[Bug c++/79464] [7 Regression] ICE in ipa_modify_formal_ parameters

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79464

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
And, to answer myself whether the middle-end does care about the METHOD_TYPE
being correct, yes, it does, e.g. that type is stored in gimple_call_fntype
on all the calls to such inherited base ctor, so it should be better correct.

On the other side there are over 120 direct TYPE_ARG_TYPES uses and almost 30
FUNCTION_FIRST_USER_PARMTYPE.  Assuming inherited base ctor clones are only
built post-instantiation, that could cut down many of those.  Perhaps
the FUNCTION_FIRST_USER_PARMTYPE macro could transparently return the
FUNCTION_FIRST_USER_PARMTYPE for the ctor it is cloned from if used on
inheriting base ctor, but a) is that correct/desirable in all cases b) what
about the direct TYPE_ARG_TYPES uses.