https://gcc.gnu.org/bugzilla/show_bug.cgi?id=16075
--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> --- This comes from the way name resolution is performed in generic instances, or rather from the fact that it is performed in generic instances at all by GNAT, which goes against the implicit resolution model of the language, where the resolution should be done only once, in the generic unit, and then propagated to its instances modulo the formal/actual substitution. As such, the problem is not reasonably fixable in GNAT; the best we can do is probably to give an ambiguity error on such a pathological code and thus refuse to compile it.
