https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120483
--- Comment #5 from anlauf at gcc dot gnu.org --- Instead of adding the SAVE attribute to the declaration, one can get the same wrong code with -fno-automatic. The decl generated by gfc_sym_type() looks suspicious in the case when the SAVE attribute is present. Reverting that particular hunk of commit r15-2131 fixes the present testcase for me, but I haven't investigated further.