Bug: Expressions in DIMENSION attribute of arrays inside a SUBROUTINE lead to an ASR verify pass error

[kmr-srbh]

Minimum Reproducible Example (MRE)

m.f90

MODULE m
   IMPLICIT NONE
   INTEGER :: nx = 2
END MODULE m

example.f90

MODULE example
   USE m
   IMPLICIT NONE
CONTAINS
   SUBROUTINE sub (n)

      INTEGER, DIMENSION(nx-1) :: n
      
   END SUBROUTINE sub
END MODULE example

PROGRAM test
   use example
END PROGRAM test

Output

(base) saurabh-kumar@Awadh:~/Projects/System/lfortran$ lfortran ./examples/example.f90
ASR verify pass error: ASR::ttype_t in ASR::FunctionType cannot be tied to a scope.
 --> ./examples/example.f90:5:4 - 9:21
  |
5 |       SUBROUTINE sub (n)
  |       ^^^^^^^^^^^^^^^^^^...
...
  |
9 |       END SUBROUTINE sub
  | ...^^^^^^^^^^^^^^^^^^^^^ failed here


Note: Please report unclear, confusing or incorrect messages as bugs at
https://github.com/lfortran/lfortran/issues.

The bug was caught when trying to compile file expxs.f90 of the SNAP package.

The subroutine expxs_slgg ( scat, map, cs ) causes the above error due to the parameter cs which is declared as REAL(r_knd), DIMENSION(cmom-1,nx), INTENT(OUT) :: cs. On further analysis, we find that the expression cmom-1 is the cause of the issue. Other similar expressions like cmom+1, cmom * 2, etc also lead to the above error. We can simplify the issue to the MRE provided above.

[assem2002]

Do you think it's due to the dimension-shape not being a constant at compile time?

[kmr-srbh]

Do you think it's due to the dimension-shape not being a constant at compile time?

Thank you @assem2002 , I had missed the module part of the issue. I have updated the code. The previous code had failed when the dimension was not an expression.

[assem2002]

It's still the same problem though. If you figure out any fix inform me, I will look up this issue too and I'll keep updated.

[kmr-srbh]

It's still the same problem though. If you figure out any fix inform me, I will look up this issue too and I'll keep updated.

The code compiles and executes properly with GFortran. It compiles with LFortran when we replace nx-1 with nx. Also, the issue does not occur outside of a subroutine. Could you please explain why you think this could be the reason?

[assem2002]

it got confusing when found out it works with nx but not with nx - 1. I'll dig into it and let you know.

[kmr-srbh]

it got confusing when found out it works with nx but not with nx - 1. I'll dig into it and let you know.

Yes! The previous code I had provided was incorrect. The issue here has a module in the picture too. The above is the correct code.

[certik]

I think a variation of this bug was fixed by @Pranavchiku in #3962. Maybe the fix does not work if the module is compiled separately?

[kmr-srbh]

I think a variation of this bug was fixed by @Pranavchiku in #3962. Maybe the fix does not work if the module is compiled separately?

I tried the MRE provided in #3962. From what I have observed, setting array dimension works with module variables, but not when we convert it to an expression. So, doing real, dimension(nx), intent(in) :: cs works, but real, dimension(nx-1), intent(in) :: cs fails.

(lf) saurabh-kumar@Awadh:~/Projects/System/lfortran$ lfortran ./examples/example.f90
ASR verify pass error: ASR::ttype_t in ASR::FunctionType cannot be tied to a scope.
 --> ./examples/example.f90:5:1 - 9:14
  |
5 |    subroutine a(cs)
  |    ^^^^^^^^^^^^^^^^...
...
  |
9 |    end subroutine
  | ...^^^^^^^^^^^^^^ failed here


Note: Please report unclear, confusing or incorrect messages as bugs at
https://github.com/lfortran/lfortran/issues.

Given that we already have a fix for it's base case in #3962, I think this should be easy to fix. I am trying to fix this.

[kmr-srbh]

@certik got this. This is an issue with handling binary integer operations inside the dimension attribute. I am pushing a fix in a while.

(base) saurabh-kumar@Awadh:~/Projects/System/lfortran$ lfortran ./examples/example.f90
integer binop
ASR verify pass error: ASR::ttype_t in ASR::FunctionType cannot be tied to a scope.
 --> ./examples/example.f90:5:1 - 9:14
  |
5 |    subroutine a(cs)
  |    ^^^^^^^^^^^^^^^^...
...
  |
9 |    end subroutine
  | ...^^^^^^^^^^^^^^ failed here


Note: Please report unclear, confusing or incorrect messages as bugs at
https://github.com/lfortran/lfortran/issues.

[Pranavchiku]

Yes, handling binary operation will do the job. Just transfer it to something like “INTEGER, DIMENSION(get_nx()-1) :: n”

[Pranavchiku]

An approach to fix this is to create a BaseExprReplacer that visits passed expression with special handling for Var_t,IntegerBinOp_t, etc. Apply same logic on visit_Var and then we can expand it.

[kmr-srbh]

An approach to fix this is to create a BaseExprReplacer that visits passed expression with special handling for Var_t,IntegerBinOp_t, etc. Apply same logic on visit_Var and then we can expand it.

Let me have a look into it! 👍

[Pranavchiku]

Let me have a look into it! 👍

i’ll ask you to get current PR to finish line.