The configure.ac script detects libintl but can depend on libiconv.

I added this to force it and to demonstrate, but I think this is not 100% correct and should be massaged into the libintl tests.

--- configure.ac.old    2013-02-16 09:34:55.000000000 +0000
+++ configure.ac        2013-02-16 09:43:22.000000000 +0000
@@ -2122,6 +2122,15 @@
                                                # pthread (first!) on Linux
 fi
 
+# Check iconv
+AC_CHECK_LIB([iconv], [iconv_open], , [ac_found_iconf=no])
+if test "x$ac_found_iconf" = "xno"; then
+  AC_CHECK_LIB([iconv], [libiconv_open], , [ac_found_iconf=no])
+fi
+if test "x$ac_found_iconf" = "xyes"; then
+  LIBS="-liconv $LIBS"
+fi
+
 # check if we need libintl for locale functions
 AC_CHECK_LIB(intl, textdomain,
        [AC_DEFINE(WITH_LIBINTL, 1,

Do you have a use case where this is needed?  As far as I know, Python itself doesn't reference libiconv directly.  libintl may but we're not building it, merely using it.  If libintl does reference it, the linker in use should try to satisfy the dependency for it.  We shouldn't have to specify the transitive closure of all dependent libraries.

Hi Yes, when Python is built as static only, on a system when no dynamic loading is available. You have to explicitly link libiconv when libintl references it.

That helps a bit but, to be able to even test this, we would still need a specific example of a platform and a configuration where this is the case.  It's not necessary or desirable to add on most platforms, I think.

Sure this is on an Atari m68k platform called FreeMiNT.

Traditionally in configure scripts you'll see it checked as *-mint*

Alan, I was about to file a bug as well because I'm not even getting a libintl dependency to be acknowledged under these build conditions.  What versions of gettext and GCC are you using?

Ned, not sure if it matters, but the libintl page does specify that -liconv must also be linked as it is a dependency:

http://www.gnu.org/software/gettext/FAQ.html#integrating_undefined

I would suggest a more straightforward patch.  My understanding is that the standard library on the m68k-atari-mint platform does not include the necessary gettext functionality.  This functionality is present in glibc, however.  A more direct test might be to see if gettext is available in the standard library.  If so, proceed.  If not, attempt to link in libintl and its dependency:

--- a/configure.ac	Sat Feb 23 18:52:51 2013 +0100
+++ b/configure.ac	Sun Feb 24 17:59:41 2013 -0500
@@ -2180,6 +2180,14 @@
 	[Define to 1 if libintl is needed for locale functions.])
         LIBS="-lintl $LIBS"])
 
+# search for gettext in either libc or libintl
+AC_CHECK_FUNC(gettext, [],
+	[AC_CHECK_LIB(intl, gettext, 
+                  [LIBS="$LIBS -lintl -liconv"],
+                  [AC_MSG_ERROR([unable to find the gettext() function])],
+                  [-liconv])
+    ])
+

Bumping. How does Jeffrey's patch look folks ?

This works for me...

--- configure.ac.old    2013-09-10 14:37:20.000000000 +0000
+++ configure.ac        2013-09-10 14:56:27.000000000 +0000
@@ -2190,7 +2190,11 @@
 AC_CHECK_LIB(intl, textdomain,
        [AC_DEFINE(WITH_LIBINTL, 1,
        [Define to 1 if libintl is needed for locale functions.])
-        LIBS="-lintl $LIBS"])
+        LIBS="-lintl $LIBS"],
+       [AC_SEARCH_LIBS(textdomain, intl,
+               [AC_DEFINE(WITH_LIBINTL, 1,
+               [Define to 1 if libintl is needed for locale functions.])
+               LIBS="-lintl -liconv $LIBS"], , "-liconv")])
 
 # checks for system dependent C++ extensions support
 case "$ac_sys_system" in