X-Git-Url: https://perl5.git.perl.org/perl5.git/blobdiff_plain/dfd77d7a1935ab78f2a7f1490ca419bcf7a2036b..a4a439fb9cd74c575855119abb55dc091955bdf4:/locale.c diff --git a/locale.c b/locale.c index 7332c4b..667a1bc 100644 --- a/locale.c +++ b/locale.c @@ -29,7 +29,9 @@ * in such scope than if not. However, various libc functions called by Perl * are affected by the LC_NUMERIC category, so there are macros in perl.h that * are used to toggle between the current locale and the C locale depending on - * the desired behavior of those functions at the moment. + * the desired behavior of those functions at the moment. And, LC_MESSAGES is + * switched to the C locale for outputting the message unless within the scope + * of 'use locale'. */ #include "EXTERN.h" @@ -117,10 +119,13 @@ Perl_set_numeric_radix(pTHX) else PL_numeric_radix_sv = NULL; - DEBUG_L(PerlIO_printf(Perl_debug_log, "Locale radix is %s\n", + DEBUG_L(PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n", (PL_numeric_radix_sv) - ? lc->decimal_point - : "NULL")); + ? SvPVX(PL_numeric_radix_sv) + : "NULL", + (PL_numeric_radix_sv) + ? cBOOL(SvUTF8(PL_numeric_radix_sv)) + : 0)); # endif /* HAS_LOCALECONV */ #endif /* USE_LOCALE_NUMERIC */ @@ -187,13 +192,17 @@ Perl_new_numeric(pTHX_ const char *newnum) } save_newnum = stdize_locale(savepv(newnum)); + + PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum); + PL_numeric_local = TRUE; + if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) { Safefree(PL_numeric_name); PL_numeric_name = save_newnum; } - - PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum); - PL_numeric_local = TRUE; + else { + Safefree(save_newnum); + } /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't * have to worry about the radix being a non-dot. (Core operations that @@ -350,7 +359,7 @@ Perl_new_ctype(pTHX_ const char *newctype) #ifdef MB_CUR_MAX /* We only handle single-byte locales (outside of UTF-8 ones; so if - * this locale requires than one byte, there are going to be + * this locale requires more than one byte, there are going to be * problems. */ if (check_for_problems && MB_CUR_MAX > 1 @@ -450,7 +459,21 @@ Perl_new_collate(pTHX_ const char *newcoll) * Any code changing the locale (outside this file) should use * POSIX::setlocale, which calls this function. Therefore this function * should be called directly only from this file and from - * POSIX::setlocale() */ + * POSIX::setlocale(). + * + * The design of locale collation is that every locale change is given an + * index 'PL_collation_ix'. The first time a string particpates in an + * operation that requires collation while locale collation is active, it + * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That + * magic includes the collation index, and the transformation of the string + * by strxfrm(), q.v. That transformation is used when doing comparisons, + * instead of the string itself. If a string changes, the magic is + * cleared. The next time the locale changes, the index is incremented, + * and so we know during a comparison that the transformation is not + * necessarily still valid, and so is recomputed. Note that if the locale + * changes enough times, the index could wrap (a U32), and it is possible + * that a transformation would improperly be considered valid, leading to + * an unlikely bug */ if (! newcoll) { if (PL_collation_name) { @@ -459,30 +482,171 @@ Perl_new_collate(pTHX_ const char *newcoll) PL_collation_name = NULL; } PL_collation_standard = TRUE; + is_standard_collation: PL_collxfrm_base = 0; PL_collxfrm_mult = 2; + PL_in_utf8_COLLATE_locale = FALSE; + *PL_strxfrm_min_char = '\0'; + PL_strxfrm_max_cp = 0; return; } + /* If this is not the same locale as currently, set the new one up */ if (! PL_collation_name || strNE(PL_collation_name, newcoll)) { ++PL_collation_ix; Safefree(PL_collation_name); PL_collation_name = stdize_locale(savepv(newcoll)); PL_collation_standard = isNAME_C_OR_POSIX(newcoll); + if (PL_collation_standard) { + goto is_standard_collation; + } + + PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE); + *PL_strxfrm_min_char = '\0'; + PL_strxfrm_max_cp = 0; + + /* A locale collation definition includes primary, secondary, tertiary, + * etc. weights for each character. To sort, the primary weights are + * used, and only if they compare equal, then the secondary weights are + * used, and only if they compare equal, then the tertiary, etc. + * + * strxfrm() works by taking the input string, say ABC, and creating an + * output transformed string consisting of first the primary weights, + * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the + * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters + * may not have weights at every level. In our example, let's say B + * doesn't have a tertiary weight, and A doesn't have a secondary + * weight. The constructed string is then going to be + * A¹B¹C¹ B²C² A³C³ .... + * This has the desired effect that strcmp() will look at the secondary + * or tertiary weights only if the strings compare equal at all higher + * priority weights. The spaces shown here, like in + * "A¹B¹C¹ * A²B²C² " + * are not just for readability. In the general case, these must + * actually be bytes, which we will call here 'separator weights'; and + * they must be smaller than any other weight value, but since these + * are C strings, only the terminating one can be a NUL (some + * implementations may include a non-NUL separator weight just before + * the NUL). Implementations tend to reserve 01 for the separator + * weights. They are needed so that a shorter string's secondary + * weights won't be misconstrued as primary weights of a longer string, + * etc. By making them smaller than any other weight, the shorter + * string will sort first. (Actually, if all secondary weights are + * smaller than all primary ones, there is no need for a separator + * weight between those two levels, etc.) + * + * The length of the transformed string is roughly a linear function of + * the input string. It's not exactly linear because some characters + * don't have weights at all levels. When we call strxfrm() we have to + * allocate some memory to hold the transformed string. The + * calculations below try to find coefficients 'm' and 'b' for this + * locale so that m*x + b equals how much space we need, given the size + * of the input string in 'x'. If we calculate too small, we increase + * the size as needed, and call strxfrm() again, but it is better to + * get it right the first time to avoid wasted expensive string + * transformations. */ { - /* 2: at most so many chars ('a', 'b'). */ - /* 50: surely no system expands a char more. */ -#define XFRMBUFSIZE (2 * 50) - char xbuf[XFRMBUFSIZE]; - const Size_t fa = strxfrm(xbuf, "a", XFRMBUFSIZE); - const Size_t fb = strxfrm(xbuf, "ab", XFRMBUFSIZE); - const SSize_t mult = fb - fa; - if (mult < 1 && !(fa == 0 && fb == 0)) - Perl_croak(aTHX_ "panic: strxfrm() gets absurd - a => %"UVuf", ab => %"UVuf, - (UV) fa, (UV) fb); - PL_collxfrm_base = (fa > (Size_t)mult) ? (fa - mult) : 0; - PL_collxfrm_mult = mult; + /* We use the string below to find how long the tranformation of it + * is. Almost all locales are supersets of ASCII, or at least the + * ASCII letters. We use all of them, half upper half lower, + * because if we used fewer, we might hit just the ones that are + * outliers in a particular locale. Most of the strings being + * collated will contain a preponderance of letters, and even if + * they are above-ASCII, they are likely to have the same number of + * weight levels as the ASCII ones. It turns out that digits tend + * to have fewer levels, and some punctuation has more, but those + * are relatively sparse in text, and khw believes this gives a + * reasonable result, but it could be changed if experience so + * dictates. */ + const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz"; + char * x_longer; /* Transformed 'longer' */ + Size_t x_len_longer; /* Length of 'x_longer' */ + + char * x_shorter; /* We also transform a substring of 'longer' */ + Size_t x_len_shorter; + + /* _mem_collxfrm() is used get the transformation (though here we + * are interested only in its length). It is used because it has + * the intelligence to handle all cases, but to work, it needs some + * values of 'm' and 'b' to get it started. For the purposes of + * this calculation we use a very conservative estimate of 'm' and + * 'b'. This assumes a weight can be multiple bytes, enough to + * hold any UV on the platform, and there are 5 levels, 4 weight + * bytes, and a trailing NUL. */ + PL_collxfrm_base = 5; + PL_collxfrm_mult = 5 * sizeof(UV); + + /* Find out how long the transformation really is */ + x_longer = _mem_collxfrm(longer, + sizeof(longer) - 1, + &x_len_longer, + + /* We avoid converting to UTF-8 in the + * called function by telling it the + * string is in UTF-8 if the locale is a + * UTF-8 one. Since the string passed + * here is invariant under UTF-8, we can + * claim it's UTF-8 even though it isn't. + * */ + PL_in_utf8_COLLATE_locale); + Safefree(x_longer); + + /* Find out how long the transformation of a substring of 'longer' + * is. Together the lengths of these transformations are + * sufficient to calculate 'm' and 'b'. The substring is all of + * 'longer' except the first character. This minimizes the chances + * of being swayed by outliers */ + x_shorter = _mem_collxfrm(longer + 1, + sizeof(longer) - 2, + &x_len_shorter, + PL_in_utf8_COLLATE_locale); + Safefree(x_shorter); + + /* If the results are nonsensical for this simple test, the whole + * locale definition is suspect. Mark it so that locale collation + * is not active at all for it. XXX Should we warn? */ + if ( x_len_shorter == 0 + || x_len_longer == 0 + || x_len_shorter >= x_len_longer) + { + PL_collxfrm_mult = 0; + PL_collxfrm_base = 0; + } + else { + SSize_t base; /* Temporary */ + + /* We have both: m * strlen(longer) + b = x_len_longer + * m * strlen(shorter) + b = x_len_shorter; + * subtracting yields: + * m * (strlen(longer) - strlen(shorter)) + * = x_len_longer - x_len_shorter + * But we have set things up so that 'shorter' is 1 byte smaller + * than 'longer'. Hence: + * m = x_len_longer - x_len_shorter + * + * But if something went wrong, make sure the multiplier is at + * least 1. + */ + if (x_len_longer > x_len_shorter) { + PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter; + } + else { + PL_collxfrm_mult = 1; + } + + /* mx + b = len + * so: b = len - mx + * but in case something has gone wrong, make sure it is + * non-negative */ + base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1); + if (base < 0) { + base = 0; + } + + /* Add 1 for the trailing NUL */ + PL_collxfrm_base = base + 1; + } } } @@ -508,7 +672,7 @@ Perl_my_setlocale(pTHX_ int category, const char* locale) * otherwise to use the particular category's variable if set; otherwise to * use the LANG variable. */ - bool override_LC_ALL = 0; + bool override_LC_ALL = FALSE; char * result; if (locale && strEQ(locale, "")) { @@ -569,6 +733,8 @@ Perl_my_setlocale(pTHX_ int category, const char* locale) } result = setlocale(category, locale); + DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__, + _setlocale_debug_string(category, locale, result))); if (! override_LC_ALL) { return result; @@ -583,41 +749,63 @@ Perl_my_setlocale(pTHX_ int category, const char* locale) result = PerlEnv_getenv("LC_TIME"); if (result && strNE(result, "")) { setlocale(LC_TIME, result); + DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_TIME, result, "not captured"))); } # endif # ifdef USE_LOCALE_CTYPE result = PerlEnv_getenv("LC_CTYPE"); if (result && strNE(result, "")) { setlocale(LC_CTYPE, result); + DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_CTYPE, result, "not captured"))); } # endif # ifdef USE_LOCALE_COLLATE result = PerlEnv_getenv("LC_COLLATE"); if (result && strNE(result, "")) { setlocale(LC_COLLATE, result); + DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_COLLATE, result, "not captured"))); } # endif # ifdef USE_LOCALE_MONETARY result = PerlEnv_getenv("LC_MONETARY"); if (result && strNE(result, "")) { setlocale(LC_MONETARY, result); + DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_MONETARY, result, "not captured"))); } # endif # ifdef USE_LOCALE_NUMERIC result = PerlEnv_getenv("LC_NUMERIC"); if (result && strNE(result, "")) { setlocale(LC_NUMERIC, result); + DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_NUMERIC, result, "not captured"))); } # endif # ifdef USE_LOCALE_MESSAGES result = PerlEnv_getenv("LC_MESSAGES"); if (result && strNE(result, "")) { setlocale(LC_MESSAGES, result); + DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_MESSAGES, result, "not captured"))); } # endif - return setlocale(LC_ALL, NULL); + result = setlocale(LC_ALL, NULL); + DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", + __FILE__, __LINE__, + _setlocale_debug_string(LC_ALL, NULL, result))); + return result; } #endif @@ -639,7 +827,42 @@ Perl_init_i18nl10n(pTHX_ int printwarn) * 1 = set ok or not applicable, * 0 = fallback to a locale of lower priority * -1 = fallback to all locales failed, not even to the C locale - */ + * + * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is + * set, debugging information is output. + * + * This looks more complicated than it is, mainly due to the #ifdefs. + * + * We try to set LC_ALL to the value determined by the environment. If + * there is no LC_ALL on this platform, we try the individual categories we + * know about. If this works, we are done. + * + * But if it doesn't work, we have to do something else. We search the + * environment variables ourselves instead of relying on the system to do + * it. We look at, in order, LC_ALL, LANG, a system default locale (if we + * think there is one), and the ultimate fallback "C". This is all done in + * the same loop as above to avoid duplicating code, but it makes things + * more complex. After the original failure, we add the fallback + * possibilities to the list of locales to try, and iterate the loop + * through them all until one succeeds. + * + * On Ultrix, the locale MUST come from the environment, so there is + * preliminary code to set it. I (khw) am not sure that it is necessary, + * and that this couldn't be folded into the loop, but barring any real + * platforms to test on, it's staying as-is + * + * A slight complication is that in embedded Perls, the locale may already + * be set-up, and we don't want to get it from the normal environment + * variables. This is handled by having a special environment variable + * indicate we're in this situation. We simply set setlocale's 2nd + * parameter to be a NULL instead of "". That indicates to setlocale that + * it is not to change anything, but to return the current value, + * effectively initializing perl's db to what the locale already is. + * + * We play the same trick with NULL if a LC_ALL succeeds. We call + * setlocale() on the individual categores with NULL to get their existing + * values for our db, instead of trying to change them. + * */ int ok = 1; @@ -654,25 +877,52 @@ Perl_init_i18nl10n(pTHX_ int printwarn) char *curnum = NULL; #endif /* USE_LOCALE_NUMERIC */ #ifdef __GLIBC__ - char * const language = PerlEnv_getenv("LANGUAGE"); + const char * const language = savepv(PerlEnv_getenv("LANGUAGE")); #endif /* NULL uses the existing already set up locale */ const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT")) ? NULL : ""; +#ifdef DEBUGGING + const bool debug = (PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")) + ? TRUE + : FALSE; +# define DEBUG_LOCALE_INIT(category, locale, result) \ + STMT_START { \ + if (debug) { \ + PerlIO_printf(Perl_debug_log, \ + "%s:%d: %s\n", \ + __FILE__, __LINE__, \ + _setlocale_debug_string(category, \ + locale, \ + result)); \ + } \ + } STMT_END +#else +# define DEBUG_LOCALE_INIT(a,b,c) +#endif const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */ unsigned int trial_locales_count; - char * const lc_all = PerlEnv_getenv("LC_ALL"); - char * const lang = PerlEnv_getenv("LANG"); + const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL")); + const char * const lang = savepv(PerlEnv_getenv("LANG")); bool setlocale_failure = FALSE; unsigned int i; char *p; - const bool locwarn = (printwarn > 1 || - (printwarn && - (!(p = PerlEnv_getenv("PERL_BADLANG")) || - grok_atou(p, NULL)))); + + /* A later getenv() could zap this, so only use here */ + const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG"); + + const bool locwarn = (printwarn > 1 + || (printwarn + && (! bad_lang_use_once + || ( + /* disallow with "" or "0" */ + *bad_lang_use_once + && strNE("0", bad_lang_use_once))))); bool done = FALSE; + char * sl_result; /* return from setlocale() */ + char * locale_param; #ifdef WIN32 /* In some systems you can find out the system default locale * and use that as the fallback locale. */ @@ -684,6 +934,7 @@ Perl_init_i18nl10n(pTHX_ int printwarn) #ifndef LOCALE_ENVIRON_REQUIRED PERL_UNUSED_VAR(done); + PERL_UNUSED_VAR(locale_param); #else /* @@ -693,55 +944,64 @@ Perl_init_i18nl10n(pTHX_ int printwarn) # ifdef LC_ALL if (lang) { - if (my_setlocale(LC_ALL, setlocale_init)) + sl_result = my_setlocale(LC_ALL, setlocale_init); + DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result); + if (sl_result) done = TRUE; else setlocale_failure = TRUE; } - if (!setlocale_failure) { + if (! setlocale_failure) { # ifdef USE_LOCALE_CTYPE - Safefree(curctype); - if (! (curctype = - my_setlocale(LC_CTYPE, - (!done && (lang || PerlEnv_getenv("LC_CTYPE"))) - ? setlocale_init : NULL))) + locale_param = (! done && (lang || PerlEnv_getenv("LC_CTYPE"))) + ? setlocale_init + : NULL; + curctype = my_setlocale(LC_CTYPE, locale_param); + DEBUG_LOCALE_INIT(LC_CTYPE, locale_param, sl_result); + if (! curctype) setlocale_failure = TRUE; else curctype = savepv(curctype); # endif /* USE_LOCALE_CTYPE */ # ifdef USE_LOCALE_COLLATE - Safefree(curcoll); - if (! (curcoll = - my_setlocale(LC_COLLATE, - (!done && (lang || PerlEnv_getenv("LC_COLLATE"))) - ? setlocale_init : NULL))) + locale_param = (! done && (lang || PerlEnv_getenv("LC_COLLATE"))) + ? setlocale_init + : NULL; + curcoll = my_setlocale(LC_COLLATE, locale_param); + DEBUG_LOCALE_INIT(LC_COLLATE, locale_param, sl_result); + if (! curcoll) setlocale_failure = TRUE; else curcoll = savepv(curcoll); # endif /* USE_LOCALE_COLLATE */ # ifdef USE_LOCALE_NUMERIC - Safefree(curnum); - if (! (curnum = - my_setlocale(LC_NUMERIC, - (!done && (lang || PerlEnv_getenv("LC_NUMERIC"))) - ? setlocale_init : NULL))) + locale_param = (! done && (lang || PerlEnv_getenv("LC_NUMERIC"))) + ? setlocale_init + : NULL; + curnum = my_setlocale(LC_NUMERIC, locale_param); + DEBUG_LOCALE_INIT(LC_NUMERIC, locale_param, sl_result); + if (! curnum) setlocale_failure = TRUE; else curnum = savepv(curnum); # endif /* USE_LOCALE_NUMERIC */ # ifdef USE_LOCALE_MESSAGES - if (! my_setlocale(LC_MESSAGES, - (!done && (lang || PerlEnv_getenv("LC_MESSAGES"))) - ? setlocale_init : NULL)) - { + locale_param = (! done && (lang || PerlEnv_getenv("LC_MESSAGES"))) + ? setlocale_init + : NULL; + sl_result = my_setlocale(LC_MESSAGES, locale_param); + DEBUG_LOCALE_INIT(LC_MESSAGES, locale_param, sl_result); + if (! sl_result) setlocale_failure = TRUE; } # endif /* USE_LOCALE_MESSAGES */ # ifdef USE_LOCALE_MONETARY - if (! my_setlocale(LC_MONETARY, - (!done && (lang || PerlEnv_getenv("LC_MONETARY"))) - ? setlocale_init : NULL)) - { + locale_param = (! done && (lang || PerlEnv_getenv("LC_MONETARY"))) + ? setlocale_init + : NULL; + sl_result = my_setlocale(LC_MONETARY, locale_param); + DEBUG_LOCALE_INIT(LC_MONETARY, locale_param, sl_result); + if (! sl_result) { setlocale_failure = TRUE; } # endif /* USE_LOCALE_MONETARY */ @@ -752,7 +1012,9 @@ Perl_init_i18nl10n(pTHX_ int printwarn) #endif /* !LOCALE_ENVIRON_REQUIRED */ /* We try each locale in the list until we get one that works, or exhaust - * the list */ + * the list. Normally the loop is executed just once. But if setting the + * locale fails, inside the loop we add fallback trials to the array and so + * will execute the loop multiple times */ trial_locales[0] = setlocale_init; trial_locales_count = 1; for (i= 0; i < trial_locales_count; i++) { @@ -775,6 +1037,7 @@ Perl_init_i18nl10n(pTHX_ int printwarn) /* Note that this may change the locale, but we are going to do * that anyway just below */ system_default_locale = setlocale(LC_ALL, ""); + DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale); /* Skip if invalid or it's already on the list of locales to * try */ @@ -794,7 +1057,9 @@ Perl_init_i18nl10n(pTHX_ int printwarn) } #ifdef LC_ALL - if (! my_setlocale(LC_ALL, trial_locale)) { + sl_result = my_setlocale(LC_ALL, trial_locale); + DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result); + if (! sl_result) { setlocale_failure = TRUE; } else { @@ -812,31 +1077,41 @@ Perl_init_i18nl10n(pTHX_ int printwarn) if (!setlocale_failure) { #ifdef USE_LOCALE_CTYPE Safefree(curctype); - if (! (curctype = my_setlocale(LC_CTYPE, trial_locale))) + curctype = my_setlocale(LC_CTYPE, trial_locale); + DEBUG_LOCALE_INIT(LC_CTYPE, trial_locale, curctype); + if (! curctype) setlocale_failure = TRUE; else curctype = savepv(curctype); #endif /* USE_LOCALE_CTYPE */ #ifdef USE_LOCALE_COLLATE Safefree(curcoll); - if (! (curcoll = my_setlocale(LC_COLLATE, trial_locale))) + curcoll = my_setlocale(LC_COLLATE, trial_locale); + DEBUG_LOCALE_INIT(LC_COLLATE, trial_locale, curcoll); + if (! curcoll) setlocale_failure = TRUE; else curcoll = savepv(curcoll); #endif /* USE_LOCALE_COLLATE */ #ifdef USE_LOCALE_NUMERIC Safefree(curnum); - if (! (curnum = my_setlocale(LC_NUMERIC, trial_locale))) + curnum = my_setlocale(LC_NUMERIC, trial_locale); + DEBUG_LOCALE_INIT(LC_NUMERIC, trial_locale, curnum); + if (! curnum) setlocale_failure = TRUE; else curnum = savepv(curnum); #endif /* USE_LOCALE_NUMERIC */ #ifdef USE_LOCALE_MESSAGES - if (! (my_setlocale(LC_MESSAGES, trial_locale))) + sl_result = my_setlocale(LC_MESSAGES, trial_locale); + DEBUG_LOCALE_INIT(LC_MESSAGES, trial_locale, sl_result); + if (! (sl_result)) setlocale_failure = TRUE; #endif /* USE_LOCALE_MESSAGES */ #ifdef USE_LOCALE_MONETARY - if (! (my_setlocale(LC_MONETARY, trial_locale))) + sl_result = my_setlocale(LC_MONETARY, trial_locale); + DEBUG_LOCALE_INIT(LC_MONETARY, trial_locale, sl_result); + if (! (sl_result)) setlocale_failure = TRUE; #endif /* USE_LOCALE_MONETARY */ @@ -861,18 +1136,18 @@ Perl_init_i18nl10n(pTHX_ int printwarn) PerlIO_printf(Perl_error_log, "perl: warning: Setting locale failed for the categories:\n\t"); -#ifdef USE_LOCALE_CTYPE +# ifdef USE_LOCALE_CTYPE if (! curctype) PerlIO_printf(Perl_error_log, "LC_CTYPE "); -#endif /* USE_LOCALE_CTYPE */ -#ifdef USE_LOCALE_COLLATE +# endif /* USE_LOCALE_CTYPE */ +# ifdef USE_LOCALE_COLLATE if (! curcoll) PerlIO_printf(Perl_error_log, "LC_COLLATE "); -#endif /* USE_LOCALE_COLLATE */ -#ifdef USE_LOCALE_NUMERIC +# endif /* USE_LOCALE_COLLATE */ +# ifdef USE_LOCALE_NUMERIC if (! curnum) PerlIO_printf(Perl_error_log, "LC_NUMERIC "); -#endif /* USE_LOCALE_NUMERIC */ +# endif /* USE_LOCALE_NUMERIC */ PerlIO_printf(Perl_error_log, "and possibly others\n"); #endif /* LC_ALL */ @@ -929,7 +1204,12 @@ Perl_init_i18nl10n(pTHX_ int printwarn) * LANG, and the C locale. We don't try the same locale twice, so * don't add to the list if already there. (On POSIX systems, the * LC_ALL element will likely be a repeat of the 0th element "", - * but there's no harm done by doing it explicitly */ + * but there's no harm done by doing it explicitly. + * + * Note that this tries the LC_ALL environment variable even on + * systems which have no LC_ALL locale setting. This may or may + * not have been originally intentional, but there's no real need + * to change the behavior. */ if (lc_all) { for (j = 0; j < trial_locales_count; j++) { if (strEQ(lc_all, trial_locales[j])) { @@ -994,14 +1274,17 @@ Perl_init_i18nl10n(pTHX_ int printwarn) #ifdef USE_LOCALE_CTYPE Safefree(curctype); curctype = savepv(setlocale(LC_CTYPE, NULL)); + DEBUG_LOCALE_INIT(LC_CTYPE, NULL, curctype); #endif /* USE_LOCALE_CTYPE */ #ifdef USE_LOCALE_COLLATE Safefree(curcoll); curcoll = savepv(setlocale(LC_COLLATE, NULL)); + DEBUG_LOCALE_INIT(LC_COLLATE, NULL, curcoll); #endif /* USE_LOCALE_COLLATE */ #ifdef USE_LOCALE_NUMERIC Safefree(curnum); curnum = savepv(setlocale(LC_NUMERIC, NULL)); + DEBUG_LOCALE_INIT(LC_NUMERIC, NULL, curnum); #endif /* USE_LOCALE_NUMERIC */ } @@ -1076,6 +1359,13 @@ Perl_init_i18nl10n(pTHX_ int printwarn) Safefree(curnum); #endif /* USE_LOCALE_NUMERIC */ +#ifdef __GLIBC__ + Safefree(language); +#endif + + Safefree(lc_all); + Safefree(lang); + #else /* !USE_LOCALE */ PERL_UNUSED_ARG(printwarn); #endif /* USE_LOCALE */ @@ -1086,66 +1376,380 @@ Perl_init_i18nl10n(pTHX_ int printwarn) #ifdef USE_LOCALE_COLLATE -/* - * mem_collxfrm() is a bit like strxfrm() but with two important - * differences. First, it handles embedded NULs. Second, it allocates - * a bit more memory than needed for the transformed data itself. - * The real transformed data begins at offset sizeof(collationix). - * Please see sv_collxfrm() to see how this is used. - */ - char * -Perl_mem_collxfrm(pTHX_ const char *s, STRLEN len, STRLEN *xlen) +Perl_mem_collxfrm(pTHX_ const char *input_string, STRLEN len, STRLEN *xlen) { - char *xbuf; - STRLEN xAlloc, xin, xout; /* xalloc is a reserved word in VC */ + /* This function is retained for compatibility in case someone outside core + * is using this (but it is undocumented) */ PERL_ARGS_ASSERT_MEM_COLLXFRM; - /* the first sizeof(collationix) bytes are used by sv_collxfrm(). */ - /* the +1 is for the terminating NUL. */ + return _mem_collxfrm(input_string, len, xlen, FALSE); +} + +char * +Perl__mem_collxfrm(pTHX_ const char *input_string, + STRLEN len, /* Length of 'input_string' */ + STRLEN *xlen, /* Set to length of returned string + (not including the collation index + prefix) */ + bool utf8 /* Is the input in UTF-8? */ + ) +{ + + /* _mem_collxfrm() is a bit like strxfrm() but with two important + * differences. First, it handles embedded NULs. Second, it allocates a bit + * more memory than needed for the transformed data itself. The real + * transformed data begins at offset sizeof(collationix). *xlen is set to + * the length of that, and doesn't include the collation index size. + * Please see sv_collxfrm() to see how this is used. */ + + char * s = (char *) input_string; + STRLEN s_strlen = strlen(input_string); + char *xbuf = NULL; + STRLEN xAlloc, xout; /* xalloc is a reserved word in VC */ + bool first_time = TRUE; /* Cleared after first loop iteration */ + + PERL_ARGS_ASSERT__MEM_COLLXFRM; + + /* Must be NUL-terminated */ + assert(*(input_string + len) == '\0'); + + /* If this locale has defective collation, skip */ + if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) { + goto bad; + } + + /* Replace any embedded NULs with the control that sorts before any others. + * This will give as good as possible results on strings that don't + * otherwise contain that character, but otherwise there may be + * less-than-perfect results with that character and NUL. This is + * unavoidable unless we replace strxfrm with our own implementation. + * + * XXX This code may be overkill. khw wrote it before realizing that if + * you change a NUL into some other character, that that may change the + * strxfrm results if that character is part of a sequence with other + * characters for weight calculations. To minimize the chances of this, + * now the replacement is restricted to another control (likely to be + * \001). But the full generality has been retained. + * + * This is one of the few places in the perl core, where we can use + * standard functions like strlen() and strcat(). It's because we're + * looking for NULs. */ + if (s_strlen < len) { + char * e = s + len; + char * sans_nuls; + STRLEN cur_min_char_len; + + /* If we don't know what control character sorts lowest for this + * locale, find it */ + if (*PL_strxfrm_min_char == '\0') { + int j; + char * cur_min_x = NULL; /* Cur cp's xfrm, (except it also + includes the collation index + prefixed. */ + + /* Look through all legal code points (NUL isn't) */ + for (j = 1; j < 256; j++) { + char * x; /* j's xfrm plus collation index */ + STRLEN x_len; /* length of 'x' */ + STRLEN trial_len = 1; + + /* Create a 1 byte string of the current code point, but with + * room to be 2 bytes */ + char cur_source[] = { (char) j, '\0' , '\0' }; + + if (PL_in_utf8_COLLATE_locale) { + if (! isCNTRL_L1(j)) { + continue; + } + + /* If needs to be 2 bytes, find them */ + if (! UVCHR_IS_INVARIANT(j)) { + char * d = cur_source; + append_utf8_from_native_byte((U8) j, (U8 **) &d); + trial_len = 2; + } + } + else if (! isCNTRL_LC(j)) { + continue; + } + + /* Then transform it */ + x = _mem_collxfrm(cur_source, trial_len, &x_len, + PL_in_utf8_COLLATE_locale); + + /* If something went wrong (which it shouldn't), just + * ignore this code point */ + if ( x_len == 0 + || strlen(x + sizeof(PL_collation_ix)) < x_len) + { + continue; + } + + /* If this character's transformation is lower than + * the current lowest, this one becomes the lowest */ + if ( cur_min_x == NULL + || strLT(x + sizeof(PL_collation_ix), + cur_min_x + sizeof(PL_collation_ix))) + { + strcpy(PL_strxfrm_min_char, cur_source); + cur_min_x = x; + } + else { + Safefree(x); + } + } /* end of loop through all bytes */ + + /* Unlikely, but possible, if there aren't any controls in the + * locale, arbitrarily use \001 */ + if (cur_min_x == NULL) { + STRLEN x_len; /* temporary */ + cur_min_x = _mem_collxfrm("\001", 1, &x_len, + PL_in_utf8_COLLATE_locale); + /* cur_min_cp was already initialized to 1 */ + } + + Safefree(cur_min_x); + } + + /* The worst case length for the replaced string would be if every + * character in it is NUL. Multiply that by the length of each + * replacement, and allow for a trailing NUL */ + cur_min_char_len = strlen(PL_strxfrm_min_char); + Newx(sans_nuls, (len * cur_min_char_len) + 1, char); + *sans_nuls = '\0'; + + + /* Replace each NUL with the lowest collating control. Loop until have + * exhausted all the NULs */ + while (s + s_strlen < e) { + strcat(sans_nuls, s); - xAlloc = sizeof(PL_collation_ix) + PL_collxfrm_base + (PL_collxfrm_mult * len) + 1; + /* Do the actual replacement */ + strcat(sans_nuls, PL_strxfrm_min_char); + + /* Move past the input NUL */ + s += s_strlen + 1; + s_strlen = strlen(s); + } + + /* And add anything that trails the final NUL */ + strcat(sans_nuls, s); + + /* Switch so below we transform this modified string */ + s = sans_nuls; + len = strlen(s); + } + + /* Make sure the UTF8ness of the string and locale match */ + if (utf8 != PL_in_utf8_COLLATE_locale) { + const char * const t = s; /* Temporary so we can later find where the + input was */ + + /* Here they don't match. Change the string's to be what the locale is + * expecting */ + + if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */ + s = (char *) bytes_to_utf8((const U8 *) s, &len); + utf8 = TRUE; + } + else { /* locale is not UTF-8; but input is; downgrade the input */ + + s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8); + + /* If the downgrade was successful we are done, but if the input + * contains things that require UTF-8 to represent, have to do + * damage control ... */ + if (UNLIKELY(utf8)) { + + /* What we do is construct a non-UTF-8 string with + * 1) the characters representable by a single byte converted + * to be so (if necessary); + * 2) and the rest converted to collate the same as the + * highest collating representable character. That makes + * them collate at the end. This is similar to how we + * handle embedded NULs, but we use the highest collating + * code point instead of the smallest. Like the NUL case, + * this isn't perfect, but is the best we can reasonably + * do. Every above-255 code point will sort the same as + * the highest-sorting 0-255 code point. If that code + * point can combine in a sequence with some other code + * points for weight calculations, us changing something to + * be it can adversely affect the results. But in most + * cases, it should work reasonably. And note that this is + * really an illegal situation: using code points above 255 + * on a locale where only 0-255 are valid. If two strings + * sort entirely equal, then the sort order for the + * above-255 code points will be in code point order. */ + + utf8 = FALSE; + + /* If we haven't calculated the code point with the maximum + * collating order for this locale, do so now */ + if (! PL_strxfrm_max_cp) { + int j; + + /* The current transformed string that collates the + * highest (except it also includes the prefixed collation + * index. */ + char * cur_max_x = NULL; + + /* Look through all legal code points (NUL isn't) */ + for (j = 1; j < 256; j++) { + char * x; + STRLEN x_len; + + /* Create a 1-char string of the current code point. */ + char cur_source[] = { (char) j, '\0' }; + + /* Then transform it */ + x = _mem_collxfrm(cur_source, 1, &x_len, FALSE); + + /* If something went wrong (which it shouldn't), just + * ignore this code point */ + if (x_len == 0) { + Safefree(x); + continue; + } + + /* If this character's transformation is higher than + * the current highest, this one becomes the highest */ + if ( cur_max_x == NULL + || strGT(x + sizeof(PL_collation_ix), + cur_max_x + sizeof(PL_collation_ix))) + { + PL_strxfrm_max_cp = j; + cur_max_x = x; + } + else { + Safefree(x); + } + } + + Safefree(cur_max_x); + } + + /* Here we know which legal code point collates the highest. + * We are ready to construct the non-UTF-8 string. The length + * will be at least 1 byte smaller than the input string + * (because we changed at least one 2-byte character into a + * single byte), but that is eaten up by the trailing NUL */ + Newx(s, len, char); + + { + STRLEN i; + STRLEN d= 0; + + for (i = 0; i < len; i+= UTF8SKIP(t + i)) { + U8 cur_char = t[i]; + if (UTF8_IS_INVARIANT(cur_char)) { + s[d++] = cur_char; + } + else if (UTF8_IS_DOWNGRADEABLE_START(cur_char)) { + s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]); + } + else { /* Replace illegal cp with highest collating + one */ + s[d++] = PL_strxfrm_max_cp; + } + } + s[d++] = '\0'; + Renew(s, d, char); /* Free up unused space */ + } + } + } + + /* Here, we have constructed a modified version of the input. It could + * be that we already had a modified copy before we did this version. + * If so, that copy is no longer needed */ + if (t != input_string) { + Safefree(t); + } + } + + /* The first element in the output is the collation id, used by + * sv_collxfrm(); then comes the space for the transformed string. The + * equation should give us a good estimate as to how much is needed */ + xAlloc = sizeof(PL_collation_ix) + + PL_collxfrm_base + + (PL_collxfrm_mult * ((utf8) + ? utf8_length((U8 *) s, (U8 *) s + len) + : len)); Newx(xbuf, xAlloc, char); - if (! xbuf) + if (UNLIKELY(! xbuf)) goto bad; + /* Store the collation id */ *(U32*)xbuf = PL_collation_ix; xout = sizeof(PL_collation_ix); - for (xin = 0; xin < len; ) { - Size_t xused; - - for (;;) { - xused = strxfrm(xbuf + xout, s + xin, xAlloc - xout); - if (xused >= PERL_INT_MAX) - goto bad; - if ((STRLEN)xused < xAlloc - xout) - break; - xAlloc = (2 * xAlloc) + 1; - Renew(xbuf, xAlloc, char); - if (! xbuf) - goto bad; - } - xin += strlen(s + xin) + 1; - xout += xused; + /* Then the transformation of the input. We loop until successful, or we + * give up */ + for (;;) { + STRLEN xused = strxfrm(xbuf + xout, s, xAlloc - xout); + + /* If the transformed string occupies less space than we told strxfrm() + * was available, it means it successfully transformed the whole + * string. */ + if (xused < xAlloc - xout) { + xout += xused; + break; + } + + if (UNLIKELY(xused >= PERL_INT_MAX)) + goto bad; + + /* A well-behaved strxfrm() returns exactly how much space it needs + * (not including the trailing NUL) when it fails due to not enough + * space being provided. Assume that this is the case unless it's been + * proven otherwise */ + if (LIKELY(PL_strxfrm_is_behaved) && first_time) { + xAlloc = xused + sizeof(PL_collation_ix) + 1; + } + else { /* Here, either: + * 1) The strxfrm() has previously shown bad behavior; or + * 2) It isn't the first time through the loop, which means + * that the strxfrm() is now showing bad behavior, because + * we gave it what it said was needed in the previous + * iteration, and it came back saying it needed still more. + * (Many versions of cygwin fit this. When the buffer size + * isn't sufficient, they return the input size instead of + * how much is needed.) + * Increase the buffer size by a fixed percentage and try again. */ + xAlloc = (2 * xAlloc) + 1; + PL_strxfrm_is_behaved = FALSE; + } + + + Renew(xbuf, xAlloc, char); + if (UNLIKELY(! xbuf)) + goto bad; - /* Embedded NULs are understood but silently skipped - * because they make no sense in locale collation. */ + first_time = FALSE; } - xbuf[xout] = '\0'; *xlen = xout - sizeof(PL_collation_ix); + + /* Free up unneeded space; retain ehough for trailing NUL */ + Renew(xbuf, xout + 1, char); + + if (s != input_string) { + Safefree(s); + } + return xbuf; bad: Safefree(xbuf); + if (s != input_string) { + Safefree(s); + } *xlen = 0; return NULL; } #endif /* USE_LOCALE_COLLATE */ - #ifdef USE_LOCALE bool @@ -1674,13 +2278,21 @@ Perl__is_in_locale_category(pTHX_ const bool compiling, const int category) char * Perl_my_strerror(pTHX_ const int errnum) { + dVAR; /* Uses C locale for the error text unless within scope of 'use locale' for * LC_MESSAGES */ #ifdef USE_LOCALE_MESSAGES if (! IN_LC(LC_MESSAGES)) { - char * save_locale = setlocale(LC_MESSAGES, NULL); + char * save_locale; + + /* We have a critical section to prevent another thread from changing + * the locale out from under us (or zapping the buffer returned from + * setlocale() ) */ + LOCALE_LOCK; + + save_locale = setlocale(LC_MESSAGES, NULL); if (! isNAME_C_OR_POSIX(save_locale)) { char *errstr; @@ -1695,8 +2307,13 @@ Perl_my_strerror(pTHX_ const int errnum) { setlocale(LC_MESSAGES, save_locale); Safefree(save_locale); + + LOCALE_UNLOCK; + return errstr; } + + LOCALE_UNLOCK; } #endif @@ -1736,14 +2353,99 @@ Perl_sync_locale(pTHX) } +#if defined(DEBUGGING) && defined(USE_LOCALE) + +char * +Perl__setlocale_debug_string(const int category, /* category number, + like LC_ALL */ + const char* const locale, /* locale name */ + + /* return value from setlocale() when attempting to + * set 'category' to 'locale' */ + const char* const retval) +{ + /* Returns a pointer to a NUL-terminated string in static storage with + * added text about the info passed in. This is not thread safe and will + * be overwritten by the next call, so this should be used just to + * formulate a string to immediately print or savepv() on. */ + + /* initialise to a non-null value to keep it out of BSS and so keep + * -DPERL_GLOBAL_STRUCT_PRIVATE happy */ + static char ret[128] = "x"; + + my_strlcpy(ret, "setlocale(", sizeof(ret)); + + switch (category) { + default: + my_snprintf(ret, sizeof(ret), "%s? %d", ret, category); + break; +# ifdef LC_ALL + case LC_ALL: + my_strlcat(ret, "LC_ALL", sizeof(ret)); + break; +# endif +# ifdef LC_CTYPE + case LC_CTYPE: + my_strlcat(ret, "LC_CTYPE", sizeof(ret)); + break; +# endif +# ifdef LC_NUMERIC + case LC_NUMERIC: + my_strlcat(ret, "LC_NUMERIC", sizeof(ret)); + break; +# endif +# ifdef LC_COLLATE + case LC_COLLATE: + my_strlcat(ret, "LC_COLLATE", sizeof(ret)); + break; +# endif +# ifdef LC_TIME + case LC_TIME: + my_strlcat(ret, "LC_TIME", sizeof(ret)); + break; +# endif +# ifdef LC_MONETARY + case LC_MONETARY: + my_strlcat(ret, "LC_MONETARY", sizeof(ret)); + break; +# endif +# ifdef LC_MESSAGES + case LC_MESSAGES: + my_strlcat(ret, "LC_MESSAGES", sizeof(ret)); + break; +# endif + } + + my_strlcat(ret, ", ", sizeof(ret)); + + if (locale) { + my_strlcat(ret, "\"", sizeof(ret)); + my_strlcat(ret, locale, sizeof(ret)); + my_strlcat(ret, "\"", sizeof(ret)); + } + else { + my_strlcat(ret, "NULL", sizeof(ret)); + } + + my_strlcat(ret, ") returned ", sizeof(ret)); + + if (retval) { + my_strlcat(ret, "\"", sizeof(ret)); + my_strlcat(ret, retval, sizeof(ret)); + my_strlcat(ret, "\"", sizeof(ret)); + } + else { + my_strlcat(ret, "NULL", sizeof(ret)); + } + + assert(strlen(ret) < sizeof(ret)); + + return ret; +} + +#endif /* - * Local variables: - * c-indentation-style: bsd - * c-basic-offset: 4 - * indent-tabs-mode: nil - * End: - * * ex: set ts=8 sts=4 sw=4 et: */