You forgot to explain what you think is not quite right.
gfortran searches only directories where it may have
installed some modules and directories that you explicitly
add via -I.

(In reply to comment #1)
> You forgot to explain what you think is not quite right.
> gfortran searches only directories where it may have
> installed some modules and directories that you explicitly
> add via -I.
Because when you do cross compilation you don't want your compiler to pick up host modules. Apart from that ^^ shouldn't frontends behave in a consistent manner?

yeah, as Kacper said, cross compilers should not be searching native paths.  the gcc and g++ frontends are working correctly (they use the sysrooted path) while the gfortran frontend is broken.

-v output shows:
Target: armv7a-unknown-linux-gnueabi
Configured with: /var/tmp/portage/cross-armv7a-unknown-linux-gnueabi/gcc-4.6.3/work/gcc-4.6.3/configure --prefix=/usr --bindir=/usr/x86_64-pc-linux-gnu/armv7a-unknown-linux-gnueabi/gcc-bin/4.6.3 --includedir=/usr/lib/gcc/armv7a-unknown-linux-gnueabi/4.6.3/include --datadir=/usr/share/gcc-data/armv7a-unknown-linux-gnueabi/4.6.3 --mandir=/usr/share/gcc-data/armv7a-unknown-linux-gnueabi/4.6.3/man --infodir=/usr/share/gcc-data/armv7a-unknown-linux-gnueabi/4.6.3/info --with-gxx-include-dir=/usr/lib/gcc/armv7a-unknown-linux-gnueabi/4.6.3/include/g++-v4 --host=x86_64-pc-linux-gnu --target=armv7a-unknown-linux-gnueabi --build=x86_64-pc-linux-gnu --disable-altivec --disable-fixed-point --without-ppl --without-cloog --enable-lto --enable-nls --without-included-gettext --with-system-zlib --enable-obsolete --disable-werror --enable-secureplt --disable-multilib --disable-libmudflap --disable-libssp --enable-libgomp --with-python-dir=/share/gcc-data/armv7a-unknown-linux-gnueabi/4.6.3/python --enable-poison-system-directories --enable-checking=release --disable-libgcj --with-arch=armv7-a --with-float=hard --with-fpu=vfpv3-d16 --enable-languages=c,c++,fortran --with-sysroot=/usr/armv7a-unknown-linux-gnueabi --disable-bootstrap --enable-__cxa_atexit --enable-clocale=gnu --with-bugurl=http://bugs.gentoo.org/ --with-pkgversion='Gentoo 4.6.3 p1.6, pie-0.5.2'

Untested draft patch. As I don't build cross compilers, I would be happy if someone could confirm that it works.

--- a/gcc/fortran/cpp.c
+++ b/gcc/fortran/cpp.c
@@ -269,3 +269,3 @@ gfc_cpp_init_options (unsigned int decoded_options_count,
   gfc_cpp_option.prefix = NULL;
-  gfc_cpp_option.sysroot = NULL;
+  gfc_cpp_option.sysroot = TARGET_SYSTEM_ROOT;

I forgot half of the patch - the second half is needed when no system root exists. The patch is still untested.

--- a/gcc/fortran/cpp.c
+++ b/gcc/fortran/cpp.c
@@ -40,2 +40,6 @@ along with GCC; see the file COPYING3.  If not see

+#ifndef TARGET_SYSTEM_ROOT
+# define TARGET_SYSTEM_ROOT NULL
+#endif
+
 #ifndef TARGET_CPU_CPP_BUILTINS
@@ -269,3 +273,3 @@ gfc_cpp_init_options (unsigned int decoded_options_count,
   gfc_cpp_option.prefix = NULL;
-  gfc_cpp_option.sysroot = NULL;
+  gfc_cpp_option.sysroot = TARGET_SYSTEM_ROOT;

(In reply to comment #5)

that's half the equation.  the other half is the build system support.

in gcc/Makefile.in, they do:
  CFLAGS-c-family/c-opts.o += @TARGET_SYSTEM_ROOT_DEFINE@

but you can't do that in gcc/fortran/Make-lang.in.  you can't use $(TARGET_SYSTEM_ROOT_DEFINE) either because the variable is only AC_SUBST().

so you could add this line to gcc/Makefile.in:
  CFLAGS-fortran/cpp.o += @TARGET_SYSTEM_ROOT_DEFINE@
and it'd work.  this seems to be how they propagate the define into the CFLAGS-c-family/c-opts.o variable, so maybe this is OK.

alternatively, gcc/Makefile.in does setup $(COMPILER_DEFINES), so adding this to gcc/fortran/Make-lang.in works:
  CFLAGS-fortran/cpp.o += $(DRIVER_DEFINES)
but i'm sure that's the wrong answer.

the other alternative is to declare a "static const char *sysroot" inside of gcc/fortran/gfortranspec.c and then initialize gfc_cpp_option.sysroot via that.  this works because gfortranspec.c is compiled already with $(COMPILER_DEFINES).

i'm wait out of my depth, so i don't know what the "right" answer is :).

Can you check whether the following patch works for you?

http://gcc.gnu.org/ml/gcc-patches/2012-10/msg01780.html

(In reply to comment #7)

that patch doesn't work as there is a typo in Make-lang.in.  it needs to be:
CFLAGS-fortran/cpp.o += $(TARGET_SYSTEM_ROOT_DEFINE)

then it works OK

Author: burnus
Date: Sun Oct 21 10:51:26 2012
New Revision: 192650

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192650
Log:
gcc/
2012-10-21  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54725
        * Makefile.in (TARGET_SYSTEM_ROOT_DEFINE): New.

gcc/fortran
2012-10-21  Tobias Burnus  <burnus@net-b.de>

        PR fortran/54725
        * Make-lang.in (CFLAGS-cpp.o): Use TARGET_SYSTEM_ROOT_DEFINE.
        * cpp.o (gfc_cpp_init_options): Use it for
        setting gfc_cpp_option.sysroot.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/Makefile.in
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/Make-lang.in
    trunk/gcc/fortran/cpp.c

FIXED on the trunk (4.8).

Thanks for the bug report, the debugging and for testing the patch!