Defect #8 in recent polymorphic assignment feature #polysegfaults

[RichBarton-Arm]

Testing out the new polymorphic assignment feature, I can trigger lots of segfaults in the fortran runtime. I am reporting these as separate failures even though they clearly could be related. Search for #polysegfaults .

Compiling the below code with flang (no additional options) produces a binary that gives a runtime segfault assigning to y from the output of the spread intrinsic.

program test
  implicit none
  type t
    integer i
  end type
  class(t),allocatable :: x(:)
  class(t),allocatable :: y(:,:)
  allocate(t::x(2))
  y = spread(x,dim=1,ncopies=1)
  select type(y)
  type is (t)
    print *,y(1,1)
  end select
end program

This has very similar backtrace to #831 with a very large size being malloced by the runtime.

Tested with eab3991 built with llvm9 branches of both llvm and flang-driver on AArch64