obtained gcc-4.6.4.tar.bz2 from upstream website;upstream

verified gcc-4.6.4.tar.bz2.sig;
imported gcc-4.6.4 source tree from verified upstream tarball.

downloading a git-generated archive based on the 'upstream' tag
should provide you with a source tree that is binary identical
to the one extracted from the above tarball.

if you have obtained the source via the command 'git clone',
however, do note that line-endings of files in your working
directory might differ from line-endings of the respective
files in the upstream repository.

1 files changed, 53 insertions, 0 deletions

diff --git a/gcc/testsuite/gfortran.dg/elemental_subroutine_3.f90 b/gcc/testsuite/gfortran.dg/elemental_subroutine_3.f90
new file mode 100644
index 000000000..1f93cd4a2
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/elemental_subroutine_3.f90
@@ -0,0 +1,53 @@
+! { dg-do run }
+! Test the fix for PR25746, in which dependency checking was not being
+! done for elemental subroutines and therefore for interface assignments.
+!
+! This test is based on
+! http://home.comcast.net/~kmbtib/Fortran_stuff/elem_assign.f90
+! as reported by Harald Anlauf <anlauf@gmx.de> in the PR.
+! 
+module elem_assign
+   implicit none
+   type mytype
+      integer x
+   end type mytype
+   interface assignment(=)
+      module procedure myassign
+   end interface assignment(=)
+   contains
+      elemental subroutine myassign(x,y)
+         type(mytype), intent(out) :: x
+         type(mytype), intent(in) :: y
+! Multiply the components by 2 to verify that this is being called.
+         x%x = y%x*2
+      end subroutine myassign
+end module elem_assign
+
+program test
+   use elem_assign
+   implicit none
+   type(mytype) :: y(6), x(6) = (/mytype(1),mytype(20),mytype(300),&
+                                  mytype(4000),mytype(50000),&
+                                  mytype(1000000)/)
+   type(mytype) :: z(2, 3)
+! The original case - dependency between lhs and rhs. 
+   x = x((/2,3,1,4,5,6/))
+   if (any(x%x .ne. (/40, 600, 2, 8000, 100000, 2000000/))) call abort ()
+! Slightly more elborate case with non-trivial array ref on lhs.
+   x(4:1:-1) = x((/1,3,2,4/))
+   if (any(x%x .ne. (/16000, 1200, 4, 80, 100000, 2000000/))) call abort ()
+! Check that no-dependence case works....
+   y = x
+   if (any(y%x .ne. (/32000, 2400, 8, 160, 200000, 4000000/))) call abort ()
+! ...and now a case that caused headaches during the preparation of the patch
+   x(2:5) = x(1:4)
+   if (any(x%x .ne. (/16000, 32000, 2400, 8, 160, 2000000/))) call abort ()
+! Check offsets are done correctly in multi-dimensional cases
+   z = reshape (x, (/2,3/))
+   z(:, 3:2:-1) = z(:, 1:2)
+   y = reshape (z, (/6/))
+   if (any(y%x .ne. (/ 64000, 128000, 19200, 64, 128000, 256000/))) call abort ()
+end program test
+
+! { dg-final { cleanup-modules "elem_assign" } }
+