https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82064
--- Comment #4 from Paul Thomas <pault at gcc dot gnu.org> --- (In reply to janus from comment #3) > It appears that the regression has been introduced by r241450, which was the > fix for PR 69834. Reverting it, in particular the changes to > resolve_select_type, makes the error go away. Strangely, my GNU Fortran (GCC) 8.0.0 20170830 gives me two "OK"s :-) That said, this is a repeat of another recent bug (sorry, I am rushing off to work - no time to find the number). I posted a work around, which translates in this case to: module mod_types type, abstract :: t end type t type, extends(t) :: t2 end type t2 class(t), allocatable :: dum1 class(t2), allocatable :: dum2 end module mod_types The permanent fix is to make sure that the vtables get produced unconditionally for module derived types. Cheers Paul