[cxx-abi-dev] IA-64 C++ ABI key function question

[Martin v. Löwis]

scott douglass <sdouglass at arm.com> writes:

> Am I correct in thinking that key function was defined this way to
> enable this particular extension to C++?  Is this a common usage?

I think the rationale is the same as for the case of the pure virtual
destructor, which could also be used as a key function, but is not:
The committee codified existing practice among several compilers, and
later could not change the ABI anymore.

For the specific issue, I recommend that a quality implementation
issues a warning that a virtual function should not be defined inline,
since it usually can't be invoked inline, anyway.

Notice that even in the case of multiple copies, COMDAT sections will
eliminate duplicates in the linker.

Regards,
Martin