http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49638
--- Comment #9 from janus at gcc dot gnu.org 2011-08-04 07:41:43 UTC --- Hi Mikael, > 4.5.7.3 (type-bound procedure overriding) has: > • Either both shall be subroutines or both shall be functions having the same > result characteristics (12.3.3). > > 12.3.3 (Characteristics of function results): > If a type parameter of a function result or a bound of a function result array > is not a constant expression, the > exact dependence on the entities in the expression is a characteristic > > > So the standards asks for same-expression-ness. > Whether a compiler should diagnose it is another question. thanks for digging out these references. That makes it pretty clear. > I understand reversed operands (for commutative operators) as having an equal > "exact dependence on the entities". And same for associative operators. > Thus it is bound to be very complicated in the general case. Yes, these are the kinds of things that I have also been worrying about. They will complicate matters a bit, but I think one can handle it. > About the functions to reuse there is also gfc_dep_compare_expr and its > sub-functions (they should handle functions, variables and arithmetic > operators). Yes, Tobias already suggested this to me, and indeed it looks very much like what we need here. > I don't know however how you will have two corresponding dummy > arguments (from different procedures) compare equal. Well, the patch in comment #7 handles this by just comparing the names of the arguments (which have to be the same in overridden procedures, so I this this will be enough): + case EXPR_VARIABLE: + if (strcmp (e1->symtree->n.sym->name, e1->symtree->n.sym->name) != 0) + return FAILURE; gfc_dep_compare_expr relies on a function called 'gfc_are_identical_variables'. This really checks for equal symbols, which is too strict for our case here (so we may add an extra argument to loosen this restriction?). Thanks for the feedback ...
