On Wed, 20 Nov 2019 at 07:34, Marc-André Lureau
<marcandre.lur...@redhat.com> wrote:
> > > +static void serial_mm_instance_init(Object *o)
> > > +{
> > > + SerialMM *self = SERIAL_MM(o);
> >
> > 'self' is not idiomatic for the name of the variable containing
> > the pointer to the object in QOM code ("git grep '\Wself\W' hw"
> > shows no uses of it at all, which is quite unusual for us --
> > usually the codebase has at least a few uses of any non-standard
> > way of writing something ;-))
> >
> > Usually we use something approximating to the abbreviation
> > of the type name, so here 'smm' would do.
>
> I would prefer something more straightforward than having to come up
> with various name mangling.
>
> Imho, we loose some readability, consistency & semantic by not naming
> the object argument of the method "self"
There are two problems here:
(1) "self" gives no hint at all about whether it's the Object*,
the DeviceState*, or the SerialMM*. All of those are the
object the method is operating on, but the type differences matter.
(2) *Absolutely nobody anywhere else in any other device model
is using the 'self' argument name*. It's not a convention if
only this one file is using it, and adopting it here gives us
absolutely no consistency -- exactly the opposite.
Item (1) is part of why we do things the way we do; item (2)
is why we should not make the 16550 UART some weird
exception from the way we do things.
thanks
-- PMM
