Thanks Steve for the confirmation. Your "always a bug" statement is of course correct, though I'd imagine it takes on a much lower priority with vendors if it is associated with invalid code. On Tue, Dec 26, 2017 at 5:08 PM, Steve Lionel < [log in to unmask]> wrote: > On 12/26/2017 7:02 PM, Neil Carlson wrote: > >> In fact it works with gfortran 8.0.0 too! (What doesn't work with >> gfortran is if subroutine SUB is in a separate source file -- you get an >> internal compiler error. The developer wants to dismiss the issue as being >> due to invalid code.) >> >> Is that code valid, or am I out to lunch on this one? >> > > The specification expression is legal for the reasons you state. But even > if it weren't, an internal compiler error is ALWAYS a compiler bug, even in > the presence of invalid code. It is true that ncells needs to be assigned a > value before the function is called for the program to be valid, but that's > beside the point. > > Steve >