If it looks like UTF-8 (either nl_langinfo or locale variables),

think UTF-8, embrace your inner UTF-8, as suggested by Larry.
(And as suggested by Markus Kuhn.)

While we are at it, document also the case of
mixed hash keys as a known potential troublemaker.
(Since it's locale-related, sometimes.)

p4raw-id: //depot/perl@15350

1 files changed, 39 insertions, 0 deletions

diff --git a/locale.c b/locale.c
index 22e903037a..2c84ab5cb2 100644
--- a/locale.c
+++ b/locale.c
@@ -25,6 +25,10 @@
 #  include <locale.h>
 #endif
 
+#ifdef I_LANGINFO
+#   include <langinfo.h>
+#endif
+
 /*
  * Standardize the locale name from a string returned by 'setlocale'.
  *
@@ -462,10 +466,45 @@ Perl_init_i18nl10n(pTHX_ int printwarn)
 #ifdef USE_LOCALE_NUMERIC
     new_numeric(curnum);
 #endif /* USE_LOCALE_NUMERIC */
+
     }
 
 #endif /* USE_LOCALE */
 
+    {
+	 bool wantutf8 = FALSE;
+	 char *codeset = NULL;
+#if defined(HAS_NL_LANGINFO) && defined(CODESET)
+	 codeset = nl_langinfo(CODESET);
+#endif
+	 if (codeset &&
+	     (ibcmp(codeset,  "UTF-8", 5) == 0 ||
+	      ibcmp(codeset,  "UTF8",  4) == 0))
+	      wantutf8 = TRUE;
+#ifdef __GLIBC__
+	 if (!wantutf8 && language &&
+	     (ibcmp(language, "UTF-8", 5) == 0 ||
+	      ibcmp(language, "UTF8",  4) == 0))
+	      wantutf8 = TRUE;
+#endif
+	 if (!wantutf8 && lc_all &&
+	     (ibcmp(lc_all,   "UTF-8", 5) == 0 ||
+	      ibcmp(lc_all,   "UTF8",  4) == 0))
+	      wantutf8 = TRUE;
+#ifdef USE_LOCALE_CTYPE
+	 if (!wantutf8 && curctype &&
+	     (ibcmp(curctype,     "UTF-8", 5) == 0 ||
+	      ibcmp(curctype,     "UTF8",  4) == 0))
+	      wantutf8 = TRUE;
+#endif
+	 if (!wantutf8 && lang &&
+	     (ibcmp(lang,     "UTF-8", 5) == 0 ||
+	      ibcmp(lang,     "UTF8",  4) == 0))
+	      wantutf8 = TRUE;
+	 if (wantutf8)
+	      PL_wantutf8 = TRUE;
+    }
+
 #ifdef USE_LOCALE_CTYPE
     if (curctype != NULL)
 	Safefree(curctype);