[Bug lto/64076] [4.9/5 Regression] ICE: in update_visibility_by_resolution_info, at ipa-visibility.c:427

[tbsaunde at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64076

tbsaunde at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tbsaunde at gcc dot gnu.org

--- Comment #2 from tbsaunde at gcc dot gnu.org ---
what happens is this:
testcase_2.o contains thunks to call s:: from vtables for each of x y and z. 
However testcase_2.C is compiled without t being a base so there's no thunk for
t.  So when testcase_1.C is compiled with t as a base of s we have an ironly
thunk for t's vtable to call s::f and prevailing regular definitions of the
other thunks. The result of that is triping the assert because we have one
externally visible ironly thunk, and others that come from testcase_2.o