X-Git-Url: http://erislabs.net/gitweb/?a=blobdiff_plain;f=lib%2Ficonv_open.c;h=f1d61b9212717c0c4205222f261430706d87de72;hb=74dd1966a5db708c66201bf952976ef27bd33e7e;hp=f78cbb48d13794d147b55bc4f7bfd7ffdbdd886d;hpb=8119dc1622611078f2fca227cf94d65ab1765ce5;p=gnulib.git

diff --git a/lib/iconv_open.c b/lib/iconv_open.c
index f78cbb48d..f1d61b921 100644
--- a/lib/iconv_open.c
+++ b/lib/iconv_open.c
@@ -23,20 +23,23 @@
 #include <errno.h>
 #include <string.h>
 #include "c-ctype.h"
+#include "c-strcase.h"
 
 #define SIZEOF(a) (sizeof(a) / sizeof(a[0]))
 
 /* Namespace cleanliness.  */
 #define mapping_lookup rpl_iconv_open_mapping_lookup
 
-/* The macro ICONV_FLAVOR is defined to one of these.  */
+/* The macro ICONV_FLAVOR is defined to one of these or undefined.  */
 
 #define ICONV_FLAVOR_AIX "iconv_open-aix.h"
 #define ICONV_FLAVOR_HPUX "iconv_open-hpux.h"
 #define ICONV_FLAVOR_IRIX "iconv_open-irix.h"
 #define ICONV_FLAVOR_OSF "iconv_open-osf.h"
 
-#include ICONV_FLAVOR
+#ifdef ICONV_FLAVOR
+# include ICONV_FLAVOR
+#endif
 
 iconv_t
 rpl_iconv_open (const char *tocode, const char *fromcode)
@@ -47,6 +50,59 @@ rpl_iconv_open (const char *tocode, const char *fromcode)
   char *fromcode_upper_end;
   char *tocode_upper_end;
 
+#if REPLACE_ICONV_UTF
+  /* Special handling of conversion between UTF-8 and UTF-{16,32}{BE,LE}.
+     Do this here, before calling the real iconv_open(), because  OSF/1 5.1
+     iconv() to these encoding inserts a BOM, which is wrong.
+     We do not need to handle conversion between arbitrary encodings and
+     UTF-{16,32}{BE,LE}, because the 'striconveh' module implements two-step
+     conversion throough UTF-8.
+     The _ICONV_* constants are chosen to be disjoint from any iconv_t
+     returned by the system's iconv_open() functions.  Recall that iconv_t
+     is a scalar type.  */
+  if (c_toupper (fromcode[0]) == 'U'
+      && c_toupper (fromcode[1]) == 'T'
+      && c_toupper (fromcode[2]) == 'F'
+      && fromcode[3] == '-')
+    {
+      if (c_toupper (tocode[0]) == 'U'
+	  && c_toupper (tocode[1]) == 'T'
+	  && c_toupper (tocode[2]) == 'F'
+	  && tocode[3] == '-')
+	{
+	  if (strcmp (fromcode + 4, "8") == 0)
+	    {
+	      if (c_strcasecmp (tocode + 4, "16BE") == 0)
+		return _ICONV_UTF8_UTF16BE;
+	      if (c_strcasecmp (tocode + 4, "16LE") == 0)
+		return _ICONV_UTF8_UTF16LE;
+	      if (c_strcasecmp (tocode + 4, "32BE") == 0)
+		return _ICONV_UTF8_UTF32BE;
+	      if (c_strcasecmp (tocode + 4, "32LE") == 0)
+		return _ICONV_UTF8_UTF32LE;
+	    }
+	  else if (strcmp (tocode + 4, "8") == 0)
+	    {
+	      if (c_strcasecmp (fromcode + 4, "16BE") == 0)
+		return _ICONV_UTF16BE_UTF8;
+	      if (c_strcasecmp (fromcode + 4, "16LE") == 0)
+		return _ICONV_UTF16LE_UTF8;
+	      if (c_strcasecmp (fromcode + 4, "32BE") == 0)
+		return _ICONV_UTF32BE_UTF8;
+	      if (c_strcasecmp (fromcode + 4, "32LE") == 0)
+		return _ICONV_UTF32LE_UTF8;
+	    }
+	}
+    }
+#endif
+
+  /* Do *not* add special support for 8-bit encodings like ASCII or ISO-8859-1
+     here.  This would lead to programs that work in some locales (such as the
+     "C" or "en_US" locales) but do not work in East Asian locales.  It is
+     better if programmers make their programs depend on GNU libiconv (except
+     on glibc systems), e.g. by using the AM_ICONV macro and documenting the
+     dependency in an INSTALL or DEPENDENCIES file.  */
+
   /* Try with the original names first.
      This covers the case when fromcode or tocode is a lowercase encoding name
      that is understood by the system's iconv_open but not listed in our
@@ -93,6 +149,7 @@ rpl_iconv_open (const char *tocode, const char *fromcode)
     tocode_upper_end = q;
   }
 
+#ifdef ICONV_FLAVOR
   /* Apply the mappings.  */
   {
     const struct mapping *m =
@@ -106,6 +163,10 @@ rpl_iconv_open (const char *tocode, const char *fromcode)
 
     tocode = (m != NULL ? m->vendor_name : tocode_upper);
   }
+#else
+  fromcode = fromcode_upper;
+  tocode = tocode_upper;
+#endif
 
   return iconv_open (tocode, fromcode);
 }