The Intel oneAPI Fortran compiler (ifx (IFX) 2024.2.0 20240602 on x86_64/Linux) produces erroneous executable, where an associated procedure pointer loses its association under certain conditions. Below the MWE example extracted from the Fortuno unit testing framework, demonstrating the issue:
module test_generator
implicit none
type :: serial_case_base
character(:), allocatable :: name
end type
type, extends(serial_case_base) :: serial_case
procedure(argless_proc), pointer, nopass :: proc => null()
end type serial_case
abstract interface
subroutine argless_proc()
end subroutine argless_proc
end interface
type :: test_item
class(serial_case_base), allocatable :: item
end type
contains
function get_tests() result(testitems)
type(test_item), allocatable :: testitems(:)
testitems = [test_item(serial_case(name="proc_call", proc=test_proc_call))]
! Test, wether procedure can be called
select type (p => testitems(1)%item)
type is (serial_case)
print *, "1:START calling proc()"
call p%proc()
print *, "1:END calling proc()"
end select
end function get_tests
subroutine test_proc_call()
print *, "hello from test_proc_call"
end subroutine test_proc_call
end module test_generator
program testapp
use test_generator, only : get_tests, test_item, serial_case
implicit none
type(test_item), allocatable :: testitems(:)
testitems = [get_tests()]
! Test, wether procedure can be called
select type (p => testitems(1)%item)
type is (serial_case)
print *, "2:START calling proc()"
call p%proc()
print *, "2:END calling proc()"
end select
end program testapp
Compiling the code with ifx (without any options) and executing it results in a segfault with following output:
1:START calling proc()
hello from test_proc_call
1:END calling proc()
2:START calling proc()
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
libc.so.6 00007F1766F56710 Unknown Unknown Unknown
a.out 0000000000404A98 Unknown Unknown Unknown
libc.so.6 00007F1766F56710 Unknown Unknown Unknown
The code is standard conforming as far as I can see, and other compilers (e.g. NAG, GNU) seem to have no problems to generate a non-segfaulting executable.
A few additional observations:
- The first call test at line 33 works out as supposed. It is the test at line 59, which causes the segfault.
- Leaving away the array constructor at line 53 (using “testitems = get_tests()” instead of “testitems = [get_tests()]”) results in a working executable.
- Moving the definition of the field “name” into the extending type (serial_case) results in a correctly working executable.
- Removing the definition of the field “name” (and adapting the structure constructor in line 27 accordingly) also leads to a working executable.