Re: [PATCH v2 2/3] v4l2-ctl: Support V4L2_CTRL_TYPE_RECT

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Tue, Nov 05, 2024 at 10:01:32AM +0100, Hans Verkuil wrote:
> On 05/11/2024 09:51, Laurent Pinchart wrote:
> > On Tue, Nov 05, 2024 at 09:30:43AM +0100, Hans Verkuil wrote:
> >> On 04/11/2024 02:24, Ming Qian(OSS) wrote:
> >>> On 2024/10/31 18:09, Laurent Pinchart wrote:
> >>>> On Thu, Oct 31, 2024 at 05:46:49PM +0800, Ming Qian(OSS) wrote:
> >>>>> On 2024/10/31 17:34, Laurent Pinchart wrote:
> >>>>>> On Thu, Oct 31, 2024 at 05:19:02PM +0800, Ming Qian(OSS) wrote:
> >>>>>>> On 2024/10/30 17:19, Hans Verkuil wrote:
> >>>>>>>> On 30/10/2024 10:03, Laurent Pinchart wrote:
> >>>>>>>>> On Wed, Oct 30, 2024 at 11:43:06AM +0900, ming.qian@oss.nxp.com wrote:
> >>>>>>>>>> From: Yunke Cao <yunkec@google.com>
> >>>>>>>>>>
> >>>>>>>>>> Tested with VIVID
> >>>>>>>>>>
> >>>>>>>>>>     ./v4l2-ctl -C rect -d 0
> >>>>>>>>>> rect: 300x400@200x100
> >>>>>>>>>>
> >>>>>>>>>>     ./v4l2-ctl -c rect=1000x2000@0x0
> >>>>>>>>>>     ./v4l2-ctl -C rect -d 0
> >>>>>>>>>> rect: 1000x2000@0x0
> >>>>>>>>>>
> >>>>>>>>>> Signed-off-by: Yunke Cao <yunkec@google.com>
> >>>>>>>>>> Signed-off-by: Ming Qian <ming.qian@oss.nxp.com>
> >>>>>>>>>> ---
> >>>>>>>>>>     utils/v4l2-ctl/v4l2-ctl-common.cpp | 12 ++++++++++++
> >>>>>>>>>>     1 file changed, 12 insertions(+)
> >>>>>>>>>>
> >>>>>>>>>> diff --git a/utils/v4l2-ctl/v4l2-ctl-common.cpp b/utils/v4l2-ctl/v4l2-ctl-common.cpp
> >>>>>>>>>> index 40667575fcc7..538e1951cf81 100644
> >>>>>>>>>> --- a/utils/v4l2-ctl/v4l2-ctl-common.cpp
> >>>>>>>>>> +++ b/utils/v4l2-ctl/v4l2-ctl-common.cpp
> >>>>>>>>>> @@ -614,6 +614,10 @@ static void print_value(int fd, const v4l2_query_ext_ctrl &qc, const v4l2_ext_co
> >>>>>>>>>>     		case V4L2_CTRL_TYPE_AREA:
> >>>>>>>>>>     			printf("%dx%d", ctrl.p_area->width, ctrl.p_area->height);
> >>>>>>>>>>     			break;
> >>>>>>>>>> +		case V4L2_CTRL_TYPE_RECT:
> >>>>>>>>>> +			printf("%ux%u@%dx%d", ctrl.p_rect->width, ctrl.p_rect->height,
> >>>>>>>>>
> >>>>>>>>> I find this notation ambiguous, it's not immediately clear when reading
> >>>>>>>>> 10x10@20x20 if we're looking at a 10x10 rectangle positioned at (20,20)
> >>>>>>>>> or the other way around. media-ctl use (20,20)/10x10 which I think would
> >>>>>>>>> be a better notation.
> >>>>>>>>
> >>>>>>>> Good point, I agree.
> >>>>>>>>
> >>>>>>>> Ming Qian, can you also update patch 1/4 of the kernel patch series to
> >>>>>>>> use the same formatting when logging the V4L2_CTRL_TYPE_RECT value?
> >>>>>>>>
> >>>>>>>> Regards,
> >>>>>>>>
> >>>>>>>> 	Hans
> >>>>>>>
> >>>>>>> There is a issue in v4l2-utils, that ',' is the ending flag in
> >>>>>>> v4l_getsubopt(), then I can't set the rect control,
> >>>>>>> for example:
> >>>>>>>
> >>>>>>> $v4l2-ctl -d 0 -c rect="(0,0)/1000x2000"
> >>>>>>> control '0)/1000x2000' without '='
> >>>>>>
> >>>>>> The should be fixable in v4l_getsubopt().
> >>>>>>
> >>>>>
> >>>>> I can see the following comments of v4l_getsubopt(),
> >>>>>
> >>>>>      Parse comma separated suboption from *OPTIONP and match against
> >>>>>      strings in TOKENS.
> >>>>>
> >>>>> I am not sure if we can change it.
> >>>>
> >>>> I think we can improve quotes handling by considering quoted substrings
> >>>> as a single value, ignoring commas. Hans any opinion ?
> >>
> >> I think commas are hard to parse. Note that v4l_getsubopt is normally a
> >> #define for getsubopt from glibc. So you can't change the behavior of
> >> that function.
> > 
> > Can't we ? Isn't it an internal function ?
> 
> No. It's there when it is compiled for systems without glibc.
> 
> But I guess we can just use our copy all the time.
> 
> >> I propose this format for parsing instead:
> >>
> >> widthxheight@(top;left)
> >>
> >> e.g.: 1000x2000@(0;0)
> >>
> >> According to this:
> >> https://www.dr-aart.nl/Geometry-coordinates.html
> >>
> >> the ';' is the separator in countries where a decimal comma is used
> >> instead of a decimal point.
> >>
> >> I prefer to have the position after the size of the rectangle, for two
> >> reasons: it feels more natural to talk about a 'rectangle of size S at position
> >> P', and it also makes it possible to allow a variant where only the size
> >> is given and the position will default to (0;0). I.e., we can support
> >> parsing either "widthxheight" or "widthxheight@(top;left)".
> >>
> >> However, logging rectangles in the kernel should use a comma instead of a
> >> semicolon. Inside v4l-utils just consistently use the semicolon.
> >>
> >> What do you think, Laurent?
> > 
> > We have a precedent of using (x,y)/WxH , both in the kernel and in
> > media-ctl. Breaking that with another syntax would cause trouble,
> > especially having different syntaxes between media-ctl and v4l2-ctl.
> > Think about the shell scripts that would need to convert from one syntax
> > to another for instance. I would very strongly like to avoid that.
> 
> Have we used that notation in the kernel? Where?

In kernel log message. We have quite a few occurences of "(%d,%d)/%ux%u"
(or the less correct "(%d,%d)/%dx%d"). That's not an ABI, but it's still
used.

For what it's worth, we're also using the same notation in libcamera
(not necessarily as a result of a careful design decision, but likely
more because it was already used by media-ctl and nobody thought twice).

> I will admit that I am not a fan of the media-ctl notation, I think it is
> weird. I think WxH@(x,y) is much more natural. But if v4l_getsubopt can be
> adapted for this, then I'm fine with it.

I don't have anything against WxH@(x,y) per-se, but having different
notations in different tools is in my opinion a big enough annoyance
that a new notation shouldn't be introduced without very compeling
reasons.

> >>> How about omitting the commas between the brackets when parsing subopt?
> >>>
> >>>
> >>>>>>>>>> +			       ctrl.p_rect->left, ctrl.p_rect->top);
> >>>>>>>>>> +			break;
> >>>>>>>>>>     		default:
> >>>>>>>>>>     			printf("unsupported payload type");
> >>>>>>>>>>     			break;
> >>>>>>>>>> @@ -702,6 +706,9 @@ static void print_qctrl(int fd, const v4l2_query_ext_ctrl &qc,
> >>>>>>>>>>     	case V4L2_CTRL_TYPE_AREA:
> >>>>>>>>>>     		printf("%31s %#8.8x (area)   :", s.c_str(), qc.id);
> >>>>>>>>>>     		break;
> >>>>>>>>>> +	case V4L2_CTRL_TYPE_RECT:
> >>>>>>>>>> +		printf("%31s %#8.8x (rect)   :", s.c_str(), qc.id);
> >>>>>>>>>> +		break;
> >>>>>>>>>>     	case V4L2_CTRL_TYPE_HDR10_CLL_INFO:
> >>>>>>>>>>     		printf("%31s %#8.8x (hdr10-cll-info):", s.c_str(), qc.id);
> >>>>>>>>>>     		break;
> >>>>>>>>>> @@ -1279,6 +1286,11 @@ void common_set(cv4l_fd &_fd)
> >>>>>>>>>>     					sscanf(set_ctrl.second.c_str(), "%ux%u",
> >>>>>>>>>>     					       &ctrl.p_area->width, &ctrl.p_area->height);
> >>>>>>>>>>     					break;
> >>>>>>>>>> +				case V4L2_CTRL_TYPE_RECT:
> >>>>>>>>>> +					sscanf(set_ctrl.second.c_str(), "%ux%u@%dx%d",
> >>>>>>>>>> +					       &ctrl.p_rect->width, &ctrl.p_rect->height,
> >>>>>>>>>> +					       &ctrl.p_rect->left, &ctrl.p_rect->top);
> >>>>>>>>>> +					break;
> >>>>>>>>>>     				default:
> >>>>>>>>>>     					fprintf(stderr, "%s: unsupported payload type\n",
> >>>>>>>>>>     							qc.name);

-- 
Regards,

Laurent Pinchart