Re: [PATCH v3 04/10] drm: Convert connector_helper_funcs->atomic_check to accept drm_atomic_state

[Ville Syrjälä]

On Thu, May 16, 2019 at 02:07:34PM +0200, Daniel Vetter wrote:
> On Thu, May 16, 2019 at 2:02 PM Laurent Pinchart
> <laurent.pinchart@xxxxxxxxxxxxxxxx> wrote:
> >
> > Hi Daniel,
> >
> > On Mon, May 13, 2019 at 04:47:47PM +0200, Daniel Vetter wrote:
> > > On Sat, May 11, 2019 at 10:12:02PM +0300, Laurent Pinchart wrote:
> > > > On Thu, May 02, 2019 at 03:49:46PM -0400, Sean Paul wrote:
> > > >> From: Sean Paul <seanpaul@xxxxxxxxxxxx>
> > > >>
> > > >> Everyone who implements connector_helper_funcs->atomic_check reaches
> > > >> into the connector state to get the atomic state. Instead of continuing
> > > >> this pattern, change the callback signature to just give atomic state
> > > >> and let the driver determine what it does and does not need from it.
> > > >>
> > > >> Eventually all atomic functions should do this, but that's just too much
> > > >> busy work for me.
> > > >
> > > > Given that drivers also access the connector state, isn't this slightly
> > > > more inefficient ?
> > >
> > > It's atomic code, we're trying to optimize for clean code at the expense
> > > of a bit of runtime overhead due to more pointer chasing. And I agree with
> > > the general push, the pile of old/new_state pointers of various objects
> > > we're passing around is confusing. Passing the overall drm_atomic_state
> > > seems much more reasonable, and with that we can get everything else. Plus
> > > it's much more obvious whether you have the old/new state (since that's
> > > explicit when you look it up from the drm_atomic_state).
> >
> > Yes, I agree it's cleaner. I just hope the atomic state tracking cost
> > can be kept under control :-)
> >
> > By the way, this is likely not going to happen as it would be way too
> > intrusive, but it would be nice to rename drm_atomic_state to
> > drm_atomic_transaction (or something similar). It doesn't model a state,
> > but a change between an old state to a new state. This confused me in
> > the past, and I'm sure it can still be confusing to newcomers.
> 
> Why are you the first to suggest this, this is awesome!

Can't quite tell if that's irony or not. Anyways, this has been
suggested before but no volunteers stepped forward.

> drm_atomic_state is indeed not a state, but a transaction representing
> how we go from the old to the new state.

On a semi-related topic, I've occasionally pondered about moving
mode_changed & co. from the obj states to the top level
state/transaction (maybe stored as a bitmask). But that would
definitely not be a trivial sed job.

-- 
Ville Syrjälä
Intel