Re: [PATCH, PR 60640] When creating virtual clones, clone thunks too

[Jason Merrill <jason at redhat dot com>]

On 04/04/2014 04:05 PM, Jan Hubicka wrote:
> Ah, yes, it is what I was discussing this with Jason, but apparently the
> discussion died out.  According to his comment __cxa_pure_virtual is a
> synonym for undefined behaviour so it may be correct to unconditionally
> devirtualize here (changing runtime behaviour from terminate() to random method
> call), but at the moment we don't do that.  Somewhere back in my head I have
> burned in from C++ lessons that calling pure virtual method should result in
> terminate () call, so I am considering __cxa_pure_virtual to be legitimate call
> target (as opposed to any non-function or __builtin_unreachable that I drop
> from the list).
>
> I would be happy to change this behaviour, since there are quite few extra
> cases where we could devirtualize well then.

It's definitely undefined.

10.4/6:

Member functions can be called from a constructor (or destructor) of an 
abstract class; the effect of making a virtual call (10.3) to a pure 
virtual function directly or indirectly for the object being created (or 
destroyed) from such a constructor (or destructor) is undefined.

Jason