https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110548

--- Comment #1 from Andrew Benson <abensonca at gcc dot gnu.org> ---
This still fails with 16.2.1 20261115 and trunk (17.0.0 20261115). Here is a
smaller test case (no finalizer is needed):

module taskerMod
   type :: helper
   end type helper
   type :: tasker
      integer :: depth=-1
    contains
      procedure :: compute => taskerCompute
   end type tasker
 contains
   recursive subroutine taskerCompute(self,helper_)
     class(tasker), intent(inout)           :: self
     class(helper), intent(inout), optional :: helper_
     !$omp atomic
     self%depth=self%depth+1
     write (*,*) "DEPTH ",self%depth
     if (self%depth < 3) then
        !$omp task untied
        call self%compute(helper_)
        !$omp end task
     end if
   end subroutine taskerCompute
 end module taskerMod
 program testTasks
   use :: taskerMod
   type(tasker) :: tasker_
   tasker_=tasker(0)
   !$omp parallel
   !$omp single
   !$omp taskgroup
   !$omp task untied
   call tasker_%compute()
   !$omp end task
   !$omp end taskgroup
   !$omp end single
   !$omp end parallel
   write (*,*) "COMPLETED"
 end program testTasks

$ gfortran -g -fopenmp pr110548.f90 && OMP_NUM_THREADS=4 ./a.out
 DEPTH            1

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0  0x150e60e3fb2f in ???
#1  0x40154f in __taskermod_MOD_taskercompute._omp_cpyfn.1
        at pr110548.f90:17
#2  0x150e6114cf9f in GOMP_task
        at libgomp/task.c:644
#3  0x40130f in __taskermod_MOD_taskercompute
        at pr110548.f90:17

The crash is specific to an absent *polymorphic* optional argument. helper_ is
a dummy argument in an orphaned task, so it is implicitly firstprivate. With
16.2.1 and 17.0:

  helper_ declared as   implicit firstprivate   firstprivate(helper_)
  type(helper)          works                   works
  class(helper)         segfault                segfault

So gfortran already handles an absent optional argument in FIRSTPRIVATE (the
copy is simply not present), and handles a present polymorphic argument, but
crashes when copying an absent polymorphic one: the segfault is in the task's
copy function (_omp_cpyfn.1), which presumably does not check for presence
before copying the class object.

shared(helper_) on the inner task avoids the crash.

Reply via email to