[libavg-devel] v4l branch code review

[OXullo Intersecans]

Hithere!
a little delay, sorry!

Il giorno 24/lug/07, alle ore 21:56, Ulrich von Zadow ha scritto:

> First, I fixed a compile error (I think it depends on the kernel  
> version
> - I have 2.6.21 here) in linux/videodev2.h. The fix is comitted on the
> branch. Please check if that doesn't break on your computer.

I had to comment the two includes linux/types.h and linux/time.h  
(V4LCamera.cpp:36-37) due to a double inclusion error.
Do you think that this would break yours? (please check against the  
latest commit)

> Then, the camera only supports V4L2_PIX_FMT_YUYV (not
> V4L2_PIX_FMT_UYUV). I haven't fixed that yet (I will), but I can get a
> picture with broken colors by just ignoring the PIX_FMT difference, so
> in principle everything else works.

[snip]
> I just fixed this on the branch by adding a new pixel format. I have a
> working USB camera here now!

that's perfect!
I had a chance to test a Logitech QuickCam PRO 5000, which inits  
correctly (YUV422 (ok YUYV)) but displays a garbled colourspace.
This cam is known to give a bunch of problem with v4l, therefore I  
don't consider this an appropriate test.
What camera did you get working?

> The next thing is that the current camera node uses the width and  
> height
> attributes as the camera image width and height for V4L. However,  
> these
> are also used as the size of the image on the screen :-(. We should
> really change that and use new attribute names for the camera  
> resolution.

good.
I added attributes capturewidth and captureheight to avg <video> node  
DTD and switched to these two instead of width/height (CameraNode.cpp: 
86-87).
I wish that this won't break up nothing, conceptually. I tried to use  
the same notation format of the attribute identifier.

> Also, while everything works here using V4L2_FIELD_INTERLACED, can you
> check if V4L2_FIELD_ANY works as well? That should work for more
> cameras, right?

Yes, this way the 'correct' format is selected contextually.
FIELD_ANY works well with bt878

> After that, I noticed that I didn't really need to buy a camera at  
> all;
> there is a test device you can enable in the kernel under Device
> Drivers->Multimedia devices->Video Capture Adapters->Virtual Video
> Driver. The test device also only supports V4L2_PIX_FMT_YUYV.

That's a good point. I had no time to test it, having a stock kernel.
The only thing that a real camera can give you in addition is an  
overview of the specific limit of the hardware.
Bt878 based boards are very admissive, they support almost all pixel  
formats, resolutions, field sequences..

>
> Still, I think we're very close to being done :-).

gooood
let me know!

--
OXullo Intersecans

0 2 L > Outside Standing Level
http://www.02L.net