Hi,
On 09/06/2012 05:13 PM, Frank Schäfer wrote:
Hi,
I'm currently looking into the gspca_pac7302-driver and how it interacts
with libv4lconvert.
This is how it currently works
- driver announces v4l2_pix_format 640x480 (width x height)
- the frames (jpeg) passed to userspace are encoded as 480x640 and this
complies with the jpeg-header we generate
- libv4lconvert checks width/height in the jpeg-header and compares them
with the image format announced by the kernel:
a) values are the same:
1) V4LCONTROL_ROTATED_90_JPEG is NOT set for the device (standard
case):
=> everything is fine, image is decoded
2) V4LCONTROL_ROTATED_90_JPEG is set for the device:
=> libv4lconvert bails out with -EIO displaying the error
message "unexpected width / height in JPEG header: expected: 640x480,
header: 480x640"
b) values are different:
1) V4LCONTROL_ROTATED_90_JPEG is NOT set:
=> libv4lconvert bails out with -EIO displaying the error
message "unexpected width / height in JPEG header: expected: 640x480,
header: 480x640"
2) V4LCONTROL_ROTATED_90_JPEG is set:
=> image is decoded and rotated correctly
Thinking about this for some minutes:
1) shouldn't the kernel always announce the real image format (size) of
the data it passes to userspace ?
It is passing the real size, the data is just in a vary funky format which
needs rotation as part of its "decoding" / decompression.
Current behavior seems inconsistent to me...
Announcing the actual image size allows applications which trust the API
value more than the value in the frame header to decode the image
correctly without using libv4lconvert (although the image would still be
rotated).
That assumes that the app would know how to decompress the data which it
will not know, these cams do not generate standard JPEG data, libv4lconvert's
decompression code is the only decompression code for this fsck-ed up JPEG-s,
short of the windows drivers code.
2) shouldn't libv4lconvert always rotate the image if
V4LCONTROL_ROTATED_90_JPEG is set for a device ?
It seems like a2) is a bug, because the expected size should be 640x480,
too.
rotating by 90 degrees also swaps the width and height, which are usually
not the same, so rotating an image which starts at 640x480 will yield
a final image of 480x640 which will not be what the app expects.
3) because all pac7302 devices are sending rotated image data, we should
add them ALL to libv4lconvert. Currently only 4 of the 14 devices are on
the list.
Do you want me to send a patch ?
I see you've send a patch in the mean time, I'll reply in more detail to
this to the patch mail.
Regards,
Hans
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
