[Flang] Unexpected runtime I/O error related to finalization

[DanielCChen]

Consider the following code

module m
    type base
        integer*4 :: id

        contains

        procedure :: makeArray => createBaseArray
        final :: finalizeBaseRank1
    end type

    type (base) :: b1_m

    contains

    function createBaseArray (b, n)
        class (base), intent(in) :: b
        integer*4, intent(in) :: n
        class (base), allocatable :: createBaseArray (:)

        allocate (createBaseArray(n))

        createBaseArray%id = b%id
    end function

    subroutine finalizeBaseRank1 (b)
        type(base) :: b(:)

        print *, 'finalizeBaseRank1'
    end subroutine

end module

program ftpbnd516a
use m
  print*, size (b1_m%makeArray (3))
end

Flang currently issues a runtime I/O error as:

> a.out

fatal Fortran runtime error(/home/cdchen/temp/t.f:28): Recursive I/O attempted on unit 6
 3IOT/Abort trap(coredump)

It seems the print statement in the main program got interfered with the print statement in the finalization routine.
All ifort, gfortran and XLF executed the program successfully.

[llvmbot]

@llvm/issue-subscribers-flang-runtime

Author: Daniel Chen (DanielCChen)

Consider the following code ``` module m type base integer*4 :: id

    contains

    procedure :: makeArray => createBaseArray
    final :: finalizeBaseRank1
end type

type (base) :: b1_m

contains

function createBaseArray (b, n)
    class (base), intent(in) :: b
    integer*4, intent(in) :: n
    class (base), allocatable :: createBaseArray (:)

    allocate (createBaseArray(n))

    createBaseArray%id = b%id
end function

subroutine finalizeBaseRank1 (b)
    type(base) :: b(:)

    print *, 'finalizeBaseRank1'
end subroutine

end module

program ftpbnd516a
use m
print*, size (b1_m%makeArray (3))
end

Flang currently issues a runtime I/O error as:

> a.out

fatal Fortran runtime error(/home/cdchen/temp/t.f:28): Recursive I/O attempted on unit 6
3IOT/Abort trap(coredump)

It seems the print statement in the main program got interfered with the print statement in the finalization routine.  
All ifort, gfortran and XLF executed the program successfully.
</details>

[eugeneepshteyn]

Well, the way flang emits I/O calls, this does end up recursive I/O and I'm not sure it's wrong. Shouldn't the allocatable array created as part of print statement be finalized before the print is done processing?

ifx 2025.0 outputs the following:

$ ./final-io-ifx.exe 
           3
 finalizeBaseRank1
$ ifx --version
ifx (IFX) 2025.0.0 20241008

It seems that the finalizer ran after the print was done.

gfortran on the other hand seems to ignore the I/O from the finalizer, so either finalizer didn't run or gfortran ignores recursive I/O:

$ ./final-io-gfortran.exe 
           3
$ gfortran --version
GNU Fortran (GCC) 14.1.0
...

[DanielCChen]

The standard requires to finalize function result that is finalizable before exiting the innermost scope of the executable construct, in this case, the main program.

If an executable construct references a nonpointer function, the result is finalized after execution of the innermost executable construct containing the reference.

Of course the compiler can choose what the best time is, but it shouldn't interfered with other code.

I think the reason for gfortran's behavior is that it doesn't call the finalizer at all for the finalizable function result.
ifix seems consistent with XLF.

[klausler]

It's the innermost executable construct, not the "innermost scope of the executable construct".

I suspect that we're not going to be able to fix this without some nontrivial changes to I/O data list lowering.