[Bug ipa/60306] [4.9 Regression] Incorrect devirtualization "pure virtual method called"

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60306

--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Created attachment 32231
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32231&action=edit
WIP patch

Hi,
I actually had code around to detect type known from function call, so it is
easy to plug it in and solve the testcase.

I am not convinced this is safe - I think we may miss the dynamic type setting
statement and then we may give wrong answer. Basically we want to know if on
_all_ or none paths we encoutered the type setting statemnt.  I do not think
the alias oracle walker can tell me if it reached top of function?

Honza