bug: passing array section in subroutine

[gxyd]

Minimal Reproducible Example

For the below fortran program extracted from the POT3D repository:

module global_dims
      implicit none
      integer, parameter :: nt_g=5
      integer :: ntm1_g=4
      integer, parameter :: np_g=7
      integer :: npm1_g=6
end module

program POT3D
      use global_dims
      implicit none
      call set_flux
end program pot3d

subroutine set_flux
      use global_dims
      implicit none
      real, dimension(:,:), allocatable :: br0_g
      allocate (br0_g(nt_g,np_g))
      call readbr (br0_g)
end subroutine

subroutine readbr (br0_g)
      use global_dims
      implicit none
      real, dimension(nt_g,np_g) :: br0_g
      real :: sum0
      call intrp2d (nt_g-2,np_g-2, br0_g(2:ntm1_g,2:npm1_g))
      sum0=sum(br0_g(2,2:npm1_g))
      print *, "sum0: ", sum0
end subroutine

subroutine intrp2d (nx,ny,f)
      implicit none
      integer :: nx,ny
      real, dimension(nx,ny) :: f
      f = 2.43214264512062073e-01
      return
end subroutine

the output printed by LFortran is:

> lfortran --version
LFortran version: 0.45.0-322-g5631279ff
Platform: macOS ARM
LLVM: 19.1.7
Default target: arm64-apple-darwin23.6.0

> lfortran test.f90
sum0:     7.29642808e-01

> gfortran test.f90 -o test.o && ./test.o
 sum0:    1.21607137

I checked also for ifx compiler as well, the output produced is the same as that for GFortran compiler: https://godbolt.org/z/bEhPYnjch

[gxyd]

If we modify the snippet above (i.e. #6324 (comment)) by using the below diff:

diff --git a/test.f90 b/test.f90
index f71fa6c00..9459cff97 100644
--- a/test.f90
+++ b/test.f90
@@ -25,7 +25,11 @@ subroutine readbr (br0_g)
       implicit none
       real, dimension(nt_g,np_g) :: br0_g
       real :: sum0
-      call intrp2d (nt_g-2,np_g-2, br0_g(2:ntm1_g,2:npm1_g))
+      real, dimension(:, :), allocatable :: f
+      allocate(f(ntm1_g - 1, npm1_g-1))
+      f = br0_g(2:ntm1_g, 2:npm1_g)
+      call intrp2d (nt_g-2,np_g-2, f)
+      br0_g(2:ntm1_g, 2:npm1_g) = f
       sum0=sum(br0_g(2,2:npm1_g))
       print *, "sum0: ", sum0
 end subroutine

we get a correct print with LFortran.

[YBansal95]

I'd like to work on this issue. Could you please elaborate a bit about what needs to be done?

[certik]

Isn't this the same issue as #6214?

[gxyd]

Isn't this the same issue as #6214?

Yes, I think so. I hadn't seen that issue until I opened this one, we can keep this issue open until we close that issue and then we can test this one also to see if the MRE reported here also works or not.