summaryrefslogtreecommitdiff
path: root/tests/setlocale.c
diff options
context:
space:
mode:
authorJörg Frings-Fürst <debian@jff.email>2022-02-07 06:57:44 +0100
committerJörg Frings-Fürst <debian@jff.email>2022-02-07 06:57:44 +0100
commit6ffaa52cf732a84c0f8259dbc9b8183960ec5c33 (patch)
treed3136c22e3e805cfc0efe78412cce905a17e0c8d /tests/setlocale.c
parent591cbf5878b7d4105b275585b9db1324c9dfce56 (diff)
parentaf151c593704a40e10efb28d1e3dd59a23a661f3 (diff)
Merge branch 'release/debian/1.0-1'debian/1.0-1
Diffstat (limited to 'tests/setlocale.c')
-rw-r--r--tests/setlocale.c787
1 files changed, 757 insertions, 30 deletions
diff --git a/tests/setlocale.c b/tests/setlocale.c
index 1ac3f4d..3ea3559 100644
--- a/tests/setlocale.c
+++ b/tests/setlocale.c
@@ -1,17 +1,17 @@
/* Set the current locale. -*- coding: utf-8 -*-
- Copyright (C) 2009, 2011-2018 Free Software Foundation, Inc.
+ Copyright (C) 2009, 2011-2022 Free Software Foundation, Inc.
- This program is free software: you can redistribute it and/or modify
- it under the terms of the GNU General Public License as published by
- the Free Software Foundation; either version 3 of the License, or
- (at your option) any later version.
+ This file is free software: you can redistribute it and/or modify
+ it under the terms of the GNU Lesser General Public License as
+ published by the Free Software Foundation; either version 3 of the
+ License, or (at your option) any later version.
- This program is distributed in the hope that it will be useful,
+ This file is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
- GNU General Public License for more details.
+ GNU Lesser General Public License for more details.
- You should have received a copy of the GNU General Public License
+ You should have received a copy of the GNU Lesser General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>. */
/* Written by Bruno Haible <bruno@clisp.org>, 2009. */
@@ -29,15 +29,59 @@
/* Specification. */
#include <locale.h>
+#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "localename.h"
+#if HAVE_CFLOCALECOPYPREFERREDLANGUAGES || HAVE_CFPREFERENCESCOPYAPPVALUE
+# if HAVE_CFLOCALECOPYPREFERREDLANGUAGES
+# include <CoreFoundation/CFLocale.h>
+# elif HAVE_CFPREFERENCESCOPYAPPVALUE
+# include <CoreFoundation/CFPreferences.h>
+# endif
+# include <CoreFoundation/CFPropertyList.h>
+# include <CoreFoundation/CFArray.h>
+# include <CoreFoundation/CFString.h>
+extern void gl_locale_name_canonicalize (char *name);
+#endif
+
#if 1
# undef setlocale
+/* Which of the replacements to activate? */
+# if NEED_SETLOCALE_IMPROVED
+# define setlocale_improved rpl_setlocale
+# elif NEED_SETLOCALE_MTSAFE
+# define setlocale_mtsafe rpl_setlocale
+# else
+# error "This file should only be compiled if NEED_SETLOCALE_IMPROVED || NEED_SETLOCALE_MTSAFE."
+# endif
+
+/* Like setlocale, but guaranteed to be multithread-safe if LOCALE == NULL. */
+# if !SETLOCALE_NULL_ALL_MTSAFE || !SETLOCALE_NULL_ONE_MTSAFE /* i.e. if NEED_SETLOCALE_MTSAFE */
+
+# if NEED_SETLOCALE_IMPROVED
+static
+# endif
+char *
+setlocale_mtsafe (int category, const char *locale)
+{
+ if (locale == NULL)
+ return (char *) setlocale_null (category);
+ else
+ return setlocale (category, locale);
+}
+# else /* !NEED_SETLOCALE_MTSAFE */
+
+# define setlocale_mtsafe setlocale
+
+# endif /* NEED_SETLOCALE_MTSAFE */
+
+# if NEED_SETLOCALE_IMPROVED
+
/* Return string representation of locale category CATEGORY. */
static const char *
category_to_name (int category)
@@ -72,7 +116,7 @@ category_to_name (int category)
return retval;
}
-# if defined _WIN32 && ! defined __CYGWIN__
+# if defined _WIN32 && ! defined __CYGWIN__
/* The native Windows setlocale() function expects locale names of the form
"German" or "German_Germany" or "DEU", but not "de" or "de_DE". We need
@@ -604,7 +648,7 @@ search (const struct table_entry *table, size_t table_size, const char *string,
{
size_t i;
- for (i = mid; i < hi; i++)
+ for (i = mid + 1; i < hi; i++)
{
if (strcmp (table[i].code, string) > 0)
{
@@ -635,13 +679,11 @@ setlocale_unixlike (int category, const char *locale)
/* The native Windows implementation of setlocale understands the special
locale name "C", but not "POSIX". Therefore map "POSIX" to "C". */
-#if defined _WIN32 && !defined __CYGWIN__
if (locale != NULL && strcmp (locale, "POSIX") == 0)
locale = "C";
-#endif
/* First, try setlocale with the original argument unchanged. */
- result = setlocale (category, locale);
+ result = setlocale_mtsafe (category, locale);
if (result != NULL)
return result;
@@ -783,11 +825,44 @@ setlocale_unixlike (int category, const char *locale)
return NULL;
}
-# else
-# define setlocale_unixlike setlocale
-# endif
+# elif defined __ANDROID__
+
+/* Like setlocale, but accept also the locale names "C" and "POSIX". */
+static char *
+setlocale_unixlike (int category, const char *locale)
+{
+ char *result = setlocale_mtsafe (category, locale);
+ if (result == NULL)
+ switch (category)
+ {
+ case LC_CTYPE:
+ case LC_NUMERIC:
+ case LC_TIME:
+ case LC_COLLATE:
+ case LC_MONETARY:
+ case LC_MESSAGES:
+ case LC_ALL:
+ case LC_PAPER:
+ case LC_NAME:
+ case LC_ADDRESS:
+ case LC_TELEPHONE:
+ case LC_MEASUREMENT:
+ if (locale == NULL
+ || strcmp (locale, "C") == 0 || strcmp (locale, "POSIX") == 0)
+ result = (char *) "C";
+ break;
+ default:
+ break;
+ }
+ return result;
+}
+# define setlocale setlocale_unixlike
-# if LC_MESSAGES == 1729
+# else
+# define setlocale_unixlike setlocale_mtsafe
+# endif
+
+# if LC_MESSAGES == 1729
/* The system does not store an LC_MESSAGES locale category. Do it here. */
static char lc_messages_name[64] = "C";
@@ -809,12 +884,533 @@ setlocale_single (int category, const char *locale)
return setlocale_unixlike (category, locale);
}
-# else
-# define setlocale_single setlocale_unixlike
-# endif
+# else
+# define setlocale_single setlocale_unixlike
+# endif
+
+# if defined __APPLE__ && defined __MACH__
+
+/* Mapping from language to main territory where that language is spoken. */
+static char const locales_with_principal_territory[][6 + 1] =
+ {
+ /* Language Main territory */
+ "ace_ID", /* Achinese Indonesia */
+ "af_ZA", /* Afrikaans South Africa */
+ "ak_GH", /* Akan Ghana */
+ "am_ET", /* Amharic Ethiopia */
+ "an_ES", /* Aragonese Spain */
+ "ang_GB", /* Old English Britain */
+ "arn_CL", /* Mapudungun Chile */
+ "as_IN", /* Assamese India */
+ "ast_ES", /* Asturian Spain */
+ "av_RU", /* Avaric Russia */
+ "awa_IN", /* Awadhi India */
+ "az_AZ", /* Azerbaijani Azerbaijan */
+ "ban_ID", /* Balinese Indonesia */
+ "be_BY", /* Belarusian Belarus */
+ "bej_SD", /* Beja Sudan */
+ "bem_ZM", /* Bemba Zambia */
+ "bg_BG", /* Bulgarian Bulgaria */
+ "bho_IN", /* Bhojpuri India */
+ "bi_VU", /* Bislama Vanuatu */
+ "bik_PH", /* Bikol Philippines */
+ "bin_NG", /* Bini Nigeria */
+ "bm_ML", /* Bambara Mali */
+ "bn_IN", /* Bengali India */
+ "bo_CN", /* Tibetan China */
+ "br_FR", /* Breton France */
+ "bs_BA", /* Bosnian Bosnia */
+ "bug_ID", /* Buginese Indonesia */
+ "ca_ES", /* Catalan Spain */
+ "ce_RU", /* Chechen Russia */
+ "ceb_PH", /* Cebuano Philippines */
+ "co_FR", /* Corsican France */
+ "cr_CA", /* Cree Canada */
+ /* Don't put "crh_UZ" or "crh_UA" here. That would be asking for fruitless
+ political discussion. */
+ "cs_CZ", /* Czech Czech Republic */
+ "csb_PL", /* Kashubian Poland */
+ "cy_GB", /* Welsh Britain */
+ "da_DK", /* Danish Denmark */
+ "de_DE", /* German Germany */
+ "din_SD", /* Dinka Sudan */
+ "doi_IN", /* Dogri India */
+ "dsb_DE", /* Lower Sorbian Germany */
+ "dv_MV", /* Divehi Maldives */
+ "dz_BT", /* Dzongkha Bhutan */
+ "ee_GH", /* Éwé Ghana */
+ "el_GR", /* Greek Greece */
+ /* Don't put "en_GB" or "en_US" here. That would be asking for fruitless
+ political discussion. */
+ "es_ES", /* Spanish Spain */
+ "et_EE", /* Estonian Estonia */
+ "fa_IR", /* Persian Iran */
+ "fi_FI", /* Finnish Finland */
+ "fil_PH", /* Filipino Philippines */
+ "fj_FJ", /* Fijian Fiji */
+ "fo_FO", /* Faroese Faeroe Islands */
+ "fon_BJ", /* Fon Benin */
+ "fr_FR", /* French France */
+ "fur_IT", /* Friulian Italy */
+ "fy_NL", /* Western Frisian Netherlands */
+ "ga_IE", /* Irish Ireland */
+ "gd_GB", /* Scottish Gaelic Britain */
+ "gon_IN", /* Gondi India */
+ "gsw_CH", /* Swiss German Switzerland */
+ "gu_IN", /* Gujarati India */
+ "he_IL", /* Hebrew Israel */
+ "hi_IN", /* Hindi India */
+ "hil_PH", /* Hiligaynon Philippines */
+ "hr_HR", /* Croatian Croatia */
+ "hsb_DE", /* Upper Sorbian Germany */
+ "ht_HT", /* Haitian Haiti */
+ "hu_HU", /* Hungarian Hungary */
+ "hy_AM", /* Armenian Armenia */
+ "id_ID", /* Indonesian Indonesia */
+ "ig_NG", /* Igbo Nigeria */
+ "ii_CN", /* Sichuan Yi China */
+ "ilo_PH", /* Iloko Philippines */
+ "is_IS", /* Icelandic Iceland */
+ "it_IT", /* Italian Italy */
+ "ja_JP", /* Japanese Japan */
+ "jab_NG", /* Hyam Nigeria */
+ "jv_ID", /* Javanese Indonesia */
+ "ka_GE", /* Georgian Georgia */
+ "kab_DZ", /* Kabyle Algeria */
+ "kaj_NG", /* Jju Nigeria */
+ "kam_KE", /* Kamba Kenya */
+ "kmb_AO", /* Kimbundu Angola */
+ "kcg_NG", /* Tyap Nigeria */
+ "kdm_NG", /* Kagoma Nigeria */
+ "kg_CD", /* Kongo Democratic Republic of Congo */
+ "kk_KZ", /* Kazakh Kazakhstan */
+ "kl_GL", /* Kalaallisut Greenland */
+ "km_KH", /* Central Khmer Cambodia */
+ "kn_IN", /* Kannada India */
+ "ko_KR", /* Korean Korea (South) */
+ "kok_IN", /* Konkani India */
+ "kr_NG", /* Kanuri Nigeria */
+ "kru_IN", /* Kurukh India */
+ "ky_KG", /* Kyrgyz Kyrgyzstan */
+ "lg_UG", /* Ganda Uganda */
+ "li_BE", /* Limburgish Belgium */
+ "lo_LA", /* Laotian Laos */
+ "lt_LT", /* Lithuanian Lithuania */
+ "lu_CD", /* Luba-Katanga Democratic Republic of Congo */
+ "lua_CD", /* Luba-Lulua Democratic Republic of Congo */
+ "luo_KE", /* Luo Kenya */
+ "lv_LV", /* Latvian Latvia */
+ "mad_ID", /* Madurese Indonesia */
+ "mag_IN", /* Magahi India */
+ "mai_IN", /* Maithili India */
+ "mak_ID", /* Makasar Indonesia */
+ "man_ML", /* Mandingo Mali */
+ "men_SL", /* Mende Sierra Leone */
+ "mfe_MU", /* Mauritian Creole Mauritius */
+ "mg_MG", /* Malagasy Madagascar */
+ "mi_NZ", /* Maori New Zealand */
+ "min_ID", /* Minangkabau Indonesia */
+ "mk_MK", /* Macedonian North Macedonia */
+ "ml_IN", /* Malayalam India */
+ "mn_MN", /* Mongolian Mongolia */
+ "mni_IN", /* Manipuri India */
+ "mos_BF", /* Mossi Burkina Faso */
+ "mr_IN", /* Marathi India */
+ "ms_MY", /* Malay Malaysia */
+ "mt_MT", /* Maltese Malta */
+ "mwr_IN", /* Marwari India */
+ "my_MM", /* Burmese Myanmar */
+ "na_NR", /* Nauru Nauru */
+ "nah_MX", /* Nahuatl Mexico */
+ "nap_IT", /* Neapolitan Italy */
+ "nb_NO", /* Norwegian Bokmål Norway */
+ "nds_DE", /* Low Saxon Germany */
+ "ne_NP", /* Nepali Nepal */
+ "nl_NL", /* Dutch Netherlands */
+ "nn_NO", /* Norwegian Nynorsk Norway */
+ "no_NO", /* Norwegian Norway */
+ "nr_ZA", /* South Ndebele South Africa */
+ "nso_ZA", /* Northern Sotho South Africa */
+ "ny_MW", /* Chichewa Malawi */
+ "nym_TZ", /* Nyamwezi Tanzania */
+ "nyn_UG", /* Nyankole Uganda */
+ "oc_FR", /* Occitan France */
+ "oj_CA", /* Ojibwa Canada */
+ "or_IN", /* Oriya India */
+ "pa_IN", /* Punjabi India */
+ "pag_PH", /* Pangasinan Philippines */
+ "pam_PH", /* Pampanga Philippines */
+ "pap_AN", /* Papiamento Netherlands Antilles - this line can be removed in 2018 */
+ "pbb_CO", /* Páez Colombia */
+ "pl_PL", /* Polish Poland */
+ "ps_AF", /* Pashto Afghanistan */
+ "pt_PT", /* Portuguese Portugal */
+ "raj_IN", /* Rajasthani India */
+ "rm_CH", /* Romansh Switzerland */
+ "rn_BI", /* Kirundi Burundi */
+ "ro_RO", /* Romanian Romania */
+ "ru_RU", /* Russian Russia */
+ "rw_RW", /* Kinyarwanda Rwanda */
+ "sa_IN", /* Sanskrit India */
+ "sah_RU", /* Yakut Russia */
+ "sas_ID", /* Sasak Indonesia */
+ "sat_IN", /* Santali India */
+ "sc_IT", /* Sardinian Italy */
+ "scn_IT", /* Sicilian Italy */
+ "sg_CF", /* Sango Central African Republic */
+ "shn_MM", /* Shan Myanmar */
+ "si_LK", /* Sinhala Sri Lanka */
+ "sid_ET", /* Sidamo Ethiopia */
+ "sk_SK", /* Slovak Slovakia */
+ "sl_SI", /* Slovenian Slovenia */
+ "sm_WS", /* Samoan Samoa */
+ "smn_FI", /* Inari Sami Finland */
+ "sms_FI", /* Skolt Sami Finland */
+ "so_SO", /* Somali Somalia */
+ "sq_AL", /* Albanian Albania */
+ "sr_RS", /* Serbian Serbia */
+ "srr_SN", /* Serer Senegal */
+ "suk_TZ", /* Sukuma Tanzania */
+ "sus_GN", /* Susu Guinea */
+ "sv_SE", /* Swedish Sweden */
+ "te_IN", /* Telugu India */
+ "tem_SL", /* Timne Sierra Leone */
+ "tet_ID", /* Tetum Indonesia */
+ "tg_TJ", /* Tajik Tajikistan */
+ "th_TH", /* Thai Thailand */
+ "ti_ER", /* Tigrinya Eritrea */
+ "tiv_NG", /* Tiv Nigeria */
+ "tk_TM", /* Turkmen Turkmenistan */
+ "tl_PH", /* Tagalog Philippines */
+ "to_TO", /* Tonga Tonga */
+ "tpi_PG", /* Tok Pisin Papua New Guinea */
+ "tr_TR", /* Turkish Turkey */
+ "tum_MW", /* Tumbuka Malawi */
+ "ug_CN", /* Uighur China */
+ "uk_UA", /* Ukrainian Ukraine */
+ "umb_AO", /* Umbundu Angola */
+ "ur_PK", /* Urdu Pakistan */
+ "uz_UZ", /* Uzbek Uzbekistan */
+ "ve_ZA", /* Venda South Africa */
+ "vi_VN", /* Vietnamese Vietnam */
+ "wa_BE", /* Walloon Belgium */
+ "wal_ET", /* Walamo Ethiopia */
+ "war_PH", /* Waray Philippines */
+ "wen_DE", /* Sorbian Germany */
+ "yao_MW", /* Yao Malawi */
+ "zap_MX" /* Zapotec Mexico */
+ };
+
+/* Compare just the language part of two locale names. */
+static int
+langcmp (const char *locale1, const char *locale2)
+{
+ size_t locale1_len;
+ size_t locale2_len;
+ int cmp;
+
+ {
+ const char *locale1_end = strchr (locale1, '_');
+ if (locale1_end != NULL)
+ locale1_len = locale1_end - locale1;
+ else
+ locale1_len = strlen (locale1);
+ }
+ {
+ const char *locale2_end = strchr (locale2, '_');
+ if (locale2_end != NULL)
+ locale2_len = locale2_end - locale2;
+ else
+ locale2_len = strlen (locale2);
+ }
+
+ if (locale1_len < locale2_len)
+ {
+ cmp = memcmp (locale1, locale2, locale1_len);
+ if (cmp == 0)
+ cmp = -1;
+ }
+ else
+ {
+ cmp = memcmp (locale1, locale2, locale2_len);
+ if (locale1_len > locale2_len && cmp == 0)
+ cmp = 1;
+ }
+
+ return cmp;
+}
+
+/* Given a locale name, return the main locale with the same language,
+ or NULL if not found.
+ For example: "fr_DE" -> "fr_FR". */
+static const char *
+get_main_locale_with_same_language (const char *locale)
+{
+# define table locales_with_principal_territory
+ /* The table is sorted. Perform a binary search. */
+ size_t hi = sizeof (table) / sizeof (table[0]);
+ size_t lo = 0;
+ while (lo < hi)
+ {
+ /* Invariant:
+ for i < lo, langcmp (table[i], locale) < 0,
+ for i >= hi, langcmp (table[i], locale) > 0. */
+ size_t mid = (hi + lo) >> 1; /* >= lo, < hi */
+ int cmp = langcmp (table[mid], locale);
+ if (cmp < 0)
+ lo = mid + 1;
+ else if (cmp > 0)
+ hi = mid;
+ else
+ {
+ /* Found an i with
+ langcmp (language_table[i], locale) == 0.
+ Verify that it is the only such i. */
+ if (mid > lo && langcmp (table[mid - 1], locale) >= 0)
+ abort ();
+ if (mid + 1 < hi && langcmp (table[mid + 1], locale) <= 0)
+ abort ();
+ return table[mid];
+ }
+ }
+# undef table
+ return NULL;
+}
+
+/* Mapping from territory to main language that is spoken in that territory. */
+static char const locales_with_principal_language[][6 + 1] =
+ {
+ /* This is based on the set of existing locales in glibc, with duplicates
+ removed, and on the Wikipedia pages named "Languages of <territory>".
+ If in doubt, use the locale that exists in macOS. For example, the only
+ "*_IN" locale in macOS 10.13 is "hi_IN", so use that. */
+ /* A useful shell function for producing a line of this table is:
+ func_line ()
+ {
+ # Usage: func_line ll_CC
+ ll=`echo "$1" | sed -e 's|_.*||'`
+ cc=`echo "$1" | sed -e 's|^.*_||'`
+ llx=`sed -n -e "s|^${ll} ||p" < gettext-tools/doc/ISO_639`
+ ccx=`expand gettext-tools/doc/ISO_3166 | sed -n -e "s|^${cc} *||p"`
+ echo " \"$1\", /$X* ${llx} ${ccx} *$X/"
+ }
+ */
+ /* Main language Territory */
+ "ca_AD", /* Catalan Andorra */
+ "ar_AE", /* Arabic United Arab Emirates */
+ "ps_AF", /* Pashto Afghanistan */
+ "en_AG", /* English Antigua and Barbuda */
+ "sq_AL", /* Albanian Albania */
+ "hy_AM", /* Armenian Armenia */
+ "pap_AN", /* Papiamento Netherlands Antilles - this line can be removed in 2018 */
+ "pt_AO", /* Portuguese Angola */
+ "es_AR", /* Spanish Argentina */
+ "de_AT", /* German Austria */
+ "en_AU", /* English Australia */
+ /* Aruba has two official languages: "nl_AW", "pap_AW". */
+ "az_AZ", /* Azerbaijani Azerbaijan */
+ "bs_BA", /* Bosnian Bosnia */
+ "bn_BD", /* Bengali Bangladesh */
+ "nl_BE", /* Dutch Belgium */
+ "fr_BF", /* French Burkina Faso */
+ "bg_BG", /* Bulgarian Bulgaria */
+ "ar_BH", /* Arabic Bahrain */
+ "rn_BI", /* Kirundi Burundi */
+ "fr_BJ", /* French Benin */
+ "es_BO", /* Spanish Bolivia */
+ "pt_BR", /* Portuguese Brazil */
+ "dz_BT", /* Dzongkha Bhutan */
+ "en_BW", /* English Botswana */
+ "be_BY", /* Belarusian Belarus */
+ "en_CA", /* English Canada */
+ "fr_CD", /* French Democratic Republic of Congo */
+ "sg_CF", /* Sango Central African Republic */
+ "de_CH", /* German Switzerland */
+ "es_CL", /* Spanish Chile */
+ "zh_CN", /* Chinese China */
+ "es_CO", /* Spanish Colombia */
+ "es_CR", /* Spanish Costa Rica */
+ "es_CU", /* Spanish Cuba */
+ /* Curaçao has three official languages: "nl_CW", "pap_CW", "en_CW". */
+ "el_CY", /* Greek Cyprus */
+ "cs_CZ", /* Czech Czech Republic */
+ "de_DE", /* German Germany */
+ /* Djibouti has two official languages: "ar_DJ" and "fr_DJ". */
+ "da_DK", /* Danish Denmark */
+ "es_DO", /* Spanish Dominican Republic */
+ "ar_DZ", /* Arabic Algeria */
+ "es_EC", /* Spanish Ecuador */
+ "et_EE", /* Estonian Estonia */
+ "ar_EG", /* Arabic Egypt */
+ "ti_ER", /* Tigrinya Eritrea */
+ "es_ES", /* Spanish Spain */
+ "am_ET", /* Amharic Ethiopia */
+ "fi_FI", /* Finnish Finland */
+ /* Fiji has three official languages: "en_FJ", "fj_FJ", "hif_FJ". */
+ "fo_FO", /* Faroese Faeroe Islands */
+ "fr_FR", /* French France */
+ "en_GB", /* English Britain */
+ "ka_GE", /* Georgian Georgia */
+ "en_GH", /* English Ghana */
+ "kl_GL", /* Kalaallisut Greenland */
+ "fr_GN", /* French Guinea */
+ "el_GR", /* Greek Greece */
+ "es_GT", /* Spanish Guatemala */
+ "zh_HK", /* Chinese Hong Kong */
+ "es_HN", /* Spanish Honduras */
+ "hr_HR", /* Croatian Croatia */
+ "ht_HT", /* Haitian Haiti */
+ "hu_HU", /* Hungarian Hungary */
+ "id_ID", /* Indonesian Indonesia */
+ "en_IE", /* English Ireland */
+ "he_IL", /* Hebrew Israel */
+ "hi_IN", /* Hindi India */
+ "ar_IQ", /* Arabic Iraq */
+ "fa_IR", /* Persian Iran */
+ "is_IS", /* Icelandic Iceland */
+ "it_IT", /* Italian Italy */
+ "ar_JO", /* Arabic Jordan */
+ "ja_JP", /* Japanese Japan */
+ "sw_KE", /* Swahili Kenya */
+ "ky_KG", /* Kyrgyz Kyrgyzstan */
+ "km_KH", /* Central Khmer Cambodia */
+ "ko_KR", /* Korean Korea (South) */
+ "ar_KW", /* Arabic Kuwait */
+ "kk_KZ", /* Kazakh Kazakhstan */
+ "lo_LA", /* Laotian Laos */
+ "ar_LB", /* Arabic Lebanon */
+ "de_LI", /* German Liechtenstein */
+ "si_LK", /* Sinhala Sri Lanka */
+ "lt_LT", /* Lithuanian Lithuania */
+ /* Luxembourg has three official languages: "lb_LU", "fr_LU", "de_LU". */
+ "lv_LV", /* Latvian Latvia */
+ "ar_LY", /* Arabic Libya */
+ "ar_MA", /* Arabic Morocco */
+ "sr_ME", /* Serbian Montenegro */
+ "mg_MG", /* Malagasy Madagascar */
+ "mk_MK", /* Macedonian North Macedonia */
+ "fr_ML", /* French Mali */
+ "my_MM", /* Burmese Myanmar */
+ "mn_MN", /* Mongolian Mongolia */
+ "mt_MT", /* Maltese Malta */
+ "mfe_MU", /* Mauritian Creole Mauritius */
+ "dv_MV", /* Divehi Maldives */
+ "ny_MW", /* Chichewa Malawi */
+ "es_MX", /* Spanish Mexico */
+ "ms_MY", /* Malay Malaysia */
+ "en_NG", /* English Nigeria */
+ "es_NI", /* Spanish Nicaragua */
+ "nl_NL", /* Dutch Netherlands */
+ "no_NO", /* Norwegian Norway */
+ "ne_NP", /* Nepali Nepal */
+ "na_NR", /* Nauru Nauru */
+ "niu_NU", /* Niuean Niue */
+ "en_NZ", /* English New Zealand */
+ "ar_OM", /* Arabic Oman */
+ "es_PA", /* Spanish Panama */
+ "es_PE", /* Spanish Peru */
+ "tpi_PG", /* Tok Pisin Papua New Guinea */
+ "fil_PH", /* Filipino Philippines */
+ "pa_PK", /* Punjabi Pakistan */
+ "pl_PL", /* Polish Poland */
+ "es_PR", /* Spanish Puerto Rico */
+ "pt_PT", /* Portuguese Portugal */
+ "es_PY", /* Spanish Paraguay */
+ "ar_QA", /* Arabic Qatar */
+ "ro_RO", /* Romanian Romania */
+ "sr_RS", /* Serbian Serbia */
+ "ru_RU", /* Russian Russia */
+ "rw_RW", /* Kinyarwanda Rwanda */
+ "ar_SA", /* Arabic Saudi Arabia */
+ "en_SC", /* English Seychelles */
+ "ar_SD", /* Arabic Sudan */
+ "sv_SE", /* Swedish Sweden */
+ "en_SG", /* English Singapore */
+ "sl_SI", /* Slovenian Slovenia */
+ "sk_SK", /* Slovak Slovakia */
+ "en_SL", /* English Sierra Leone */
+ "fr_SN", /* French Senegal */
+ "so_SO", /* Somali Somalia */
+ "ar_SS", /* Arabic South Sudan */
+ "es_SV", /* Spanish El Salvador */
+ "ar_SY", /* Arabic Syria */
+ "th_TH", /* Thai Thailand */
+ "tg_TJ", /* Tajik Tajikistan */
+ "tk_TM", /* Turkmen Turkmenistan */
+ "ar_TN", /* Arabic Tunisia */
+ "to_TO", /* Tonga Tonga */
+ "tr_TR", /* Turkish Turkey */
+ "zh_TW", /* Chinese Taiwan */
+ "sw_TZ", /* Swahili Tanzania */
+ "uk_UA", /* Ukrainian Ukraine */
+ "lg_UG", /* Ganda Uganda */
+ "en_US", /* English United States of America */
+ "es_UY", /* Spanish Uruguay */
+ "uz_UZ", /* Uzbek Uzbekistan */
+ "es_VE", /* Spanish Venezuela */
+ "vi_VN", /* Vietnamese Vietnam */
+ "bi_VU", /* Bislama Vanuatu */
+ "sm_WS", /* Samoan Samoa */
+ "ar_YE", /* Arabic Yemen */
+ "en_ZA", /* English South Africa */
+ "en_ZM", /* English Zambia */
+ "en_ZW" /* English Zimbabwe */
+ };
+
+/* Compare just the territory part of two locale names. */
+static int
+terrcmp (const char *locale1, const char *locale2)
+{
+ const char *territory1 = strrchr (locale1, '_') + 1;
+ const char *territory2 = strrchr (locale2, '_') + 1;
+
+ return strcmp (territory1, territory2);
+}
+
+/* Given a locale name, return the locale corresponding to the main language
+ with the same territory, or NULL if not found.
+ For example: "fr_DE" -> "de_DE". */
+static const char *
+get_main_locale_with_same_territory (const char *locale)
+{
+ if (strrchr (locale, '_') != NULL)
+ {
+# define table locales_with_principal_language
+ /* The table is sorted. Perform a binary search. */
+ size_t hi = sizeof (table) / sizeof (table[0]);
+ size_t lo = 0;
+ while (lo < hi)
+ {
+ /* Invariant:
+ for i < lo, terrcmp (table[i], locale) < 0,
+ for i >= hi, terrcmp (table[i], locale) > 0. */
+ size_t mid = (hi + lo) >> 1; /* >= lo, < hi */
+ int cmp = terrcmp (table[mid], locale);
+ if (cmp < 0)
+ lo = mid + 1;
+ else if (cmp > 0)
+ hi = mid;
+ else
+ {
+ /* Found an i with
+ terrcmp (language_table[i], locale) == 0.
+ Verify that it is the only such i. */
+ if (mid > lo && terrcmp (table[mid - 1], locale) >= 0)
+ abort ();
+ if (mid + 1 < hi && terrcmp (table[mid + 1], locale) <= 0)
+ abort ();
+ return table[mid];
+ }
+ }
+# undef table
+ }
+ return NULL;
+}
+
+# endif
char *
-rpl_setlocale (int category, const char *locale)
+setlocale_improved (int category, const char *locale)
{
if (locale != NULL && locale[0] == '\0')
{
@@ -824,6 +1420,7 @@ rpl_setlocale (int category, const char *locale)
/* Set LC_CTYPE first. Then the other categories. */
static int const categories[] =
{
+ LC_CTYPE,
LC_NUMERIC,
LC_TIME,
LC_COLLATE,
@@ -850,18 +1447,31 @@ rpl_setlocale (int category, const char *locale)
if (base_name == NULL)
base_name = gl_locale_name_default ();
- if (setlocale_unixlike (LC_ALL, base_name) == NULL)
- goto fail;
-# if defined _WIN32 && ! defined __CYGWIN__
+ if (setlocale_unixlike (LC_ALL, base_name) != NULL)
+ {
+ /* LC_CTYPE category already set. */
+ i = 1;
+ }
+ else
+ {
+ /* On Mac OS X, "UTF-8" is a valid locale name for LC_CTYPE but
+ not for LC_ALL. Therefore this call may fail. So, try
+ another base_name. */
+ base_name = "C";
+ if (setlocale_unixlike (LC_ALL, base_name) == NULL)
+ goto fail;
+ i = 0;
+ }
+# if defined _WIN32 && ! defined __CYGWIN__
/* On native Windows, setlocale(LC_ALL,...) may succeed but set the
LC_CTYPE category to an invalid value ("C") when it does not
support the specified encoding. Report a failure instead. */
if (strchr (base_name, '.') != NULL
&& strcmp (setlocale (LC_CTYPE, NULL), "C") == 0)
goto fail;
-# endif
+# endif
- for (i = 0; i < sizeof (categories) / sizeof (categories[0]); i++)
+ for (; i < sizeof (categories) / sizeof (categories[0]); i++)
{
int cat = categories[i];
const char *name;
@@ -873,12 +1483,127 @@ rpl_setlocale (int category, const char *locale)
/* If name is the same as base_name, it has already been set
through the setlocale call before the loop. */
if (strcmp (name, base_name) != 0
-# if LC_MESSAGES == 1729
+# if LC_MESSAGES == 1729
|| cat == LC_MESSAGES
-# endif
+# endif
)
if (setlocale_single (cat, name) == NULL)
+# if defined __APPLE__ && defined __MACH__
+ {
+ /* On Mac OS X 10.13, some locales can be set through
+ System Preferences > Language & Region, that are not
+ supported by libc. The system's setlocale() falls
+ back to "C" for these locale categories. We can do
+ better, by trying an existing locale with the same
+ language or an existing locale with the same territory.
+ If we can't, print a warning, to limit user
+ expectations. */
+ int warn = 0;
+
+ if (cat == LC_CTYPE)
+ warn = (setlocale_single (cat, "UTF-8") == NULL);
+ else if (cat == LC_MESSAGES)
+ {
+# if HAVE_CFLOCALECOPYPREFERREDLANGUAGES || HAVE_CFPREFERENCESCOPYAPPVALUE /* MacOS X 10.4 or newer */
+ /* Take the primary language preference. */
+# if HAVE_CFLOCALECOPYPREFERREDLANGUAGES /* MacOS X 10.5 or newer */
+ CFArrayRef prefArray = CFLocaleCopyPreferredLanguages ();
+# elif HAVE_CFPREFERENCESCOPYAPPVALUE /* MacOS X 10.4 or newer */
+ CFTypeRef preferences =
+ CFPreferencesCopyAppValue (CFSTR ("AppleLanguages"),
+ kCFPreferencesCurrentApplication);
+ if (preferences != NULL
+ && CFGetTypeID (preferences) == CFArrayGetTypeID ())
+ {
+ CFArrayRef prefArray = (CFArrayRef)preferences;
+# endif
+ int n = CFArrayGetCount (prefArray);
+ if (n > 0)
+ {
+ char buf[256];
+ CFTypeRef element = CFArrayGetValueAtIndex (prefArray, 0);
+ if (element != NULL
+ && CFGetTypeID (element) == CFStringGetTypeID ()
+ && CFStringGetCString ((CFStringRef)element,
+ buf, sizeof (buf),
+ kCFStringEncodingASCII))
+ {
+ /* Remove the country.
+ E.g. "zh-Hans-DE" -> "zh-Hans". */
+ char *last_minus = strrchr (buf, '-');
+ if (last_minus != NULL)
+ *last_minus = '\0';
+
+ /* Convert to Unix locale name.
+ E.g. "zh-Hans" -> "zh_CN". */
+ gl_locale_name_canonicalize (buf);
+
+ /* Try setlocale with this value. */
+ if (setlocale_single (cat, buf) == NULL)
+ {
+ const char *last_try =
+ get_main_locale_with_same_language (buf);
+
+ if (last_try == NULL
+ || setlocale_single (cat, last_try) == NULL)
+ warn = 1;
+ }
+ }
+ }
+# if HAVE_CFLOCALECOPYPREFERREDLANGUAGES /* MacOS X 10.5 or newer */
+ CFRelease (prefArray);
+# elif HAVE_CFPREFERENCESCOPYAPPVALUE /* MacOS X 10.4 or newer */
+ }
+# endif
+# else
+ const char *last_try =
+ get_main_locale_with_same_language (name);
+
+ if (last_try == NULL
+ || setlocale_single (cat, last_try) == NULL)
+ warn = 1;
+# endif
+ }
+ else
+ {
+ /* For LC_NUMERIC, the application should use the locale
+ properties kCFLocaleDecimalSeparator,
+ kCFLocaleGroupingSeparator.
+ For LC_TIME, the application should use the locale
+ property kCFLocaleCalendarIdentifier.
+ For LC_COLLATE, the application should use the locale
+ properties kCFLocaleCollationIdentifier,
+ kCFLocaleCollatorIdentifier.
+ For LC_MONETARY, the applicationshould use the locale
+ properties kCFLocaleCurrencySymbol,
+ kCFLocaleCurrencyCode.
+ But since most applications don't have macOS specific
+ code like this, try an existing locale with the same
+ territory. */
+ const char *last_try =
+ get_main_locale_with_same_territory (name);
+
+ if (last_try == NULL
+ || setlocale_single (cat, last_try) == NULL)
+ warn = 1;
+ }
+
+ if (warn)
+ {
+ /* Warn only if the environment variable
+ SETLOCALE_VERBOSE is set. Otherwise these warnings
+ are just annoyances, since normal users won't invoke
+ 'localedef'. */
+ const char *verbose = getenv ("SETLOCALE_VERBOSE");
+ if (verbose != NULL && verbose[0] != '\0')
+ fprintf (stderr,
+ "Warning: Failed to set locale category %s to %s.\n",
+ category_to_name (cat), name);
+ }
+ }
+# else
goto fail;
+# endif
}
/* All steps were successful. */
@@ -903,7 +1628,7 @@ rpl_setlocale (int category, const char *locale)
}
else
{
-# if defined _WIN32 && ! defined __CYGWIN__
+# if defined _WIN32 && ! defined __CYGWIN__
if (category == LC_ALL && locale != NULL && strchr (locale, '.') != NULL)
{
char *saved_locale;
@@ -938,9 +1663,11 @@ rpl_setlocale (int category, const char *locale)
return setlocale (LC_ALL, NULL);
}
else
-# endif
+# endif
return setlocale_single (category, locale);
}
}
+# endif /* NEED_SETLOCALE_IMPROVED */
+
#endif