[Fortran] Visit all the symbols and body in the Module and fix other bugs

[Thirumalai-Shaktivel]

No description provided.

[certik]

Fixes like these will work most of the time, until they don't -- how will this handle if there is already a variable v__t_1?

[certik]

I will review this carefully later, but this approach is good, I think that's the way to do it.

[certik]

@Shaikh-Ubaid this is an example of a pass that we might want to run in the --backend=fortran, where we need GFortran to be able to compile it. But we don't want to run it by default, since we want any ASR to be able to print without additional passes.

[Shaikh-Ubaid]

I think we also need to handle @ in functions from modules, for example:

$ cat examples/expr2.f90 
program expr2
implicit none

print *, repeat("abc", 3)

end program
$ cat pass_27_insert_deallocate.f90 
! Fortran code after applying the pass: insert_deallocate
...
program expr2
implicit none
print *, repeat@repeati32("abc", 3)

end program expr2

[certik]

Yes.

I really like this approach of the pass --- that way we are not closing any doors and we can later restrict ASR names if we decide. But right now we don't need to do that, we simply print the ASR as repeat@repeati32("abc", 3) and that's still readable, and if we want the code to compile, we call the pass from this PR, and possibly other passes as needed.

[Thirumalai-Shaktivel]

With the current changes, I get:

Repeat

program expr2
implicit none
print *, repeat("abc", 3)
end program

--show-fortran --dump-all-passes-fortran

! Fortran code after applying the pass: insert_deallocate
module lfortran_intrinsic_iso_fortran_env
implicit none
integer(4), dimension(1), parameter :: character_kinds = [1]
integer(4), parameter :: error_unit = 0
integer(4), parameter :: input_unit = 5
integer(4), parameter :: int16 = 2
integer(4), parameter :: int32 = 4
integer(4), parameter :: int64 = 8
integer(4), parameter :: int8 = 1
integer(4), dimension(2), parameter :: integer_kinds = [4, 8]
integer(4), dimension(1), parameter :: logical_kinds = [4]
integer(4), parameter :: output_unit = 6
integer(4), parameter :: real128 = -1
integer(4), parameter :: real32 = 4
integer(4), parameter :: real64 = 8
integer(4), dimension(2), parameter :: real_kinds = [4, 8]
end module lfortran_intrinsic_iso_fortran_env

module lfortran_intrinsic_builtin
implicit none
end module lfortran_intrinsic_builtin

module lfortran_intrinsic_string
use lfortran_intrinsic_iso_fortran_env, only: int64
implicit none
interface repeat
    module procedure repeati32
end interface repeat

contains

elemental function len_repeati32(n) result(r)
    integer(4), intent(in) :: n
    integer(4) :: r
    r = n
end function len_repeati32
function repeati32(s, n) result(r)
    integer(4) :: i
    integer(4) :: i1
    integer(4), intent(in) :: n
    character(len=len_repeati32(n), kind=1) :: r
    character(len=1, kind=1), intent(in) :: s
    i1 = 1
    i = 1 - 1
    do while (i + 1 <= n)
        i = i + 1
        r(i:i) = s(i1:i1)
    end do
end function repeati32
end module lfortran_intrinsic_string

program expr2
use lfortran_intrinsic_string, only: len_repeati32
use lfortran_intrinsic_string, only: repeat
use lfortran_intrinsic_string, only: repeati32
implicit none
print *, repeat("abc", 3)
end program expr2

MaxVal

program expr2
implicit none
integer :: x(3)
x = 23
end program

--show-fortran --dump-all-passes-fortran --apply-fortran-mangling

! Fortran code after applying the pass: insert_deallocate
program expr2
implicit none
integer(4) :: v__1_t
integer(4) :: v__1_v
integer(4), dimension(3) :: x
v__1_v = lbound(x, 1)
v__1_t = lbound(x, 1) - 1
do while (v__1_t + 1 <= ubound(x, 1))
    v__1_t = v__1_t + 1
    x(v__1_t) = 23
    v__1_v = v__1_v + 1
end do
end program expr2

Both outputs get compiled with gfortran and print the expected output

This PR is ready for review; if there is anything else to be implemented, do let me know.

[certik]

As long as this mangling is not applied by default, which I think it is not, then this is fine. Thanks!

[certik]

Excellent. I think it looks good in general. Go ahead and finish the PR and I give it a final review.

[Thirumalai-Shaktivel]

We can use git rebase main --committer-date-is-author-date to not change the git commit date; that's a good one.

[Thirumalai-Shaktivel]

Ready for review!

[certik]

Why is fortran removed?

[Thirumalai-Shaktivel]

When we link with the clang, gfortran doesn't generate the stack trace for the failure(It hangs), so I remove it.

[certik]

It seems this PR is implementing several unrelated fixes:

• Improvements to the Fortran backend (however no new tests are added, and old tests are removed)
• Different way of calling gfortran
• changes to the frontend's symbol table visitor
• improvements to the unique symbol pass to handle underscores

Can you split these into separate PRs? Then for each PR we'll make sure tests are added, etc. and we'll merge those that we can merge.

[certik]

Thanks for splitting it. Now let's add tests for this.

[Thirumalai-Shaktivel]

Consider the test https://github.com/lfortran/lfortran/blob/main/integration_tests/complex_06.f90, which is already enabled to test the Fortran backend. It imports the lfortran_*.f90 modules inside the program.
All the changes in this PR were introduced because of compiling the lfortran_*.f90 module. So, I think everything is already tested, do you want me to enable other specific tests as well?

cc @certik

[certik]

@Thirumalai-Shaktivel go ahead and rebase this and enable Fortran tests that now work.

[certik]

I think that this looks good to merge, thanks!