Call nl_langinfo(CODESET) to get the name of the locale encoding on Unix

3 files changed, 26 insertions, 3 deletions

diff --git a/libraries/base/GHC/IO/Encoding/Iconv.hs b/libraries/base/GHC/IO/Encoding/Iconv.hs
index 237468a76f..25e22ab6bb 100644
--- a/libraries/base/GHC/IO/Encoding/Iconv.hs
+++ b/libraries/base/GHC/IO/Encoding/Iconv.hs
@@ -1,4 +1,4 @@
-{-# OPTIONS_GHC -fno-implicit-prelude -#include "HsBase.h" #-}
+{-# OPTIONS_GHC -fno-implicit-prelude #-}
 -----------------------------------------------------------------------------
 -- |
 -- Module      :  GHC.IO.Encoding.Iconv
@@ -25,6 +25,8 @@ module GHC.IO.Encoding.Iconv (
 #endif
  ) where
 
+#include "HsBaseConfig.h"
+
 #if !defined(mingw32_HOST_OS)
 
 #undef DEBUG_DUMP
@@ -98,7 +100,15 @@ utf32be = unsafePerformIO (mkTextEncoding "UTF32BE")
 
 {-# NOINLINE localeEncoding #-}
 localeEncoding :: TextEncoding
-localeEncoding = unsafePerformIO (mkTextEncoding "")
+localeEncoding = unsafePerformIO $ do
+#if HAVE_LANGINFO_H
+   cstr <- c_localeEncoding -- use nl_langinfo(CODESET) to get the encoding
+                               -- if we have it
+   r <- peekCString cstr
+   mkTextEncoding r
+#else
+   mkTextEncoding "" -- GNU iconv accepts "" to mean the -- locale encoding.
+#endif
 
 -- We hope iconv_t is a storable type.  It should be, since it has at least the
 -- value -1, which is a possible return value from iconv_open.
@@ -114,6 +124,9 @@ foreign import ccall unsafe "iconv"
     iconv :: IConv -> Ptr CString -> Ptr CSize -> Ptr CString -> Ptr CSize
 	  -> IO CSize
 
+foreign import ccall unsafe "localeEncoding"
+    c_localeEncoding :: IO CString
+
 haskellChar :: String
 #ifdef WORDS_BIGENDIAN
 haskellChar | charSize == 2 = "UTF16BE"
diff --git a/libraries/base/cbits/PrelIOUtils.c b/libraries/base/cbits/PrelIOUtils.c
index de7fed4387..6444bd04e6 100644
--- a/libraries/base/cbits/PrelIOUtils.c
+++ b/libraries/base/cbits/PrelIOUtils.c
@@ -24,4 +24,14 @@ void debugBelch2(const char*s, char *t)
 {
     debugBelch(s,t);
 }
+
+// Use a C wrapper for this because we avoid hsc2hs in base
+#if HAVE_LANGINFO_H
+#include <langinfo.h>
+char *localeEncoding (void)
+{
+    return nl_langinfo(CODESET);
+}
+#endif
+
 #endif /* __GLASGOW_HASKELL__ */
diff --git a/libraries/base/configure.ac b/libraries/base/configure.ac
index 4ae0be1cb5..cd153c3bc0 100644
--- a/libraries/base/configure.ac
+++ b/libraries/base/configure.ac
@@ -17,7 +17,7 @@ dnl ** check for full ANSI header (.h) files
 AC_HEADER_STDC
 
 # check for specific header (.h) files that we are interested in
-AC_CHECK_HEADERS([ctype.h dirent.h errno.h fcntl.h inttypes.h limits.h signal.h sys/resource.h sys/select.h sys/stat.h sys/syscall.h sys/time.h sys/timeb.h sys/timers.h sys/times.h sys/types.h sys/utsname.h sys/wait.h termios.h time.h unistd.h utime.h windows.h winsock.h])
+AC_CHECK_HEADERS([ctype.h dirent.h errno.h fcntl.h inttypes.h limits.h signal.h sys/resource.h sys/select.h sys/stat.h sys/syscall.h sys/time.h sys/timeb.h sys/timers.h sys/times.h sys/types.h sys/utsname.h sys/wait.h termios.h time.h unistd.h utime.h windows.h winsock.h langinfo.h])
 
 # Enable large file support. Do this before testing the types ino_t, off_t, and
 # rlim_t, because it will affect the result of that test.