Re: [Qemu-devel] [RFC] Introduce module API to QEMU

[Anthony Liguori]

M. Warner Losh wrote:
> In message: <address@hidden>
>             Anthony Liguori <address@hidden> writes:
> : malc wrote:
> : > You are using _identifier_ whose name violates 7.1.3, full stop, you
> : > can not do that in _any_ context.
> : >   
> : 
> : So then we cannot use __func__ either or __attribute__ by your logic.
>
> No.  That's not true.  Those are defined by the standard or by the
> implementation.  he's objecting to your using an identifier that
> starts with __ when really, there's no reason to do that.
>   

The identifier is defined by the *implementation*.  That's the whole point.

When you use the __attribute__((section,"FOO")), it forces the variable 
to be in section FOO.  For all sections, the linker (IOW the 
*implementation*) creates two variables named __start_SECTION and 
__stop_SECTION.  In this case, it's __start_FOO and __stop_FOO.

You have to *declare* this variable yourself because the linker is not 
the frontend and the frontend doesn't know about these variables.  The 
linker is what *defines* these variables.

It doesn't matter because I've switched to __attribute__((constructor)), 
but this is not a violation of the standard.  You cannot rename these 
variables because you have no control over the definition.  They are 
there whether you like them or not.

Regards,

Anthony Liguori

> Warner
>