Received-SPF: pass (google.com: domain of linux-kernel+bounces-28363-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Date: Wed, 17 Jan 2024 00:56:18 +0200
From: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
To: Sakari Ailus <sakari.ailus@linux.intel.com>
Cc: Jacopo Mondi <jacopo.mondi@ideasonboard.com>,
	Vinay Varma <varmavinaym@gmail.com>,
	Dave Stevenson <dave.stevenson@raspberrypi.com>,
	Mauro Carvalho Chehab <mchehab@kernel.org>,
	"open list:SONY IMX219 SENSOR DRIVER" <linux-media@vger.kernel.org>,
	open list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] media: i2c: imx219: implement the v4l2 selection api
Message-ID: <20240116225618.GA4860@pendragon.ideasonboard.com>
References: <kv6yfyahbud474e75y4jaczg64pcowvlz7i52kikknuh6wje5o@4k2hikwcueoy>
 <ZZu2C_lu6TAh-LOf@kekkonen.localdomain>
 <3q6andka2su7i43xz2ok44ejvtb3hdjdn6xretyde7sdcvtd7l@lz2syngckivi>
 <ZabUh0ozhQq-GtEC@kekkonen.localdomain>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <ZabUh0ozhQq-GtEC@kekkonen.localdomain>

Hello,

On Tue, Jan 16, 2024 at 07:09:59PM +0000, Sakari Ailus wrote:
> On Mon, Jan 08, 2024 at 10:19:35AM +0100, Jacopo Mondi wrote:
> > Hi Sakari, Vinay,
> > 
> >    a more foundamental question is how this usage of the crop/compose
> > API plays with the fact we enumerate only a limited set of frame
> > sizes, and now you can get an arbitrary output size. We could get away
> > by modifying enum_frame_sizes to return a size range (or ranges) but I
> > wonder if it wouldn't be better to introduce an internal pad to
> > represent the pixel array where to apply TGT_CROP in combination with
> > a source pad where we could apply TGT_COMPOSE and an output format.

I'm working on patches that implement an internal image pad, as part of
the work to add embedded data support. I hope to post this in the near
future.

> My earlier review wasn't focussed on the interface at all...
> 
> To depart from the current restrictions on single-subdev sensor drivers,
> this is one option.
> 
> Sensors implement various steps in different orders and different drivers
> have different capabilities, too. Mainly there are two classes: freely
> configurable drivers such cas CCS and then register list based drivers
> where virtually any dependencies between configurations are possible.
> 
> We probably can't support both classes with the same API semantics and due
> to hardware differencies. The sensor UAPI will be less than uniform it has
> been in the past but I don't think this should be an issue.
> 
> I wonder how much common understanding we have at this point on how this
> API would look like. Probably not much?

-- 
Regards,

Laurent Pinchart