https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108450
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> --- Interestingly, it is resolved twice. First for: (gdb) p gfc_debug_expr(e) maxloc[((p:a) (mask = p:l))] via resolve_all_program_units → gfc_resolve → gfc_resolve → resolve_codes → gfc_resolve_code → gfc_resolve_blocks → gfc_resolve_code → gfc_resolve_expr And then via: ... → resolve_omp_atomic → gfc_resolve_expr The problem is that in the latter case, the 'e' is now: maxloc[((p:a(FULL)) (mask = p:l) (.false.))] and the unprefixed .false. seems to cause the problems. The intrinsic uses: RESULT = MAXLOC(ARRAY, DIM [, MASK] [,KIND] [,BACK]) RESULT = MAXLOC(ARRAY [, MASK] [,KIND] [,BACK]) During the first call to check_specific, maxloc[((p:a(FULL)) (mask = p:l))] is turned into maxloc[((p:a(FULL)) ((arg not-present)) (mask = p:l) ((arg not-present)) ((arg not-present)))] via 4774 if (!sort_actual (specific->name, ap, specific->formal, &expr->where)) The null-args are later removed via 'remove_nullargs (ap);' * * * Thus, it looks as if resolving the RHS expression twice is wrong.
