emulate nl_langinfo as best we can */
5877 /* The other completion is where we have to emulate nl_langinfo(). There
5878 * are various possibilities depending on the Configuration */
5880 # ifdef USE_LOCALE_CTYPE
5882 const char * orig_CTYPE_locale = toggle_locale_c(LC_CTYPE, locale);
5886 const char * orig_switched_locale = toggle_locale_i(cat_index, locale);
5888 /* Here, we are in the locale we want information about */
5890 /* Almost all the items will have ASCII return values. Set that here, and
5891 * override if necessary */
5892 utf8ness_t is_utf8 = UTF8NESS_IMMATERIAL;
5896 assert(item < 0); /* Make sure using perl_langinfo.h */
5902 # if defined(HAS_SNPRINTF) \
5903 && (! defined(HAS_LOCALECONV) || defined(TS_W32_BROKEN_LOCALECONV))
5906 /* snprintf() can be used to find the radix character by outputting
5907 * a known simple floating point number to a buffer, and parsing
5908 * it, inferring the radix as the bytes separating the integer and
5909 * fractional parts. But localeconv() is more direct, not
5910 * requiring inference, so use it instead of the code just below,
5911 * if (likely) it is available and works ok */
5913 char * floatbuf = NULL;
5914 const Size_t initial_size = 10;
5916 Newx(floatbuf, initial_size, char);
5918 /* 1.5 is exactly representable on binary computers */
5919 Size_t needed_size = snprintf(floatbuf, initial_size, "%.1f", 1.5);
5921 /* If our guess wasn't big enough, increase and try again, based on
5922 * the real number that snprintf() is supposed to return */
5923 if (UNLIKELY(needed_size >= initial_size)) {
5924 needed_size++; /* insurance */
5925 Renew(floatbuf, needed_size, char);
5926 Size_t new_needed = snprintf(floatbuf, needed_size, "%.1f", 1.5);
5927 assert(new_needed <= needed_size);
5928 needed_size = new_needed;
5931 char * s = floatbuf;
5932 char * e = floatbuf + needed_size;
5935 while (s < e && *s != '1') {
5939 if (LIKELY(s < e)) {
5944 char * item_start = s;
5945 while (s < e && *s != '5') {
5949 /* Everything in between is the radix string */
5950 if (LIKELY(s < e)) {
5952 retval = save_to_buffer(item_start, retbufp, retbuf_sizep);
5956 is_utf8 = get_locale_string_utf8ness_i(retval,
5957 LOCALE_UTF8NESS_UNKNOWN,
5967 # ifdef HAS_LOCALECONV /* snprintf() failed; drop down to use
5972 # else /* snprintf() failed and no localeconv() */
5974 retval = C_decimal_point;
5979 # ifdef HAS_LOCALECONV
5981 /* These items are available from localeconv(). (To avoid using
5982 * TS_W32_BROKEN_LOCALECONV, one could use GetNumberFormat and
5983 * GetCurrencyFormat; patches welcome) */
5985 # define P_CS_PRECEDES "p_cs_precedes"
5986 # define CURRENCY_SYMBOL "currency_symbol"
5988 /* case RADIXCHAR: // May drop down to here in some configurations */
5993 /* The hash gets populated with just the field(s) related to 'item'. */
5994 HV * result_hv = my_localeconv(item);
5997 if (item != CRNCYSTR) {
5999 /* These items have been populated with just one key => value */
6000 (void) hv_iterinit(result_hv);
6001 HE * entry = hv_iternext(result_hv);
6002 string = hv_iterval(result_hv, entry);
6006 /* But CRNCYSTR localeconv() returns a slightly different value
6007 * than the nl_langinfo() API calls for, so have to modify this one
6008 * to conform. We need another value from localeconv() to know
6009 * what to change it to. my_localeconv() has populated the hash
6010 * with exactly both fields. Delete this one, leaving just the
6011 * CRNCYSTR one in the hash */
6012 SV* precedes = hv_delete(result_hv,
6013 P_CS_PRECEDES, STRLENs(P_CS_PRECEDES),
6016 locale_panic_("my_localeconv() unexpectedly didn't return"
6017 " a value for " P_CS_PRECEDES);
6020 /* The modification is to prefix the localeconv() return with a
6021 * single byte, calculated as follows: */
6022 char prefix = (LIKELY(SvIV(precedes) != -1))
6023 ? ((precedes != 0) ? '-' : '+')
6025 /* khw couldn't find any documentation that
6026 * CHAR_MAX (which we modify to -1) is the signal,
6027 * but cygwin uses it thusly, and it makes sense
6028 * given that CHAR_MAX indicates the value isn't
6029 * used, so it neither precedes nor succeeds */
6032 /* Now get CRNCYSTR */
6033 (void) hv_iterinit(result_hv);
6034 HE * entry = hv_iternext(result_hv);
6035 string = hv_iterval(result_hv, entry);
6037 /* And perform the modification */
6038 Perl_sv_setpvf(aTHX_ string, "%c%s", prefix, SvPV_nolen(string));
6041 /* Here, 'string' contains the value we want to return */
6042 retval = save_to_buffer(SvPV_nolen(string), retbufp, retbuf_sizep);
6045 is_utf8 = (SvUTF8(string))
6047 : (is_utf8_invariant_string( (U8 *) retval,
6049 ? UTF8NESS_IMMATERIAL
6057 # endif /* Some form of localeconv */
6058 # ifdef HAS_STRFTIME
6060 /* These formats are only available in later strftime's */
6061 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
6063 /* The rest can be gotten from most versions of strftime(). */
6064 case ABDAY_1: case ABDAY_2: case ABDAY_3:
6065 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
6067 case AM_STR: case PM_STR:
6068 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
6069 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
6070 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
6071 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
6072 case DAY_5: case DAY_6: case DAY_7:
6073 case MON_1: case MON_2: case MON_3: case MON_4:
6074 case MON_5: case MON_6: case MON_7: case MON_8:
6075 case MON_9: case MON_10: case MON_11: case MON_12:
6077 const char * format;
6078 bool return_format = FALSE;
6083 GCC_DIAG_IGNORE_STMT(-Wimplicit-fallthrough);
6087 locale_panic_(Perl_form(aTHX_ "switch case: %d problem", item));
6088 NOT_REACHED; /* NOTREACHED */
6090 case PM_STR: hour = 18;
6094 case ABDAY_7: mday++;
6095 case ABDAY_6: mday++;
6096 case ABDAY_5: mday++;
6097 case ABDAY_4: mday++;
6098 case ABDAY_3: mday++;
6099 case ABDAY_2: mday++;
6112 case ABMON_12: mon++;
6113 case ABMON_11: mon++;
6114 case ABMON_10: mon++;
6115 case ABMON_9: mon++;
6116 case ABMON_8: mon++;
6117 case ABMON_7: mon++;
6118 case ABMON_6: mon++;
6119 case ABMON_5: mon++;
6120 case ABMON_4: mon++;
6121 case ABMON_3: mon++;
6122 case ABMON_2: mon++;
6142 return_format = TRUE;
6146 return_format = TRUE;
6150 return_format = TRUE;
6154 return_format = TRUE;
6157 format = "%Ow"; /* Find the alternate digit for 0 */
6161 GCC_DIAG_RESTORE_STMT;
6163 /* The year was deliberately chosen so that January 1 is on the
6164 * first day of the week. Since we're only getting one thing at a
6165 * time, it all works */
6167 ints_to_tm(&mytm, 30, 30, hour, mday, mon, 2011, 0, 0, 0);
6170 temp = strftime8(format,
6172 UTF8NESS_IMMATERIAL, /* All possible formats
6176 false /* not calling from sv_strftime */
6180 temp = strftime_tm(format, &mytm);
6183 retval = save_to_buffer(temp, retbufp, retbuf_sizep);
6186 /* If the item is 'ALT_DIGITS', '*retbuf' contains the alternate
6187 * format for wday 0. If the value is the same as the normal 0,
6188 * there isn't an alternate, so clear the buffer.
6190 * (wday was chosen because its range is all a single digit.
6191 * Things like tm_sec have two digits as the minimum: '00'.) */
6192 if (item == ALT_DIGITS && strEQ(*retbufp, "0")) {
6197 /* ALT_DIGITS is problematic. Experiments on it showed that
6198 * strftime() did not always work properly when going from alt-9 to
6199 * alt-10. Only a few locales have this item defined, and in all
6200 * of them on Linux that khw was able to find, nl_langinfo() merely
6201 * returned the alt-0 character, possibly doubled. Most Unicode
6202 * digits are in blocks of 10 consecutive code points, so that is
6203 * sufficient information for such scripts, as we can infer alt-1,
6204 * alt-2, .... But for a Japanese locale, a CJK ideographic 0 is
6205 * returned, and the CJK digits are not in code point order, so you
6206 * can't really infer anything. The localedef for this locale did
6207 * specify the succeeding digits, so that strftime() works properly
6208 * on them, without needing to infer anything. But the
6209 * nl_langinfo() return did not give sufficient information for the
6210 * caller to understand what's going on. So until there is
6211 * evidence that it should work differently, this returns the alt-0
6212 * string for ALT_DIGITS. */
6214 if (return_format) {
6216 /* If to return the format, not the value, overwrite the buffer
6217 * with it. But some strftime()s will keep the original format
6218 * if illegal, so change those to "" */
6219 if (strEQ(*retbufp, format)) {
6226 /* A format is always in ASCII */
6227 is_utf8 = UTF8NESS_IMMATERIAL;
6234 # ifdef USE_LOCALE_CTYPE
6238 /* The trivial case */
6239 if (isNAME_C_OR_POSIX(locale)) {
6246 /* This function retrieves the code page. It is subject to change, but
6247 * is documented and has been stable for many releases */
6248 UINT ___lc_codepage_func(void);
6250 # ifndef WIN32_USE_FAKE_OLD_MINGW_LOCALES
6252 retval = save_to_buffer(Perl_form(aTHX_ "%d", ___lc_codepage_func()),
6253 retbufp, retbuf_sizep);
6256 retval = save_to_buffer(nl_langinfo(CODESET),
6257 retbufp, retbuf_sizep);
6260 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "locale='%s' cp=%s\n",
6266 /* The codeset is important, but khw did not figure out a way for it to
6267 * be retrieved on non-Windows boxes without nl_langinfo(). But even
6268 * if we can't get it directly, we can usually determine if it is a
6269 * UTF-8 locale or not. If it is UTF-8, we (correctly) use that for
6272 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
6274 /* If libc mbtowc() evaluates the bytes that form the REPLACEMENT
6275 * CHARACTER as that Unicode code point, this has to be a UTF-8 locale.
6278 (void) Perl_mbtowc_(aTHX_ NULL, NULL, 0);/* Reset shift state */
6279 int mbtowc_ret = Perl_mbtowc_(aTHX_ &wc,
6280 STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
6281 if (mbtowc_ret >= 0 && wc == UNICODE_REPLACEMENT) {
6282 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
6283 "mbtowc returned REPLACEMENT\n"));
6288 /* Here, it isn't a UTF-8 locale. */
6290 # else /* mbtowc() is not available. The chances of this code getting
6291 compiled are very small, as it is a C99 required function,
6292 and we are now requiring C99; perhaps if it is a defective
6293 implementation. But if so, there are other libc functions
6294 that could be used instead. */
6296 /* Sling together several possibilities, depending on platform
6297 * capabilities and what we found.
6299 * For non-English locales or non-dollar currency locales, we likely
6300 * will find out whether a locale is UTF-8 or not */
6302 utf8ness_t strings_utf8ness = UTF8NESS_UNKNOWN;
6303 char * scratch_buf = NULL;
6305 # if defined(USE_LOCALE_MONETARY) && defined(HAS_LOCALECONV)
6306 # define LANGINFO_RECURSED_MONETARY 0x1
6308 /* Can't use this method unless localeconv() is available, as that's
6309 * the way we find out the currency symbol.
6311 * First try looking at the currency symbol (via a recursive call) to
6312 * see if it disambiguates things. Often that will be in the native
6313 * script, and if the symbol isn't legal UTF-8, we know that the locale
6316 * The recursion calls my_localeconv() to find CRNCYSTR, and that can
6317 * call is_locale_utf8() which will call my_langinfo(CODESET) which
6318 * will get to here again, ad infinitum. The guard prevents that.
6320 if ((PL_langinfo_recursed & LANGINFO_RECURSED_MONETARY) == 0) {
6321 PL_langinfo_recursed |= LANGINFO_RECURSED_MONETARY;
6322 (void) my_langinfo_c(CRNCYSTR, LC_MONETARY, locale, &scratch_buf,
6323 NULL, &strings_utf8ness);
6324 PL_langinfo_recursed &= ~LANGINFO_RECURSED_MONETARY;
6327 Safefree(scratch_buf);
6330 # ifdef USE_LOCALE_TIME
6331 # define LANGINFO_RECURSED_TIME 0x2
6332 # ifdef LANGINFO_RECURSED_MONETARY
6333 # if LANGINFO_RECURSED_MONETARY == LANGINFO_RECURSED_TIME
6334 # error LANGINFO_RECURSED_MONETARY must differ from LANGINFO_RECURSED_TIME
6338 /* If we have ruled out being UTF-8, no point in checking further. */
6339 if ( strings_utf8ness != UTF8NESS_NO
6340 && (PL_langinfo_recursed & LANGINFO_RECURSED_TIME) == 0)
6342 /* But otherwise do check more. This is done even if the currency
6343 * symbol looks to be UTF-8, just in case that's a false positive.
6345 * Look at the LC_TIME entries, like the names of the months or
6346 * weekdays. We quit at the first one that is illegal UTF-8
6348 * The recursion guard is because the recursed my_langinfo_c() will
6349 * call strftime8() to find the LC_TIME value passed to it, and
6350 * that will call my_langinfo(CODESET) for non-ASCII returns,
6351 * which will get here again, ad infinitum
6354 utf8ness_t this_is_utf8 = UTF8NESS_UNKNOWN;
6355 const int times[] = {
6356 DAY_1, DAY_2, DAY_3, DAY_4, DAY_5, DAY_6, DAY_7,
6357 MON_1, MON_2, MON_3, MON_4, MON_5, MON_6, MON_7, MON_8,
6358 MON_9, MON_10, MON_11, MON_12,
6359 ALT_DIGITS, AM_STR, PM_STR,
6360 ABDAY_1, ABDAY_2, ABDAY_3, ABDAY_4, ABDAY_5, ABDAY_6,
6362 ABMON_1, ABMON_2, ABMON_3, ABMON_4, ABMON_5, ABMON_6,
6363 ABMON_7, ABMON_8, ABMON_9, ABMON_10, ABMON_11, ABMON_12
6366 /* The code in the recursive call can handle switching the locales,
6367 * but by doing it here, we avoid switching each iteration of the
6369 const char * orig_TIME_locale = toggle_locale_c(LC_TIME, locale);
6371 PL_langinfo_recursed |= LANGINFO_RECURSED_TIME;
6372 for (PERL_UINT_FAST8_T i = 0; i < C_ARRAY_LENGTH(times); i++) {
6374 (void) my_langinfo_c(times[i], LC_TIME, locale, &scratch_buf,
6375 NULL, &this_is_utf8);
6376 Safefree(scratch_buf);
6377 if (this_is_utf8 == UTF8NESS_NO) {
6378 strings_utf8ness = UTF8NESS_NO;
6382 if (this_is_utf8 == UTF8NESS_YES) {
6383 strings_utf8ness = UTF8NESS_YES;
6386 PL_langinfo_recursed &= ~LANGINFO_RECURSED_TIME;
6388 /* Here we have gone through all the LC_TIME elements.
6389 * strings_utf8ness has been set as follows:
6390 * UTF8NESS_NO If at least one isn't legal UTF-8
6391 * UTF8NESS_IMMMATERIAL If all are ASCII
6392 * UTF8NESS_YES If all are legal UTF-8 (including
6393 * ASCII), and at least one isn't
6396 restore_toggled_locale_c(LC_TIME, orig_TIME_locale);
6399 # endif /* LC_TIME */
6401 /* If nothing examined above rules out it being UTF-8, and at least one
6402 * thing fits as UTF-8 (and not plain ASCII), assume the codeset is
6404 if (strings_utf8ness == UTF8NESS_YES) {
6409 /* Here, nothing examined indicates that the codeset is UTF-8. But
6410 * what is it? The other locale categories are not likely to be of
6413 * LC_NUMERIC Only a few locales in the world have a non-ASCII radix
6414 * or group separator.
6415 * LC_CTYPE This code wouldn't be compiled if mbtowc() existed and
6416 * was reliable. This is unlikely in C99. There are
6417 * other functions that could be used instead, but are
6418 * they going to exist, and be able to distinguish between
6419 * UTF-8 and 8859-1? Deal with this only if it becomes
6421 * LC_MESSAGES The strings returned from strerror() would seem likely
6422 * candidates, but experience has shown that many systems
6423 * don't actually have translations installed for them.
6424 * They are instead always in English, so everything in
6425 * them is ASCII, which is of no help to us. A Configure
6426 * probe could possibly be written to see if this platform
6427 * has non-ASCII error messages. But again, wait until it
6428 * turns out to be an actual problem. */
6430 # endif /* ! mbtowc() */
6432 /* Rejoin the mbtowc available/not-available cases.
6434 * We got here only because we haven't been able to find the codeset.
6435 * The only other option khw could think of is to see if the codeset is
6436 * part of the locale name. This is very less than ideal; often there
6437 * is no code set in the name; and at other times they even lie.
6439 * But there is an XPG standard syntax, which many locales follow:
6441 * language[_territory[.codeset]][@modifier]
6443 * So we take the part between the dot and any '@' */
6444 retval = (const char *) strchr(locale, '.');
6446 retval = ""; /* Alas, no dot */
6450 /* Don't include the dot */
6453 /* And stop before any '@' */
6454 const char * modifier = strchr(retval, '@');
6456 char * code_set_name;
6457 const Size_t name_len = modifier - retval;
6458 Newx(code_set_name, name_len + 1, char); /* +1 for NUL */
6459 my_strlcpy(code_set_name, retval, name_len + 1);
6460 SAVEFREEPV(code_set_name);
6461 retval = code_set_name;
6464 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
6466 /* When these functions, are available, they were tried earlier and
6467 * indicated that the locale did not act like a proper UTF-8 one. So
6468 * if it claims to be UTF-8, it is a lie */
6469 if (is_codeset_name_UTF8(retval)) {
6476 /* Otherwise the code set name is considered to be everything between
6477 * the dot and the '@' */
6478 retval = save_to_buffer(retval, retbufp, retbuf_sizep);
6482 # endif /* ! WIN32 */
6483 # endif /* USE_LOCALE_CTYPE */
6485 } /* Giant switch() of nl_langinfo() items */
6487 restore_toggled_locale_i(cat_index, orig_switched_locale);
6489 # ifdef USE_LOCALE_CTYPE
6490 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_locale);
6494 *utf8ness = is_utf8;
6499 # endif /* All the implementations of my_langinfo() */
6501 /*--------------------------------------------------------------------------*/
6503 } /* my_langinfo() */
6505 #endif /* USE_LOCALE */
6508 =for apidoc_section $time
6509 =for apidoc sv_strftime_tm
6510 =for apidoc_item sv_strftime_ints
6511 =for apidoc_item my_strftime
6513 These implement the libc strftime(), but with a different API so that the return
6514 value is a pointer to the formatted result (which MUST be arranged to be FREED
6515 BY THE CALLER). This allows these functions to increase the buffer size as
6516 needed, so that the caller doesn't have to worry about that.
6518 On failure they return NULL, and set errno to C<EINVAL>.
6520 C<sv_strftime_tm> and C<sv_strftime_ints> are preferred, as they transparently
6521 handle the UTF-8ness of the current locale, the input C<fmt>, and the returned
6522 result. Only if the current C<LC_TIME> locale is a UTF-8 one (and S<C<use
6523 bytes>> is not in effect) will the result be marked as UTF-8. These differ
6524 only in the form of their inputs. C<sv_strftime_tm> takes a filled-in
6525 S<C<struct tm>> parameter. C<sv_strftime_ints> takes a bunch of integer
6526 parameters that together completely define a given time.
6528 C<my_strftime> is kept for backwards compatibility. Knowing if the result
6529 should be considered UTF-8 or not requires significant extra logic.
6531 Note that C<yday> and C<wday> effectively are ignored by C<sv_strftime_ints>
6532 and C<my_strftime>, as mini_mktime() overwrites them
6534 Also note that all three functions are always executed in the underlying
6535 C<LC_TIME> locale of the program, giving results based on that locale.
6541 S_ints_to_tm(pTHX_ struct tm * mytm,
6542 int sec, int min, int hour, int mday, int mon, int year,
6543 int wday, int yday, int isdst)
6545 /* Create a struct tm structure from the input time-related integer
6548 /* Override with the passed-in values */
6549 Zero(mytm, 1, struct tm);
6552 mytm->tm_hour = hour;
6553 mytm->tm_mday = mday;
6555 mytm->tm_year = year;
6556 mytm->tm_wday = wday;
6557 mytm->tm_yday = yday;
6558 mytm->tm_isdst = isdst;
6561 /* use libc to get the values for tm_gmtoff and tm_zone on platforms that
6562 * have them [perl #18238] */
6563 #if defined(HAS_MKTIME) \
6564 && (defined(HAS_TM_TM_GMTOFF) || defined(HAS_TM_TM_ZONE))
6565 struct tm mytm2 = *mytm;
6569 # ifdef HAS_TM_TM_GMTOFF
6570 mytm->tm_gmtoff = mytm2.tm_gmtoff;
6572 # ifdef HAS_TM_TM_ZONE
6573 mytm->tm_zone = mytm2.tm_zone;
6581 S_strftime_tm(pTHX_ const char *fmt, const struct tm *mytm)
6583 PERL_ARGS_ASSERT_STRFTIME_TM;
6585 /* Execute strftime() based on the input struct tm */
6587 /* An empty format yields an empty result */
6588 const int fmtlen = strlen(fmt);
6591 Newxz (ret, 1, char);
6595 #ifndef HAS_STRFTIME
6596 Perl_croak(aTHX_ "panic: no strftime");
6598 # if defined(USE_LOCALE_CTYPE) && defined(USE_LOCALE_TIME)
6600 const char * orig_CTYPE_LOCALE = toggle_locale_c(LC_CTYPE,
6601 querylocale_c(LC_TIME));
6604 /* Guess an initial size for the returned string based on an expansion
6605 * factor of the input format, but with a minimum that should handle most
6606 * common cases. If this guess is too small, we will try again with a
6608 int bufsize = MAX(fmtlen * 2, 64);
6610 char *buf = NULL; /* Makes Renew() act as Newx() on the first iteration */
6612 Renew(buf, bufsize, char);
6614 /* allowing user-supplied (rather than literal) formats is normally
6615 * frowned upon as a potential security risk; but this is part of the
6616 * API so we have to allow it (and the available formats have a much
6617 * lower chance of doing something bad than the ones for printf etc. */
6618 GCC_DIAG_IGNORE_STMT(-Wformat-nonliteral);
6621 int len = strftime(buf, bufsize, fmt, mytm);
6624 GCC_DIAG_RESTORE_STMT;
6626 /* A non-zero return indicates success. But to make sure we're not
6627 * dealing with some rogue strftime that returns how much space it
6628 * needs instead of 0 when there isn't enough, check that the return
6629 * indicates we have at least one byte of spare space (which will be
6630 * used for the terminating NUL). */
6631 if (inRANGE(len, 1, bufsize - 1)) {
6632 goto strftime_return;
6635 /* There are several possible reasons for a 0 return code for a
6636 * non-empty format, and they are not trivial to tease apart. This
6637 * issue is a known bug in the strftime() API. What we do to cope is
6638 * to assume that the reason is not enough space in the buffer, so
6639 * increase it and try again. */
6642 /* But don't just keep increasing the size indefinitely. Stop when it
6643 * becomes obvious that the reason for failure is something besides not
6644 * enough space. The most likely largest expanding format is %c. On
6645 * khw's Linux box, the maximum result of this is 67 characters, in the
6646 * km_KH locale. If a new script comes along that uses 4 UTF-8 bytes
6647 * per character, and with a similar expansion factor, that would be a
6648 * 268:2 byte ratio, or a bit more than 128:1 = 2**7:1. Some strftime
6649 * implementations allow you to say %1000c to pad to 1000 bytes. This
6650 * shows that it is impossible to implement this without a heuristic
6651 * (which can fail). But it indicates we need to be generous in the
6652 * upper limit before failing. The previous heuristic used was too
6653 * stingy. Since the size doubles per iteration, it doesn't take many
6654 * to reach the limit */
6655 } while (bufsize < ((1 << 11) + 1) * fmtlen);
6657 /* Here, strftime() returned 0, and it likely wasn't for lack of space.
6658 * There are two possible reasons:
6660 * First is that the result is legitimately 0 length. This can happen
6661 * when the format is precisely "%p". That is the only documented format
6662 * that can have an empty result. */
6663 if (strEQ(fmt, "%p")) {
6664 Renew(buf, 1, char);
6666 goto strftime_return;
6669 /* The other reason is that the format string is malformed. Probably it is
6670 * that the string is syntactically invalid for the locale. On some
6671 * platforms an invalid conversion specifier '%?' (for all illegal '?') is
6672 * treated as a literal, but others may fail when '?' is illegal */
6679 # if defined(USE_LOCALE_CTYPE) && defined(USE_LOCALE_TIME)
6681 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_LOCALE);
6692 S_strftime8(pTHX_ const char * fmt,
6693 const struct tm * mytm,
6694 const utf8ness_t fmt_utf8ness,
6695 utf8ness_t * result_utf8ness,
6696 const bool came_from_sv)
6698 PERL_ARGS_ASSERT_STRFTIME8;
6700 /* Wrap strftime_tm, taking into account the input and output UTF-8ness */
6702 #ifdef USE_LOCALE_TIME
6703 # define INDEX_TO_USE LC_TIME_INDEX_
6705 const char * locale = querylocale_c(LC_TIME);
6706 locale_utf8ness_t locale_utf8ness = LOCALE_UTF8NESS_UNKNOWN;
6709 # define INDEX_TO_USE LC_ALL_INDEX_ /* Effectively out of bounds */
6711 const char * locale = "C";
6712 locale_utf8ness_t locale_utf8ness = LOCALE_NOT_UTF8;
6716 switch (fmt_utf8ness) {
6717 case UTF8NESS_IMMATERIAL:
6720 case UTF8NESS_NO: /* Known not to be UTF-8; must not be UTF-8 locale */
6721 if (is_locale_utf8(locale)) {
6726 locale_utf8ness = LOCALE_NOT_UTF8;
6729 case UTF8NESS_YES: /* Known to be UTF-8; must be UTF-8 locale if can't
6731 if (! is_locale_utf8(locale)) {
6732 locale_utf8ness = LOCALE_NOT_UTF8;
6734 bool is_utf8 = true;
6735 Size_t fmt_len = strlen(fmt);
6736 fmt = (char *) bytes_from_utf8((U8 *) fmt, &fmt_len, &is_utf8);
6745 locale_utf8ness = LOCALE_IS_UTF8;
6750 case UTF8NESS_UNKNOWN:
6751 if (! is_locale_utf8(locale)) {
6752 locale_utf8ness = LOCALE_NOT_UTF8;
6755 locale_utf8ness = LOCALE_IS_UTF8;
6758 /* Upgrade 'fmt' to UTF-8 for a UTF-8 locale. Otherwise the
6759 * locale would find any UTF-8 variant characters to be
6761 Size_t fmt_len = strlen(fmt);
6762 fmt = (char *) bytes_to_utf8((U8 *) fmt, &fmt_len);
6770 char * retval = strftime_tm(fmt, mytm);
6771 *result_utf8ness = get_locale_string_utf8ness_i(retval,
6775 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
6776 "fmt=%s, retval=%s; utf8ness=%d",
6778 ((is_utf8_string((U8 *) retval, 0))
6780 :_byte_dump_string((U8 *) retval, strlen(retval),0)),
6791 S_give_perl_locale_control(pTHX_
6793 const char * lc_all_string,
6795 const char ** locales,
6797 const line_t caller_line)
6799 PERL_UNUSED_ARG(caller_line);
6801 /* This is called when the program is in the global locale and are
6802 * switching to per-thread (if available). And it is called at
6803 * initialization time to do the same.
6806 # if defined(WIN32) && defined(USE_THREAD_SAFE_LOCALE)
6808 /* On Windows, convert to per-thread behavior. This isn't necessary in
6809 * POSIX 2008, as the conversion gets done automatically in the
6810 * void_setlocale_i() calls below. */
6811 if (_configthreadlocale(_ENABLE_PER_THREAD_LOCALE) == -1) {
6812 locale_panic_("_configthreadlocale returned an error");
6816 # if ! defined(USE_THREAD_SAFE_LOCALE) \
6817 && ! defined(USE_POSIX_2008_LOCALE)
6818 # if defined(LC_ALL)
6819 PERL_UNUSED_ARG(lc_all_string);
6821 PERL_UNUSED_ARG(locales);
6825 /* This platform has per-thread locale handling. Do the conversion. */
6827 # if defined(LC_ALL)
6829 void_setlocale_c_with_caller(LC_ALL, lc_all_string, __FILE__, caller_line);
6833 for_all_individual_category_indexes(i) {
6834 void_setlocale_i_with_caller(i, locales[i], __FILE__, caller_line);
6840 /* Finally, update our remaining records. 'true' => force recalculation.
6841 * This is needed because we don't know what's happened while Perl hasn't
6842 * had control, so we need to figure out the current state */
6844 # if defined(LC_ALL)
6846 new_LC_ALL(lc_all_string, true);
6850 new_LC_ALL(calculate_LC_ALL_string(locales,
6860 S_output_check_environment_warning(pTHX_ const char * const language,
6861 const char * const lc_all,
6862 const char * const lang)
6864 PerlIO_printf(Perl_error_log,
6865 "perl: warning: Please check that your locale settings:\n");
6869 PerlIO_printf(Perl_error_log, "\tLANGUAGE = %c%s%c,\n",
6870 language ? '"' : '(',
6871 language ? language : "unset",
6872 language ? '"' : ')');
6874 PERL_UNUSED_ARG(language);
6877 PerlIO_printf(Perl_error_log, "\tLC_ALL = %c%s%c,\n",
6879 lc_all ? lc_all : "unset",
6880 lc_all ? '"' : ')');
6882 for_all_individual_category_indexes(i) {
6883 const char * value = PerlEnv_getenv(category_names[i]);
6884 PerlIO_printf(Perl_error_log,
6888 value ? value : "unset",
6892 PerlIO_printf(Perl_error_log, "\tLANG = %c%s%c\n",
6894 lang ? lang : "unset",
6896 PerlIO_printf(Perl_error_log,
6897 " are supported and installed on your system.\n");
6902 /* A helper macro for the next function. Needed because would be called in two
6903 * places. Knows about the internal workings of the function */
6904 #define GET_DESCRIPTION(trial, name) \
6905 ((isNAME_C_OR_POSIX(name)) \
6906 ? "the standard locale" \
6907 : ((trial == (system_default_trial) \
6908 ? "the system default locale" \
6909 : "a fallback locale")))
6912 * Initialize locale awareness.
6915 Perl_init_i18nl10n(pTHX_ int printwarn)
6918 * 0 if not to output warning when setup locale is bad
6919 * 1 if to output warning based on value of PERL_BADLANG
6920 * >1 if to output regardless of PERL_BADLANG
6923 * 1 = set ok or not applicable,
6924 * 0 = fallback to a locale of lower priority
6925 * -1 = fallback to all locales failed, not even to the C locale
6927 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
6928 * set, debugging information is output.
6930 * This routine effectively does the following in most cases:
6932 * basic initialization;
6933 * asserts that the compiled tables are consistent;
6934 * initialize data structures;
6935 * make sure we are in the global locale;
6936 * setlocale(LC_ALL, "");
6937 * switch to per-thread locale if applicable;
6939 * The "" causes the locale to be set to what the environment variables at
6940 * the time say it should be.
6942 * To handle possible failures, the setlocale is expanded to be like:
6944 * trial_locale = pre-first-trial;
6945 * while (has_another_trial()) {
6946 * trial_locale = next_trial();
6947 * if setlocale(LC_ALL, trial_locale) {
6952 * had_failure = true;
6956 * if (had_failure) {
6958 * if (! ok) warn_still_more();
6961 * The first trial is either:
6962 * "" to examine the environment variables for the locale
6963 * NULL to use the values already set for the locale by the program
6964 * embedding this perl instantiation.
6966 * Something is wrong if this trial fails, but there is a sequence of
6967 * fallbacks to try should that happen. They are given in the enum below.
6969 * If there is no LC_ALL defined on the system, the setlocale() above is
6970 * replaced by a loop setting each individual category separately.
6972 * In a non-embeded environment, this code is executed exactly once. It
6973 * sets up the global locale environment. At the end, if some sort of
6974 * thread-safety is in effect, it will turn thread 0 into using that, with
6975 * the same locale as the global initially. thread 0 can then change its
6976 * locale at will without affecting the global one.
6978 * At destruction time, thread 0 will revert to the global locale as the
6979 * other threads die.
6981 * Care must be taken in an embedded environment. This code will be
6982 * executed for each instantiation. Since it changes the global locale, it
6983 * could clash with another running instantiation that isn't using
6984 * per-thread locales. perlembed suggests having the controlling program
6985 * set each instantiation's locale and set PERL_SKIP_LOCALE_INIT so this
6986 * code uses that without actually changing anything. Then the onus is on
6987 * the controlling program to prevent any races. The code below does
6988 * enough locking so as to prevent system calls from overwriting data
6989 * before it is safely copied here, but that isn't a general solution.
6994 PERL_UNUSED_ARG(printwarn);
6997 #else /* USE_LOCALE to near the end of the routine */
7003 const char * const language = PerlEnv_getenv("LANGUAGE");
7006 const char * const language = NULL; /* Unused placeholder */
7009 /* A later getenv() could zap this, so only use here */
7010 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
7012 const bool locwarn = (printwarn > 1
7014 && ( ! bad_lang_use_once
7016 /* disallow with "" or "0" */
7018 && strNE("0", bad_lang_use_once)))));
7021 # define DEBUG_LOCALE_INIT(a,b,c)
7024 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7026 # define DEBUG_LOCALE_INIT(cat_index, locale, result) \
7027 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s\n", \
7028 setlocale_debug_string_i(cat_index, locale, result)));
7031 assert(categories[LC_ALL_INDEX_] == LC_ALL);
7032 assert(strEQ(category_names[LC_ALL_INDEX_], "LC_ALL"));
7033 # ifdef USE_POSIX_2008_LOCALE
7034 assert(category_masks[LC_ALL_INDEX_] == LC_ALL_MASK);
7038 for_all_individual_category_indexes(i) {
7039 assert(category_name_lengths[i] == strlen(category_names[i]));
7042 # endif /* DEBUGGING */
7044 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
7045 * why these particular incantations are used. */
7047 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
7050 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
7053 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
7055 # ifdef USE_PL_CURLOCALES
7057 for (unsigned int i = 0; i <= LC_ALL_INDEX_; i++) {
7058 PL_curlocales[i] = savepv("C");
7062 # ifdef USE_PL_CUR_LC_ALL
7064 PL_cur_LC_ALL = savepv("C");
7067 # if ! defined(PERL_LC_ALL_USES_NAME_VALUE_PAIRS) && defined(LC_ALL)
7071 /* If we haven't done so already, translate the LC_ALL positions of
7072 * categories into our internal indices. */
7073 if (map_LC_ALL_position_to_index[0] == LC_ALL_INDEX_) {
7075 /* Use this array, initialized by a config.h constant */
7076 int lc_all_category_positions[] = PERL_LC_ALL_CATEGORY_POSITIONS_INIT;
7077 STATIC_ASSERT_STMT( C_ARRAY_LENGTH(lc_all_category_positions)
7080 for (unsigned int i = 0;
7081 i < C_ARRAY_LENGTH(lc_all_category_positions);
7084 map_LC_ALL_position_to_index[i] =
7085 get_category_index(lc_all_category_positions[i]);
7092 # ifdef USE_POSIX_2008_LOCALE
7094 /* This is a global, so be sure to keep another instance from zapping it */
7096 if (PL_C_locale_obj) {
7100 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
7101 if (! PL_C_locale_obj) {
7103 locale_panic_(Perl_form(aTHX_
7104 "Cannot create POSIX 2008 C locale object"));
7108 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "created C object %p\n",
7112 /* Switch to using the POSIX 2008 interface now. This would happen below
7113 * anyway, but deferring it can lead to leaks of memory that would also get
7114 * malloc'd in the interim. We arbitrarily switch to the C locale,
7115 * overridden below */
7116 if (! uselocale(PL_C_locale_obj)) {
7117 locale_panic_(Perl_form(aTHX_
7118 "Can't uselocale(%p), LC_ALL supposed to"
7123 # ifdef MULTIPLICITY
7125 PL_cur_locale_obj = PL_C_locale_obj;
7130 /* Now initialize some data structures. This is entirely so that
7131 * later-executed code doesn't have to concern itself with things not being
7132 * initialized. Arbitrarily use the C locale (which we know has to exist
7133 * on the system). */
7135 # ifdef USE_LOCALE_NUMERIC
7137 PL_numeric_radix_sv = newSV(1);
7138 PL_underlying_radix_sv = newSV(1);
7139 Newxz(PL_numeric_name, 1, char); /* Single NUL character */
7142 # ifdef USE_LOCALE_COLLATE
7144 Newxz(PL_collation_name, 1, char);
7147 # ifdef USE_LOCALE_CTYPE
7149 Newxz(PL_ctype_name, 1, char);
7153 new_LC_ALL("C", true /* Don't shortcut */);
7155 /*===========================================================================*/
7157 /* Now ready to override the initialization with the values that the user
7158 * wants. This is done in the global locale as explained in the
7159 * introductory comments to this function */
7160 switch_to_global_locale();
7162 const char * const lc_all = PerlEnv_getenv("LC_ALL");
7163 const char * const lang = PerlEnv_getenv("LANG");
7165 /* We try each locale in the enum, in order, until we get one that works,
7166 * or exhaust the list. Normally the loop is executed just once.
7168 * Each enum value is +1 from the previous */
7171 environment_trial = 0, /* "" or NULL; code below assumes value
7172 0 is the first real trial */
7173 LC_ALL_trial, /* ENV{LC_ALL} */
7174 LANG_trial, /* ENV{LANG} */
7175 system_default_trial, /* Windows .ACP */
7176 C_trial, /* C locale */
7181 SSize_t already_checked = 0;
7182 const char * checked[C_trial];
7185 const char * lc_all_string;
7187 const char * curlocales[LC_ALL_INDEX_];
7190 /* Loop through the initial setting and all the possible fallbacks,
7191 * breaking out of the loop on success */
7192 trial = dummy_trial;
7193 while (trial != beyond_final_trial) {
7195 /* Each time through compute the next trial to use based on the one in
7196 * the previous iteration and switch to the new one. This enforces the
7197 * order in which the fallbacks are applied */
7199 trial = (trials) ((int) trial + 1); /* Casts are needed for g++ */
7201 const char * locale = NULL;
7203 /* Set up the parameters for this trial */
7206 locale_panic_("Unexpectedly got 'dummy_trial");
7209 case environment_trial:
7210 /* This is either "" to get the values from the environment, or
7211 * NULL if the calling program has initialized the values already.
7213 locale = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
7219 if (! lc_all || strEQ(lc_all, "")) {
7220 continue; /* No-op */
7227 if (! lang || strEQ(lang, "")) {
7228 continue; /* No-op */
7234 case system_default_trial:
7236 # if ! defined(WIN32) || ! defined(LC_ALL)
7238 continue; /* No-op */
7241 /* For Windows, we also try the system default locale before "C".
7242 * (If there exists a Windows without LC_ALL we skip this because
7243 * it gets too complicated. For those, "C" is the next fallback
7253 case beyond_final_trial:
7254 continue; /* No-op, causes loop to exit */
7257 /* If the locale is a substantive name, don't try the same locale
7259 if (locale && strNE(locale, "")) {
7260 for (unsigned int i = 0; i < already_checked; i++) {
7261 if (strEQ(checked[i], locale)) {
7266 /* And, for future iterations, indicate we've tried this locale */
7267 checked[already_checked] = savepv(locale);
7268 SAVEFREEPV(checked[already_checked]);
7274 STDIZED_SETLOCALE_LOCK;
7275 lc_all_string = savepv(stdized_setlocale(LC_ALL, locale));
7276 STDIZED_SETLOCALE_UNLOCK;
7278 DEBUG_LOCALE_INIT(LC_ALL_INDEX_, locale, lc_all_string);
7280 if (LIKELY(lc_all_string)) { /* Succeeded */
7285 if (trial == 0 && locwarn) {
7286 PerlIO_printf(Perl_error_log,
7287 "perl: warning: Setting locale failed.\n");
7288 output_check_environment_warning(language, lc_all, lang);
7291 # else /* Below is ! LC_ALL */
7293 bool setlocale_failure = FALSE; /* This trial hasn't failed so far */
7294 bool dowarn = trial == 0 && locwarn;
7296 for_all_individual_category_indexes(j) {
7297 STDIZED_SETLOCALE_LOCK;
7298 curlocales[j] = savepv(stdized_setlocale(categories[j], locale));
7299 STDIZED_SETLOCALE_UNLOCK;
7301 DEBUG_LOCALE_INIT(j, locale, curlocales[j]);
7303 if (UNLIKELY(! curlocales[j])) {
7304 setlocale_failure = TRUE;
7306 /* If are going to warn below, continue to loop so all failures
7307 * are included in the message */
7314 if (LIKELY(! setlocale_failure)) { /* All succeeded */
7316 break; /* Exit trial_locales loop */
7319 /* Here, this trial failed */
7322 PerlIO_printf(Perl_error_log,
7323 "perl: warning: Setting locale failed for the categories:\n");
7325 for_all_individual_category_indexes(j) {
7326 if (! curlocales[j]) {
7327 PerlIO_printf(Perl_error_log, "\t%s\n", category_names[j]);
7331 output_check_environment_warning(language, lc_all, lang);
7332 } /* end of warning on first failure */
7334 # endif /* LC_ALL */
7336 } /* end of looping through the trial locales */
7338 /* If we had to do more than the first trial, it means that one failed, and
7339 * we may need to output a warning, and, if none worked, do more */
7340 if (UNLIKELY(trial != 0)) {
7342 const char * description = "a fallback locale";
7343 const char * name = NULL;;
7345 /* If we didn't find a good fallback, list all we tried */
7346 if (! ok && already_checked > 0) {
7347 PerlIO_printf(Perl_error_log, "perl: warning: Failed to fall"
7349 if (already_checked > 1) { /* more than one was tried */
7350 PerlIO_printf(Perl_error_log, "any of:\n");
7353 while (already_checked > 0) {
7354 name = checked[--already_checked];
7355 description = GET_DESCRIPTION(trial, name);
7356 PerlIO_printf(Perl_error_log, "%s (\"%s\")\n",
7363 /* Here, a fallback worked. So we have saved its name, and the
7364 * trial that succeeded is still valid */
7366 const char * individ_locales[LC_ALL_INDEX_] = { NULL };
7368 /* Even though we know the valid string for LC_ALL that worked,
7369 * translate it into our internal format, which is the
7370 * name=value pairs notation. This is easier for a human to
7371 * decipher than the positional notation. Some platforms
7372 * can return "C C C C C C" for LC_ALL. This code also
7373 * standardizes that result into plain "C". */
7374 switch (parse_LC_ALL_string(lc_all_string,
7375 (const char **) &individ_locales,
7377 false, /* Return only [0] if
7379 false, /* Don't panic on error */
7384 /* Here, the parse failed, which shouldn't happen, but if
7385 * it does, we have an easy fallback that allows us to keep
7387 name = lc_all_string;
7390 case no_array: /* The original is a single locale */
7391 name = lc_all_string;
7394 case only_element_0: /* element[0] is a single locale valid
7395 for all categories */
7396 SAVEFREEPV(individ_locales[0]);
7397 name = individ_locales[0];
7401 name = calculate_LC_ALL_string(individ_locales,
7405 for_all_individual_category_indexes(j) {
7406 Safefree(individ_locales[j]);
7410 name = calculate_LC_ALL_string(curlocales,
7415 description = GET_DESCRIPTION(trial, name);
7419 /* Nothing seems to be working, yet we want to continue
7420 * executing. It may well be that locales are mostly
7421 * irrelevant to this particular program, and there must be
7422 * some locale underlying the program. Figure it out as best
7423 * we can, by querying the system's current locale */
7427 STDIZED_SETLOCALE_LOCK;
7428 name = stdized_setlocale(LC_ALL, NULL);
7429 STDIZED_SETLOCALE_UNLOCK;
7431 if (UNLIKELY(! name)) {
7432 name = "locale name not determinable";
7435 # else /* Below is ! LC_ALL */
7437 const char * system_locales[LC_ALL_INDEX_] = { NULL };
7439 for_all_individual_category_indexes(j) {
7440 STDIZED_SETLOCALE_LOCK;
7441 system_locales[j] = savepv(stdized_setlocale(categories[j],
7443 STDIZED_SETLOCALE_UNLOCK;
7445 if (UNLIKELY(! system_locales[j])) {
7446 system_locales[j] = "not determinable";
7450 /* We use the name=value form for the string, as that is more
7451 * human readable than the positional notation */
7452 name = calculate_LC_ALL_string(system_locales,
7456 description = "what the system says";
7458 for_all_individual_category_indexes(j) {
7459 Safefree(system_locales[j]);
7464 PerlIO_printf(Perl_error_log,
7465 "perl: warning: Falling back to %s (\"%s\").\n",
7468 /* Here, ok being true indicates that the first attempt failed, but
7469 * a fallback succeeded; false => nothing working. Translate to
7470 * API return values. */
7477 give_perl_locale_control(lc_all_string, __LINE__);
7478 Safefree(lc_all_string);
7482 give_perl_locale_control((const char **) &curlocales, __LINE__);
7484 for_all_individual_category_indexes(j) {
7485 Safefree(curlocales[j]);
7489 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
7491 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
7492 * locale is UTF-8. give_perl_locale_control() just above has already
7493 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
7494 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
7495 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
7496 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
7497 PL_utf8locale = PL_in_utf8_CTYPE_locale;
7499 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
7500 This is an alternative to using the -C command line switch
7501 (the -C if present will override this). */
7503 const char *p = PerlEnv_getenv("PERL_UNICODE");
7504 PL_unicode = p ? parse_unicode_opts(&p) : 0;
7505 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
7510 # if defined(USE_POSIX_2008_LOCALE) && defined(MULTIPLICITY)
7511 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
7512 "finished Perl_init_i18nl10n; actual obj=%p,"
7513 " expected obj=%p, initial=%s\n",
7514 uselocale(0), PL_cur_locale_obj,
7515 get_LC_ALL_display()));
7518 /* So won't continue to output stuff */
7519 DEBUG_INITIALIZATION_set(FALSE);
7521 #endif /* USE_LOCALE */
7526 #undef GET_DESCRIPTION
7527 #ifdef USE_LOCALE_COLLATE
7530 S_compute_collxfrm_coefficients(pTHX)
7533 /* A locale collation definition includes primary, secondary, tertiary,
7534 * etc. weights for each character. To sort, the primary weights are used,
7535 * and only if they compare equal, then the secondary weights are used, and
7536 * only if they compare equal, then the tertiary, etc.
7538 * strxfrm() works by taking the input string, say ABC, and creating an
7539 * output transformed string consisting of first the primary weights,
7540 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the tertiary,
7541 * etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters may not have
7542 * weights at every level. In our example, let's say B doesn't have a
7543 * tertiary weight, and A doesn't have a secondary weight. The constructed
7544 * string is then going to be
7545 * A¹B¹C¹ B²C² A³C³ ....
7546 * This has the desired effect that strcmp() will look at the secondary or
7547 * tertiary weights only if the strings compare equal at all higher
7548 * priority weights. The spaces shown here, like in
7550 * are not just for readability. In the general case, these must actually
7551 * be bytes, which we will call here 'separator weights'; and they must be
7552 * smaller than any other weight value, but since these are C strings, only
7553 * the terminating one can be a NUL (some implementations may include a
7554 * non-NUL separator weight just before the NUL). Implementations tend to
7555 * reserve 01 for the separator weights. They are needed so that a shorter
7556 * string's secondary weights won't be misconstrued as primary weights of a
7557 * longer string, etc. By making them smaller than any other weight, the
7558 * shorter string will sort first. (Actually, if all secondary weights are
7559 * smaller than all primary ones, there is no need for a separator weight
7560 * between those two levels, etc.)
7562 * The length of the transformed string is roughly a linear function of the
7563 * input string. It's not exactly linear because some characters don't
7564 * have weights at all levels. When we call strxfrm() we have to allocate
7565 * some memory to hold the transformed string. The calculations below try
7566 * to find coefficients 'm' and 'b' for this locale so that m*x + b equals
7567 * how much space we need, given the size of the input string in 'x'. If
7568 * we calculate too small, we increase the size as needed, and call
7569 * strxfrm() again, but it is better to get it right the first time to
7570 * avoid wasted expensive string transformations.
7572 * We use the string below to find how long the transformation of it is.
7573 * Almost all locales are supersets of ASCII, or at least the ASCII
7574 * letters. We use all of them, half upper half lower, because if we used
7575 * fewer, we might hit just the ones that are outliers in a particular
7576 * locale. Most of the strings being collated will contain a preponderance
7577 * of letters, and even if they are above-ASCII, they are likely to have
7578 * the same number of weight levels as the ASCII ones. It turns out that
7579 * digits tend to have fewer levels, and some punctuation has more, but
7580 * those are relatively sparse in text, and khw believes this gives a
7581 * reasonable result, but it could be changed if experience so dictates. */
7582 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
7583 char * x_longer; /* Transformed 'longer' */
7584 Size_t x_len_longer; /* Length of 'x_longer' */
7586 char * x_shorter; /* We also transform a substring of 'longer' */
7587 Size_t x_len_shorter;
7589 PL_in_utf8_COLLATE_locale = (PL_collation_standard)
7591 : is_locale_utf8(PL_collation_name);
7592 PL_strxfrm_NUL_replacement = '\0';
7593 PL_strxfrm_max_cp = 0;
7595 /* mem_collxfrm_() is used get the transformation (though here we are
7596 * interested only in its length). It is used because it has the
7597 * intelligence to handle all cases, but to work, it needs some values of
7598 * 'm' and 'b' to get it started. For the purposes of this calculation we
7599 * use a very conservative estimate of 'm' and 'b'. This assumes a weight
7600 * can be multiple bytes, enough to hold any UV on the platform, and there
7601 * are 5 levels, 4 weight bytes, and a trailing NUL. */
7602 PL_collxfrm_base = 5;
7603 PL_collxfrm_mult = 5 * sizeof(UV);
7605 /* Find out how long the transformation really is */
7606 x_longer = mem_collxfrm_(longer,
7610 /* We avoid converting to UTF-8 in the called
7611 * function by telling it the string is in UTF-8
7612 * if the locale is a UTF-8 one. Since the string
7613 * passed here is invariant under UTF-8, we can
7614 * claim it's UTF-8 even if it isn't. */
7615 PL_in_utf8_COLLATE_locale);
7618 /* Find out how long the transformation of a substring of 'longer' is.
7619 * Together the lengths of these transformations are sufficient to
7620 * calculate 'm' and 'b'. The substring is all of 'longer' except the
7621 * first character. This minimizes the chances of being swayed by outliers
7623 x_shorter = mem_collxfrm_(longer + 1,
7626 PL_in_utf8_COLLATE_locale);
7627 Safefree(x_shorter);
7629 /* If the results are nonsensical for this simple test, the whole locale
7630 * definition is suspect. Mark it so that locale collation is not active
7631 * at all for it. XXX Should we warn? */
7632 if ( x_len_shorter == 0
7633 || x_len_longer == 0
7634 || x_len_shorter >= x_len_longer)
7636 PL_collxfrm_mult = 0;
7637 PL_collxfrm_base = 1;
7638 DEBUG_L(PerlIO_printf(Perl_debug_log,
7639 "Disabling locale collation for LC_COLLATE='%s';"
7640 " length for shorter sample=%zu; longer=%zu\n",
7641 PL_collation_name, x_len_shorter, x_len_longer));
7644 SSize_t base; /* Temporary */
7646 /* We have both: m * strlen(longer) + b = x_len_longer
7647 * m * strlen(shorter) + b = x_len_shorter;
7648 * subtracting yields:
7649 * m * (strlen(longer) - strlen(shorter))
7650 * = x_len_longer - x_len_shorter
7651 * But we have set things up so that 'shorter' is 1 byte smaller than
7653 * m = x_len_longer - x_len_shorter
7655 * But if something went wrong, make sure the multiplier is at least 1.
7657 if (x_len_longer > x_len_shorter) {
7658 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
7661 PL_collxfrm_mult = 1;
7666 * but in case something has gone wrong, make sure it is non-negative
7668 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
7673 /* Add 1 for the trailing NUL */
7674 PL_collxfrm_base = base + 1;
7677 DEBUG_L(PerlIO_printf(Perl_debug_log,
7678 "?UTF-8 locale=%d; x_len_shorter=%zu, "
7680 " collate multipler=%zu, collate base=%zu\n",
7681 PL_in_utf8_COLLATE_locale,
7682 x_len_shorter, x_len_longer,
7683 PL_collxfrm_mult, PL_collxfrm_base));
7687 Perl_mem_collxfrm_(pTHX_ const char *input_string,
7688 STRLEN len, /* Length of 'input_string' */
7689 STRLEN *xlen, /* Set to length of returned string
7690 (not including the collation index
7692 bool utf8 /* Is the input in UTF-8? */
7695 /* mem_collxfrm_() is like strxfrm() but with two important differences.
7696 * First, it handles embedded NULs. Second, it allocates a bit more memory
7697 * than needed for the transformed data itself. The real transformed data
7698 * begins at offset COLLXFRM_HDR_LEN. *xlen is set to the length of that,
7699 * and doesn't include the collation index size.
7701 * It is the caller's responsibility to eventually free the memory returned
7704 * Please see sv_collxfrm() to see how this is used. */
7706 # define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
7708 char * s = (char *) input_string;
7709 STRLEN s_strlen = strlen(input_string);
7711 STRLEN xAlloc; /* xalloc is a reserved word in VC */
7712 STRLEN length_in_chars;
7713 bool first_time = TRUE; /* Cleared after first loop iteration */
7715 # ifdef USE_LOCALE_CTYPE
7716 const char * orig_CTYPE_locale = NULL;
7719 # if defined(USE_POSIX_2008_LOCALE) && defined HAS_STRXFRM_L
7720 locale_t constructed_locale = (locale_t) 0;
7723 PERL_ARGS_ASSERT_MEM_COLLXFRM_;
7725 /* Must be NUL-terminated */
7726 assert(*(input_string + len) == '\0');
7728 if (PL_collxfrm_mult == 0) { /* unknown or bad */
7729 if (PL_collxfrm_base != 0) { /* bad collation => skip */
7730 DEBUG_L(PerlIO_printf(Perl_debug_log,
7731 "mem_collxfrm_: locale's collation is defective\n"));
7735 /* (mult, base) == (0,0) means we need to calculate mult and base
7736 * before proceeding */
7737 S_compute_collxfrm_coefficients(aTHX);
7740 /* Replace any embedded NULs with the control that sorts before any others.
7741 * This will give as good as possible results on strings that don't
7742 * otherwise contain that character, but otherwise there may be
7743 * less-than-perfect results with that character and NUL. This is
7744 * unavoidable unless we replace strxfrm with our own implementation. */
7745 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
7749 STRLEN sans_nuls_len;
7750 int try_non_controls;
7751 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
7752 making sure 2nd byte is NUL.
7754 STRLEN this_replacement_len;
7756 /* If we don't know what non-NUL control character sorts lowest for
7757 * this locale, find it */
7758 if (PL_strxfrm_NUL_replacement == '\0') {
7760 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
7761 includes the collation index
7764 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
7766 /* Unlikely, but it may be that no control will work to replace
7767 * NUL, in which case we instead look for any character. Controls
7768 * are preferred because collation order is, in general, context
7769 * sensitive, with adjoining characters affecting the order, and
7770 * controls are less likely to have such interactions, allowing the
7771 * NUL-replacement to stand on its own. (Another way to look at it
7772 * is to imagine what would happen if the NUL were replaced by a
7773 * combining character; it wouldn't work out all that well.) */
7774 for (try_non_controls = 0;
7775 try_non_controls < 2;
7779 # ifdef USE_LOCALE_CTYPE
7781 /* In this case we use isCNTRL_LC() below, which relies on
7782 * LC_CTYPE, so that must be switched to correspond with the
7783 * LC_COLLATE locale */
7784 if (! try_non_controls && ! PL_in_utf8_COLLATE_locale) {
7785 orig_CTYPE_locale = toggle_locale_c(LC_CTYPE,
7789 /* Look through all legal code points (NUL isn't) */
7790 for (j = 1; j < 256; j++) {
7791 char * x; /* j's xfrm plus collation index */
7792 STRLEN x_len; /* length of 'x' */
7793 STRLEN trial_len = 1;
7794 char cur_source[] = { '\0', '\0' };
7796 /* Skip non-controls the first time through the loop. The
7797 * controls in a UTF-8 locale are the L1 ones */
7798 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
7805 /* Create a 1-char string of the current code point */
7806 cur_source[0] = (char) j;
7808 /* Then transform it */
7809 x = mem_collxfrm_(cur_source, trial_len, &x_len,
7810 0 /* The string is not in UTF-8 */);
7812 /* Ignore any character that didn't successfully transform.
7818 /* If this character's transformation is lower than
7819 * the current lowest, this one becomes the lowest */
7820 if ( cur_min_x == NULL
7821 || strLT(x + COLLXFRM_HDR_LEN,
7822 cur_min_x + COLLXFRM_HDR_LEN))
7824 PL_strxfrm_NUL_replacement = j;
7825 Safefree(cur_min_x);
7831 } /* end of loop through all 255 characters */
7833 # ifdef USE_LOCALE_CTYPE
7834 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_locale);
7837 /* Stop looking if found */
7842 /* Unlikely, but possible, if there aren't any controls that
7843 * work in the locale, repeat the loop, looking for any
7844 * character that works */
7845 DEBUG_L(PerlIO_printf(Perl_debug_log,
7846 "mem_collxfrm_: No control worked. Trying non-controls\n"));
7847 } /* End of loop to try first the controls, then any char */
7850 DEBUG_L(PerlIO_printf(Perl_debug_log,
7851 "mem_collxfrm_: Couldn't find any character to replace"
7852 " embedded NULs in locale %s with", PL_collation_name));
7856 DEBUG_L(PerlIO_printf(Perl_debug_log,
7857 "mem_collxfrm_: Replacing embedded NULs in locale %s with "
7858 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
7860 Safefree(cur_min_x);
7861 } /* End of determining the character that is to replace NULs */
7863 /* If the replacement is variant under UTF-8, it must match the
7864 * UTF8-ness of the original */
7865 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
7866 this_replacement_char[0] =
7867 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
7868 this_replacement_char[1] =
7869 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
7870 this_replacement_len = 2;
7873 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
7874 /* this_replacement_char[1] = '\0' was done at initialization */
7875 this_replacement_len = 1;
7878 /* The worst case length for the replaced string would be if every
7879 * character in it is NUL. Multiply that by the length of each
7880 * replacement, and allow for a trailing NUL */
7881 sans_nuls_len = (len * this_replacement_len) + 1;
7882 Newx(sans_nuls, sans_nuls_len, char);
7885 /* Replace each NUL with the lowest collating control. Loop until have
7886 * exhausted all the NULs */
7887 while (s + s_strlen < e) {
7888 my_strlcat(sans_nuls, s, sans_nuls_len);
7890 /* Do the actual replacement */
7891 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
7893 /* Move past the input NUL */
7895 s_strlen = strlen(s);
7898 /* And add anything that trails the final NUL */
7899 my_strlcat(sans_nuls, s, sans_nuls_len);
7901 /* Switch so below we transform this modified string */
7904 } /* End of replacing NULs */
7906 /* Make sure the UTF8ness of the string and locale match */
7907 if (utf8 != PL_in_utf8_COLLATE_locale) {
7908 /* XXX convert above Unicode to 10FFFF? */
7909 const char * const t = s; /* Temporary so we can later find where the
7912 /* Here they don't match. Change the string's to be what the locale is
7915 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
7916 s = (char *) bytes_to_utf8((const U8 *) s, &len);
7919 else { /* locale is not UTF-8; but input is; downgrade the input */
7921 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
7923 /* If the downgrade was successful we are done, but if the input
7924 * contains things that require UTF-8 to represent, have to do
7925 * damage control ... */
7926 if (UNLIKELY(utf8)) {
7928 /* What we do is construct a non-UTF-8 string with
7929 * 1) the characters representable by a single byte converted
7930 * to be so (if necessary);
7931 * 2) and the rest converted to collate the same as the
7932 * highest collating representable character. That makes
7933 * them collate at the end. This is similar to how we
7934 * handle embedded NULs, but we use the highest collating
7935 * code point instead of the smallest. Like the NUL case,
7936 * this isn't perfect, but is the best we can reasonably
7937 * do. Every above-255 code point will sort the same as
7938 * the highest-sorting 0-255 code point. If that code
7939 * point can combine in a sequence with some other code
7940 * points for weight calculations, us changing something to
7941 * be it can adversely affect the results. But in most
7942 * cases, it should work reasonably. And note that this is
7943 * really an illegal situation: using code points above 255
7944 * on a locale where only 0-255 are valid. If two strings
7945 * sort entirely equal, then the sort order for the
7946 * above-255 code points will be in code point order. */
7950 /* If we haven't calculated the code point with the maximum
7951 * collating order for this locale, do so now */
7952 if (! PL_strxfrm_max_cp) {
7955 /* The current transformed string that collates the
7956 * highest (except it also includes the prefixed collation
7958 char * cur_max_x = NULL;
7960 /* Look through all legal code points (NUL isn't) */
7961 for (j = 1; j < 256; j++) {
7964 char cur_source[] = { '\0', '\0' };
7966 /* Create a 1-char string of the current code point */
7967 cur_source[0] = (char) j;
7969 /* Then transform it */
7970 x = mem_collxfrm_(cur_source, 1, &x_len, FALSE);
7972 /* If something went wrong (which it shouldn't), just
7973 * ignore this code point */
7978 /* If this character's transformation is higher than
7979 * the current highest, this one becomes the highest */
7980 if ( cur_max_x == NULL
7981 || strGT(x + COLLXFRM_HDR_LEN,
7982 cur_max_x + COLLXFRM_HDR_LEN))
7984 PL_strxfrm_max_cp = j;
7985 Safefree(cur_max_x);
7994 DEBUG_L(PerlIO_printf(Perl_debug_log,
7995 "mem_collxfrm_: Couldn't find any character to"
7996 " replace above-Latin1 chars in locale %s with",
7997 PL_collation_name));
8001 DEBUG_L(PerlIO_printf(Perl_debug_log,
8002 "mem_collxfrm_: highest 1-byte collating character"
8003 " in locale %s is 0x%02X\n",
8005 PL_strxfrm_max_cp));
8007 Safefree(cur_max_x);
8010 /* Here we know which legal code point collates the highest.
8011 * We are ready to construct the non-UTF-8 string. The length
8012 * will be at least 1 byte smaller than the input string
8013 * (because we changed at least one 2-byte character into a
8014 * single byte), but that is eaten up by the trailing NUL */
8020 char * e = (char *) t + len;
8022 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
8024 if (UTF8_IS_INVARIANT(cur_char)) {
8027 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
8028 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
8030 else { /* Replace illegal cp with highest collating
8032 s[d++] = PL_strxfrm_max_cp;
8036 Renew(s, d, char); /* Free up unused space */
8041 /* Here, we have constructed a modified version of the input. It could
8042 * be that we already had a modified copy before we did this version.
8043 * If so, that copy is no longer needed */
8044 if (t != input_string) {
8049 length_in_chars = (utf8)
8050 ? utf8_length((U8 *) s, (U8 *) s + len)
8053 /* The first element in the output is the collation id, used by
8054 * sv_collxfrm(); then comes the space for the transformed string. The
8055 * equation should give us a good estimate as to how much is needed */
8056 xAlloc = COLLXFRM_HDR_LEN
8058 + (PL_collxfrm_mult * length_in_chars);
8059 Newx(xbuf, xAlloc, char);
8060 if (UNLIKELY(! xbuf)) {
8061 DEBUG_L(PerlIO_printf(Perl_debug_log,
8062 "mem_collxfrm_: Couldn't malloc %zu bytes\n", xAlloc));
8066 /* Store the collation id */
8067 *(PERL_UINTMAX_T *)xbuf = PL_collation_ix;
8069 # if defined(USE_POSIX_2008_LOCALE) && defined HAS_STRXFRM_L
8070 # ifdef USE_LOCALE_CTYPE
8072 constructed_locale = newlocale(LC_CTYPE_MASK, PL_collation_name,
8073 duplocale(use_curlocale_scratch()));
8076 constructed_locale = duplocale(use_curlocale_scratch());
8079 # define my_strxfrm(dest, src, n) strxfrm_l(dest, src, n, \
8081 # define CLEANUP_STRXFRM \
8083 if (constructed_locale != (locale_t) 0) \
8084 freelocale(constructed_locale); \
8087 # define my_strxfrm(dest, src, n) strxfrm(dest, src, n)
8088 # ifdef USE_LOCALE_CTYPE
8090 orig_CTYPE_locale = toggle_locale_c(LC_CTYPE, PL_collation_name);
8092 # define CLEANUP_STRXFRM \
8093 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_locale)
8095 # define CLEANUP_STRXFRM NOOP
8099 /* Then the transformation of the input. We loop until successful, or we
8104 *xlen = my_strxfrm(xbuf + COLLXFRM_HDR_LEN,
8106 xAlloc - COLLXFRM_HDR_LEN);
8109 /* If the transformed string occupies less space than we told strxfrm()
8110 * was available, it means it transformed the whole string. */
8111 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
8113 /* But there still could have been a problem */
8115 DEBUG_L(PerlIO_printf(Perl_debug_log,
8116 "strxfrm failed for LC_COLLATE=%s; errno=%d, input=%s\n",
8117 PL_collation_name, errno,
8118 _byte_dump_string((U8 *) s, len, 0)));
8122 /* Here, the transformation was successful. Some systems include a
8123 * trailing NUL in the returned length. Ignore it, using a loop in
8124 * case multiple trailing NULs are returned. */
8126 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
8131 /* If the first try didn't get it, it means our prediction was low.
8132 * Modify the coefficients so that we predict a larger value in any
8133 * future transformations */
8135 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
8136 STRLEN computed_guess = PL_collxfrm_base
8137 + (PL_collxfrm_mult * length_in_chars);
8139 /* On zero-length input, just keep current slope instead of
8141 const STRLEN new_m = (length_in_chars != 0)
8142 ? needed / length_in_chars
8145 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8146 "initial size of %zu bytes for a length "
8147 "%zu string was insufficient, %zu needed\n",
8148 computed_guess, length_in_chars, needed));
8150 /* If slope increased, use it, but discard this result for
8151 * length 1 strings, as we can't be sure that it's a real slope
8153 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
8157 STRLEN old_m = PL_collxfrm_mult;
8158 STRLEN old_b = PL_collxfrm_base;
8162 PL_collxfrm_mult = new_m;
8163 PL_collxfrm_base = 1; /* +1 For trailing NUL */
8164 computed_guess = PL_collxfrm_base
8165 + (PL_collxfrm_mult * length_in_chars);
8166 if (computed_guess < needed) {
8167 PL_collxfrm_base += needed - computed_guess;
8170 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8171 "slope is now %zu; was %zu, base "
8172 "is now %zu; was %zu\n",
8173 PL_collxfrm_mult, old_m,
8174 PL_collxfrm_base, old_b));
8176 else { /* Slope didn't change, but 'b' did */
8177 const STRLEN new_b = needed
8180 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8181 "base is now %zu; was %zu\n", new_b, PL_collxfrm_base));
8182 PL_collxfrm_base = new_b;
8189 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
8190 DEBUG_L(PerlIO_printf(Perl_debug_log,
8191 "mem_collxfrm_: Needed %zu bytes, max permissible is %u\n",
8192 *xlen, PERL_INT_MAX));
8196 /* A well-behaved strxfrm() returns exactly how much space it needs
8197 * (usually not including the trailing NUL) when it fails due to not
8198 * enough space being provided. Assume that this is the case unless
8199 * it's been proven otherwise */
8200 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
8201 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
8203 else { /* Here, either:
8204 * 1) The strxfrm() has previously shown bad behavior; or
8205 * 2) It isn't the first time through the loop, which means
8206 * that the strxfrm() is now showing bad behavior, because
8207 * we gave it what it said was needed in the previous
8208 * iteration, and it came back saying it needed still more.
8209 * (Many versions of cygwin fit this. When the buffer size
8210 * isn't sufficient, they return the input size instead of
8211 * how much is needed.)
8212 * Increase the buffer size by a fixed percentage and try again.
8214 xAlloc += (xAlloc / 4) + 1;
8215 PL_strxfrm_is_behaved = FALSE;
8217 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8218 "mem_collxfrm_ required more space than previously"
8219 " calculated for locale %s, trying again with new"
8221 PL_collation_name, COLLXFRM_HDR_LEN,
8222 xAlloc - COLLXFRM_HDR_LEN));
8225 Renew(xbuf, xAlloc, char);
8226 if (UNLIKELY(! xbuf)) {
8227 DEBUG_L(PerlIO_printf(Perl_debug_log,
8228 "mem_collxfrm_: Couldn't realloc %zu bytes\n", xAlloc));
8237 DEBUG_L(print_collxfrm_input_and_return(s, s + len, xbuf, *xlen, utf8));
8239 /* Free up unneeded space; retain enough for trailing NUL */
8240 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
8242 if (s != input_string) {
8251 DEBUG_L(print_collxfrm_input_and_return(s, s + len, NULL, 0, utf8));
8254 if (s != input_string) {
8265 S_print_collxfrm_input_and_return(pTHX_
8273 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
8275 PerlIO_printf(Perl_debug_log,
8276 "mem_collxfrm_[ix %" UVuf "] for locale '%s':\n"
8277 " input=%s\n return=%s\n return len=%zu\n",
8278 (UV) PL_collation_ix, PL_collation_name,
8279 get_displayable_string(s, e, is_utf8),
8284 : _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
8289 # endif /* DEBUGGING */
8292 Perl_strxfrm(pTHX_ SV * src)
8294 PERL_ARGS_ASSERT_STRXFRM;
8296 /* For use by POSIX::strxfrm(). If they differ, toggle LC_CTYPE to
8297 * LC_COLLATE to avoid potential mojibake.
8299 * If we can't calculate a collation, 'src' is instead returned, so that
8300 * future comparisons will be by code point order */
8302 # ifdef USE_LOCALE_CTYPE
8304 const char * orig_ctype = toggle_locale_c(LC_CTYPE,
8305 querylocale_c(LC_COLLATE));
8311 const char *p = SvPV_const(src, srclen);
8312 const U32 utf8_flag = SvUTF8(src);
8313 char *d = mem_collxfrm_(p, srclen, &dstlen, cBOOL(utf8_flag));
8315 assert(utf8_flag == 0 || utf8_flag == SVf_UTF8);
8319 dst =newSVpvn_flags(d + COLLXFRM_HDR_LEN,
8320 dstlen, SVs_TEMP|utf8_flag);
8324 # ifdef USE_LOCALE_CTYPE
8326 restore_toggled_locale_c(LC_CTYPE, orig_ctype);
8333 #endif /* USE_LOCALE_COLLATE */
8337 S_toggle_locale_i(pTHX_ const locale_category_index cat_index,
8338 const char * new_locale,
8339 const line_t caller_line)
8341 PERL_ARGS_ASSERT_TOGGLE_LOCALE_I;
8342 assert(cat_index <= LC_ALL_INDEX_);
8344 /* Changes the locale for the category specified by 'index' to 'new_locale,
8345 * if they aren't already the same.
8347 * Returns a copy of the name of the original locale for 'cat_index'
8348 * so can be switched back to with the companion function
8349 * restore_toggled_locale_i(), (NULL if no restoral is necessary.) */
8351 /* Find the original locale of the category we may need to change, so that
8352 * it can be restored to later */
8353 const char * locale_to_restore_to = querylocale_i(cat_index);
8355 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8356 "(%" LINE_Tf "): toggle_locale_i: index=%d(%s), wanted=%s,"
8358 caller_line, cat_index, category_names[cat_index],
8359 new_locale, locale_to_restore_to));
8361 if (! locale_to_restore_to) {
8362 locale_panic_via_(Perl_form(aTHX_
8363 "Could not find current %s locale",
8364 category_names[cat_index]),
8365 __FILE__, caller_line);
8368 /* If the locales are the same, there's nothing to do */
8369 if (strEQ(locale_to_restore_to, new_locale)) {
8370 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8371 "(%" LINE_Tf "): %s locale unchanged as %s\n",
8372 caller_line, category_names[cat_index],
8378 /* Finally, change the locale to the new one */
8379 void_setlocale_i_with_caller(cat_index, new_locale, __FILE__, caller_line);
8381 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8382 "(%" LINE_Tf "): %s locale switched to %s\n",
8383 caller_line, category_names[cat_index], new_locale));
8385 return locale_to_restore_to;
8388 PERL_UNUSED_ARG(caller_line);
8394 S_restore_toggled_locale_i(pTHX_ const locale_category_index cat_index,
8395 const char * restore_locale,
8396 const line_t caller_line)
8398 /* Restores the locale for LC_category corresponding to cat_index to
8399 * 'restore_locale' (which is a copy that will be freed by this function),
8400 * or do nothing if the latter parameter is NULL */
8402 PERL_ARGS_ASSERT_RESTORE_TOGGLED_LOCALE_I;
8403 assert(cat_index <= LC_ALL_INDEX_);
8405 if (restore_locale == NULL) {
8406 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8407 "(%" LINE_Tf "): No need to restore %s\n",
8408 caller_line, category_names[cat_index]));
8412 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
8413 "(%" LINE_Tf "): %s restoring locale to %s\n",
8414 caller_line, category_names[cat_index],
8417 void_setlocale_i_with_caller(cat_index, restore_locale,
8418 __FILE__, caller_line);
8421 PERL_UNUSED_ARG(caller_line);
8426 # ifdef USE_LOCALE_CTYPE
8429 S_is_codeset_name_UTF8(const char * name)
8431 /* Return a boolean as to if the passed-in name indicates it is a UTF-8
8432 * code set. Several variants are possible */
8433 const Size_t len = strlen(name);
8435 PERL_ARGS_ASSERT_IS_CODESET_NAME_UTF8;
8439 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
8440 if (memENDs(name, len, "65001")) {
8445 /* 'UTF8' or 'UTF-8' */
8446 return ( inRANGE(len, 4, 5)
8447 && name[len-1] == '8'
8448 && ( memBEGINs(name, len, "UTF")
8449 || memBEGINs(name, len, "utf"))
8450 && (len == 4 || name[3] == '-'));
8454 #endif /* USE_LOCALE */
8457 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
8459 /* Internal function which returns if we are in the scope of a pragma that
8460 * enables the locale category 'category'. 'compiling' should indicate if
8461 * this is during the compilation phase (TRUE) or not (FALSE). */
8463 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
8465 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0);
8466 if (! these_categories || these_categories == &PL_sv_placeholder) {
8470 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
8471 * a valid unsigned */
8472 assert(category >= -1);
8473 return cBOOL(SvUV(these_categories) & (1U << (category + 1)));
8476 /* my_strerror() returns a mortalized copy of the text of the error message
8477 * associated with 'errnum'.
8479 * If not called from within the scope of 'use locale', it uses the text from
8480 * the C locale. If Perl is compiled to not pay attention to LC_CTYPE nor
8481 * LC_MESSAGES, it uses whatever strerror() returns. Otherwise the text is
8482 * derived from the locale, LC_MESSAGES if we have that; LC_CTYPE if not.
8484 * It returns in *utf8ness the result's UTF-8ness
8486 * The function just calls strerror(), but temporarily switches locales, if
8487 * needed. Many platforms require LC_CTYPE and LC_MESSAGES to be in the same
8488 * CODESET in order for the return from strerror() to not contain '?' symbols,
8489 * or worse, mojibaked. It's cheaper to just use the stricter criteria of
8490 * being in the same locale. So the code below uses a common locale for both
8491 * categories. Again, that is C if not within 'use locale' scope; or the
8492 * LC_MESSAGES locale if in scope and we have that category; and LC_CTYPE if we
8493 * don't have LC_MESSAGES; and whatever strerror returns if we don't have
8496 * There are two sets of implementations. The first below is if we have
8497 * strerror_l(). This is the simpler. We just use the already-built C locale
8498 * object if not in locale scope, or build up a custom one otherwise.
8500 * When strerror_l() is not available, we may have to swap locales temporarily
8501 * to bring the two categories into sync with each other, and possibly to the C
8504 * Because the prepropessing directives to conditionally compile this function
8505 * would greatly obscure the logic of the various implementations, the whole
8506 * function is repeated for each configuration, with some common macros. */
8508 /* Used to shorten the definitions of the following implementations of
8510 #define DEBUG_STRERROR_ENTER(errnum, in_locale) \
8511 DEBUG_Lv(PerlIO_printf(Perl_debug_log, \
8512 "my_strerror called with errnum %d;" \
8513 " Within locale scope=%d\n", \
8516 #define DEBUG_STRERROR_RETURN(errstr, utf8ness) \
8517 DEBUG_Lv(PerlIO_printf(Perl_debug_log, \
8518 "Strerror returned; saving a copy: '%s';" \
8520 get_displayable_string(errstr, \
8521 errstr + strlen(errstr), \
8525 /* On platforms that have precisely one of these categories (Windows
8526 * qualifies), these yield the correct one */
8527 #if defined(USE_LOCALE_CTYPE)
8528 # define WHICH_LC_INDEX LC_CTYPE_INDEX_
8529 #elif defined(USE_LOCALE_MESSAGES)
8530 # define WHICH_LC_INDEX LC_MESSAGES_INDEX_
8533 /*===========================================================================*/
8534 /* First set of implementations, when have strerror_l() */
8536 #if defined(USE_POSIX_2008_LOCALE) && defined(HAS_STRERROR_L)
8538 # if ! defined(USE_LOCALE_CTYPE) && ! defined(USE_LOCALE_MESSAGES)
8540 /* Here, neither category is defined: use the C locale */
8542 Perl_my_strerror(pTHX_ const int errnum, utf8ness_t * utf8ness)
8544 PERL_ARGS_ASSERT_MY_STRERROR;
8546 DEBUG_STRERROR_ENTER(errnum, 0);
8548 const char *errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
8549 *utf8ness = UTF8NESS_IMMATERIAL;
8551 DEBUG_STRERROR_RETURN(errstr, utf8ness);
8557 # elif ! defined(USE_LOCALE_CTYPE) || ! defined(USE_LOCALE_MESSAGES)
8559 /*--------------------------------------------------------------------------*/
8561 /* Here one or the other of CTYPE or MESSAGES is defined, but not both. If we
8562 * are not within 'use locale' scope of the only one defined, we use the C
8563 * locale; otherwise use the current locale object */
8566 Perl_my_strerror(pTHX_ const int errnum, utf8ness_t * utf8ness)
8568 PERL_ARGS_ASSERT_MY_STRERROR;
8570 DEBUG_STRERROR_ENTER(errnum, IN_LC(categories[WHICH_LC_INDEX]));
8572 /* Use C if not within locale scope; Otherwise, use current locale */
8573 const locale_t which_obj = (IN_LC(categories[WHICH_LC_INDEX]))
8575 : use_curlocale_scratch();
8577 const char *errstr = savepv(strerror_l(errnum, which_obj));
8578 *utf8ness = get_locale_string_utf8ness_i(errstr, LOCALE_UTF8NESS_UNKNOWN,
8579 NULL, WHICH_LC_INDEX);
8580 DEBUG_STRERROR_RETURN(errstr, utf8ness);
8586 /*--------------------------------------------------------------------------*/
8587 # else /* Are using both categories. Place them in the same CODESET,
8588 * either C or the LC_MESSAGES locale */
8591 Perl_my_strerror(pTHX_ const int errnum, utf8ness_t * utf8ness)
8593 PERL_ARGS_ASSERT_MY_STRERROR;
8595 DEBUG_STRERROR_ENTER(errnum, IN_LC(LC_MESSAGES));
8598 if (! IN_LC(LC_MESSAGES)) { /* Use C if not within locale scope */
8599 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
8600 *utf8ness = UTF8NESS_IMMATERIAL;
8602 else { /* Otherwise, use the LC_MESSAGES locale, making sure LC_CTYPE
8604 locale_t cur = duplocale(use_curlocale_scratch());
8606 cur = newlocale(LC_CTYPE_MASK, querylocale_c(LC_MESSAGES), cur);
8607 errstr = savepv(strerror_l(errnum, cur));
8608 *utf8ness = get_locale_string_utf8ness_i(errstr,
8609 LOCALE_UTF8NESS_UNKNOWN,
8610 NULL, LC_MESSAGES_INDEX_);
8614 DEBUG_STRERROR_RETURN(errstr, utf8ness);
8619 # endif /* Above is using strerror_l */
8620 /*===========================================================================*/
8621 #else /* Below is not using strerror_l */
8622 # if ! defined(USE_LOCALE_CTYPE) && ! defined(USE_LOCALE_MESSAGES)
8624 /* If not using using either of the categories, return plain, unadorned
8628 Perl_my_strerror(pTHX_ const int errnum, utf8ness_t * utf8ness)
8630 PERL_ARGS_ASSERT_MY_STRERROR;
8632 DEBUG_STRERROR_ENTER(errnum, 0);
8634 const char *errstr = savepv(Strerror(errnum));
8635 *utf8ness = UTF8NESS_IMMATERIAL;
8637 DEBUG_STRERROR_RETURN(errstr, utf8ness);
8643 /*--------------------------------------------------------------------------*/
8644 # elif ! defined(USE_LOCALE_CTYPE) || ! defined(USE_LOCALE_MESSAGES)
8646 /* Here one or the other of CTYPE or MESSAGES is defined, but not both. If we
8647 * are not within 'use locale' scope of the only one defined, we use the C
8648 * locale; otherwise use the current locale */
8651 Perl_my_strerror(pTHX_ const int errnum, utf8ness_t * utf8ness)
8653 PERL_ARGS_ASSERT_MY_STRERROR;
8655 DEBUG_STRERROR_ENTER(errnum, IN_LC(categories[WHICH_LC_INDEX]));
8658 if (IN_LC(categories[WHICH_LC_INDEX])) {
8659 errstr = savepv(Strerror(errnum));
8660 *utf8ness = get_locale_string_utf8ness_i(errstr,
8661 LOCALE_UTF8NESS_UNKNOWN,
8662 NULL, WHICH_LC_INDEX);
8668 const char * orig_locale = toggle_locale_i(WHICH_LC_INDEX, "C");
8670 errstr = savepv(Strerror(errnum));
8672 restore_toggled_locale_i(WHICH_LC_INDEX, orig_locale);
8676 *utf8ness = UTF8NESS_IMMATERIAL;
8679 DEBUG_STRERROR_RETURN(errstr, utf8ness);
8685 /*--------------------------------------------------------------------------*/
8688 /* Below, have both LC_CTYPE and LC_MESSAGES. Place them in the same CODESET,
8689 * either C or the LC_MESSAGES locale */
8692 Perl_my_strerror(pTHX_ const int errnum, utf8ness_t * utf8ness)
8694 PERL_ARGS_ASSERT_MY_STRERROR;
8696 DEBUG_STRERROR_ENTER(errnum, IN_LC(LC_MESSAGES));
8698 const char * desired_locale = (IN_LC(LC_MESSAGES))
8699 ? querylocale_c(LC_MESSAGES)
8701 /* XXX Can fail on z/OS */
8705 const char* orig_CTYPE_locale = toggle_locale_c(LC_CTYPE,
8707 const char* orig_MESSAGES_locale = toggle_locale_c(LC_MESSAGES,
8709 const char *errstr = savepv(Strerror(errnum));
8711 restore_toggled_locale_c(LC_MESSAGES, orig_MESSAGES_locale);
8712 restore_toggled_locale_c(LC_CTYPE, orig_CTYPE_locale);
8716 *utf8ness = get_locale_string_utf8ness_i(errstr, LOCALE_UTF8NESS_UNKNOWN,
8717 NULL, LC_MESSAGES_INDEX_);
8718 DEBUG_STRERROR_RETURN(errstr, utf8ness);
8724 /*--------------------------------------------------------------------------*/
8725 # endif /* end of not using strerror_l() */
8726 #endif /* end of all the my_strerror() implementations */
8730 =for apidoc switch_to_global_locale
8732 This function copies the locale state of the calling thread into the program's
8733 global locale, and converts the thread to use that global locale.
8735 It is intended so that Perl can safely be used with C libraries that access the
8736 global locale and which can't be converted to not access it. Effectively, this
8737 means libraries that call C<L<setlocale(3)>> on non-Windows systems. (For
8738 portability, it is a good idea to use it on Windows as well.)
8740 A downside of using it is that it disables the services that Perl provides to
8741 hide locale gotchas from your code. The service you most likely will miss
8742 regards the radix character (decimal point) in floating point numbers. Code
8743 executed after this function is called can no longer just assume that this
8744 character is correct for the current circumstances.
8746 To return to Perl control, and restart the gotcha prevention services, call
8747 C<L</sync_locale>>. Behavior is undefined for any pure Perl code that executes
8748 while the switch is in effect.
8750 The global locale and the per-thread locales are independent. As long as just
8751 one thread converts to the global locale, everything works smoothly. But if
8752 more than one does, they can easily interfere with each other, and races are
8753 likely. On Windows systems prior to Visual Studio 15 (at which point Microsoft
8754 fixed a bug), races can occur (even if only one thread has been converted to
8755 the global locale), but only if you use the following operations:
8759 =item L<POSIX::localeconv|POSIX/localeconv>
8761 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
8763 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
8767 The first item is not fixable (except by upgrading to a later Visual Studio
8768 release), but it would be possible to work around the latter two items by
8769 having Perl change its algorithm for calculating these to use Windows API
8770 functions (likely C<GetNumberFormat> and C<GetCurrencyFormat>); patches
8773 XS code should never call plain C<setlocale>, but should instead be converted
8774 to either call L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in
8775 for the system C<setlocale>) or use the methods given in L<perlcall> to call
8776 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
8777 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
8782 #if defined(WIN32) && defined(USE_THREAD_SAFE_LOCALE)
8783 # define CHANGE_SYSTEM_LOCALE_TO_GLOBAL \
8785 if (_configthreadlocale(_DISABLE_PER_THREAD_LOCALE) == -1) { \
8786 locale_panic_("_configthreadlocale returned an error"); \
8789 #elif defined(USE_POSIX_2008_LOCALE)
8790 # define CHANGE_SYSTEM_LOCALE_TO_GLOBAL \
8792 locale_t old_locale = uselocale(LC_GLOBAL_LOCALE); \
8793 if (! old_locale) { \
8794 locale_panic_("Could not change to global locale"); \
8797 /* Free the per-thread memory */ \
8798 if ( old_locale != LC_GLOBAL_LOCALE \
8799 && old_locale != PL_C_locale_obj) \
8801 freelocale(old_locale); \
8805 # define CHANGE_SYSTEM_LOCALE_TO_GLOBAL
8809 Perl_switch_to_global_locale(pTHX)
8814 DEBUG_L(PerlIO_printf(Perl_debug_log, "Entering switch_to_global; %s\n",
8815 get_LC_ALL_display()));
8817 /* In these cases, we use the system state to determine if we are in the
8818 * global locale or not. */
8819 # ifdef USE_POSIX_2008_LOCALE
8821 const bool perl_controls = (LC_GLOBAL_LOCALE != uselocale((locale_t) 0));
8823 # elif defined(USE_THREAD_SAFE_LOCALE) && defined(WIN32)
8825 int config_return = _configthreadlocale(0);
8826 if (config_return == -1) {
8827 locale_panic_("_configthreadlocale returned an error");
8829 const bool perl_controls = (config_return == _ENABLE_PER_THREAD_LOCALE);
8833 const bool perl_controls = false;
8837 /* No-op if already in global */
8838 if (! perl_controls) {
8844 const char * thread_locale = calculate_LC_ALL_string(NULL,
8845 EXTERNAL_FORMAT_FOR_SET,
8848 CHANGE_SYSTEM_LOCALE_TO_GLOBAL;
8849 posix_setlocale(LC_ALL, thread_locale);
8851 # else /* Must be USE_POSIX_2008_LOCALE) */
8853 const char * cur_thread_locales[LC_ALL_INDEX_];
8855 /* Save each category's current per-thread state */
8856 for_all_individual_category_indexes(i) {
8857 cur_thread_locales[i] = querylocale_i(i);
8860 CHANGE_SYSTEM_LOCALE_TO_GLOBAL;
8862 /* Set the global to what was our per-thread state */
8863 POSIX_SETLOCALE_LOCK;
8864 for_all_individual_category_indexes(i) {
8865 posix_setlocale(categories[i], cur_thread_locales[i]);
8867 POSIX_SETLOCALE_UNLOCK;
8870 # ifdef USE_LOCALE_NUMERIC
8872 /* Switch to the underlying C numeric locale; the application is on its
8874 POSIX_SETLOCALE_LOCK;
8875 posix_setlocale(LC_NUMERIC, PL_numeric_name);
8876 POSIX_SETLOCALE_UNLOCK;
8885 =for apidoc sync_locale
8887 This function copies the state of the program global locale into the calling
8888 thread, and converts that thread to using per-thread locales, if it wasn't
8889 already, and the platform supports them. The LC_NUMERIC locale is toggled into
8890 the standard state (using the C locale's conventions), if not within the
8891 lexical scope of S<C<use locale>>.
8893 Perl will now consider itself to have control of the locale.
8895 Since unthreaded perls have only a global locale, this function is a no-op
8898 This function is intended for use with C libraries that do locale manipulation.
8899 It allows Perl to accommodate the use of them. Call this function before
8900 transferring back to Perl space so that it knows what state the C code has left
8903 XS code should not manipulate the locale on its own. Instead,
8904 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
8905 change the locale (though changing the locale is antisocial and dangerous on
8906 multi-threaded systems that don't have multi-thread safe locale operations.
8907 (See L<perllocale/Multi-threaded operation>).
8909 Using the libc L<C<setlocale(3)>> function should be avoided. Nevertheless,
8910 certain non-Perl libraries called from XS, do call it, and their behavior may
8911 not be able to be changed. This function, along with
8912 C<L</switch_to_global_locale>>, can be used to get seamless behavior in these
8913 circumstances, as long as only one thread is involved.
8915 If the library has an option to turn off its locale manipulation, doing that is
8916 preferable to using this mechanism. C<Gtk> is such a library.
8918 The return value is a boolean: TRUE if the global locale at the time of call
8919 was in effect for the caller; and FALSE if a per-thread locale was in effect.
8925 Perl_sync_locale(pTHX)
8934 bool was_in_global = TRUE;
8936 # ifdef USE_THREAD_SAFE_LOCALE
8939 int config_return = _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
8940 if (config_return == -1) {
8941 locale_panic_("_configthreadlocale returned an error");
8943 was_in_global = (config_return == _DISABLE_PER_THREAD_LOCALE);
8945 # elif defined(USE_POSIX_2008_LOCALE)
8947 was_in_global = (LC_GLOBAL_LOCALE == uselocale(LC_GLOBAL_LOCALE));
8950 # error Unexpected Configuration
8952 # endif /* USE_THREAD_SAFE_LOCALE */
8954 /* Here, we are in the global locale. Get and save the values for each
8955 * category, and convert the current thread to use them */
8959 STDIZED_SETLOCALE_LOCK;
8960 const char * lc_all_string = savepv(stdized_setlocale(LC_ALL, NULL));
8961 STDIZED_SETLOCALE_UNLOCK;
8963 give_perl_locale_control(lc_all_string, __LINE__);
8964 Safefree(lc_all_string);
8968 const char * current_globals[LC_ALL_INDEX_];
8969 for_all_individual_category_indexes(i) {
8970 STDIZED_SETLOCALE_LOCK;
8971 current_globals[i] = savepv(stdized_setlocale(categories[i], NULL));
8972 STDIZED_SETLOCALE_UNLOCK;
8975 give_perl_locale_control((const char **) ¤t_globals, __LINE__);
8977 for_all_individual_category_indexes(i) {
8978 Safefree(current_globals[i]);
8983 return was_in_global;
8989 #if defined(DEBUGGING) && defined(USE_LOCALE)
8992 S_my_setlocale_debug_string_i(pTHX_
8993 const locale_category_index cat_index,
8994 const char* locale, /* Optional locale name */
8996 /* return value from setlocale() when attempting
8997 * to set 'category' to 'locale' */
9002 /* Returns a pointer to a NUL-terminated string in static storage with
9003 * added text about the info passed in. This is not thread safe and will
9004 * be overwritten by the next call, so this should be used just to
9005 * formulate a string to immediately print or savepv() on. */
9007 const char * locale_quote;
9008 const char * retval_quote;
9010 assert(cat_index <= LC_ALL_INDEX_);
9012 if (locale == NULL) {
9017 locale_quote = "\"";
9020 if (retval == NULL) {
9025 retval_quote = "\"";
9028 # ifdef USE_LOCALE_THREADS
9029 # define THREAD_FORMAT "%p:"
9030 # define THREAD_ARGUMENT aTHX_
9032 # define THREAD_FORMAT
9033 # define THREAD_ARGUMENT
9036 return Perl_form(aTHX_
9037 "%s:%" LINE_Tf ": " THREAD_FORMAT
9038 " setlocale(%s[%d], %s%s%s) returned %s%s%s\n",
9040 __FILE__, line, THREAD_ARGUMENT
9041 category_names[cat_index], categories[cat_index],
9042 locale_quote, locale, locale_quote,
9043 retval_quote, retval, retval_quote);
9047 #ifdef USE_PERL_SWITCH_LOCALE_CONTEXT
9050 Perl_switch_locale_context(pTHX)
9052 /* libc keeps per-thread locale status information in some configurations.
9053 * So, we can't just switch out aTHX to switch to a new thread. libc has
9054 * to follow along. This routine does that based on per-interpreter
9055 * variables we keep just for this purpose.
9057 * There are two implementations where this is an issue. For the other
9058 * implementations, it doesn't matter because libc is using global values
9059 * that all threads know about.
9061 * The two implementations are where libc keeps thread-specific information
9062 * on its own. These are
9064 * POSIX 2008: The current locale is kept by libc as an object. We save
9065 * a copy of that in the per-thread PL_cur_locale_obj, and so
9066 * this routine uses that copy to tell the thread it should be
9067 * operating with that object
9068 * Windows thread-safe locales: A given thread in Windows can be being run
9069 * with per-thread locales, or not. When the thread context
9070 * changes, libc doesn't automatically know if the thread is
9071 * using per-thread locales, nor does it know what the new
9072 * thread's locale is. We keep that information in the
9073 * per-thread variables:
9074 * PL_controls_locale indicates if this thread is using
9075 * per-thread locales or not
9076 * PL_cur_LC_ALL indicates what the the locale
9077 * should be if it is a per-thread
9081 if (UNLIKELY( PL_veto_switch_non_tTHX_context
9082 || PL_phase == PERL_PHASE_CONSTRUCT))
9087 # ifdef USE_POSIX_2008_LOCALE
9089 if (! uselocale(PL_cur_locale_obj)) {
9090 locale_panic_(Perl_form(aTHX_
9091 "Can't uselocale(%p), LC_ALL supposed to"
9093 PL_cur_locale_obj, get_LC_ALL_display()));
9096 # elif defined(WIN32)
9098 if (! bool_setlocale_c(LC_ALL, PL_cur_LC_ALL)) {
9099 locale_panic_(Perl_form(aTHX_ "Can't setlocale(%s)", PL_cur_LC_ALL));
9109 Perl_thread_locale_init(pTHX)
9112 #ifdef USE_THREAD_SAFE_LOCALE
9113 # ifdef USE_POSIX_2008_LOCALE
9115 /* Called from a thread on startup.
9117 * The operations here have to be done from within the calling thread, as
9118 * they affect libc's knowledge of the thread; libc has no knowledge of
9121 DEBUG_L(PerlIO_printf(Perl_debug_log,
9122 "new thread, initial locale is %s;"
9123 " calling setlocale(LC_ALL, \"C\")\n",
9124 get_LC_ALL_display()));
9126 if (! uselocale(PL_C_locale_obj)) {
9128 /* Not being able to change to the C locale is severe; don't keep
9130 locale_panic_(Perl_form(aTHX_
9131 "Can't uselocale(%p), 'C'", PL_C_locale_obj));
9132 NOT_REACHED; /* NOTREACHED */
9135 # ifdef MULTIPLICITY
9137 PL_cur_locale_obj = PL_C_locale_obj;
9140 # elif defined(WIN32)
9142 /* On Windows, make sure new thread has per-thread locales enabled */
9143 if (_configthreadlocale(_ENABLE_PER_THREAD_LOCALE) == -1) {
9144 locale_panic_("_configthreadlocale returned an error");
9146 void_setlocale_c(LC_ALL, "C");
9154 Perl_thread_locale_term(pTHX)
9156 /* Called from a thread as it gets ready to terminate.
9158 * The operations here have to be done from within the calling thread, as
9159 * they affect libc's knowledge of the thread; libc has no knowledge of
9162 #if defined(USE_POSIX_2008_LOCALE) && defined(USE_THREADS)
9164 /* Switch to the global locale, so can free up the per-thread object */
9165 locale_t actual_obj = uselocale(LC_GLOBAL_LOCALE);
9166 if (actual_obj != LC_GLOBAL_LOCALE && actual_obj != PL_C_locale_obj) {
9167 freelocale(actual_obj);
9170 /* Prevent leaks even if something has gone wrong */
9171 locale_t expected_obj = PL_cur_locale_obj;
9172 if (UNLIKELY( expected_obj != actual_obj
9173 && expected_obj != LC_GLOBAL_LOCALE
9174 && expected_obj != PL_C_locale_obj))
9176 freelocale(expected_obj);
9179 PL_cur_locale_obj = LC_GLOBAL_LOCALE;
9182 #ifdef WIN32_USE_FAKE_OLD_MINGW_LOCALES
9184 /* When faking the mingw implementation, we coerce this function into doing
9185 * something completely different from its intent -- namely to free up our
9186 * static buffer to avoid a leak. This function gets called for each
9187 * thread that is terminating, so will give us a chance to free the buffer
9188 * from the appropriate pool. On unthreaded systems, it gets called by the
9189 * mutex termination code. */
9191 # ifdef MULTIPLICITY
9193 if (aTHX != wsetlocale_buf_aTHX) {
9199 if (wsetlocale_buf_size > 0) {
9200 Safefree(wsetlocale_buf);
9201 wsetlocale_buf_size = 0;
9209 * ex: set ts=8 sts=4 sw=4 et: