V4L capture support could use some improvements (Patch #336)

Hi,

I'll review this patch and install opensolaris (homefully this week) to check these and my changes.

I do agree this is quite unstable code and I'm currently planning to rewrite the whole library; (want to check with Vadim if it's ok).

I'll apply your patch and check if it's ok on the linux boxes I have here.
Will post changes to SVN as soon as I finish the tests.

Trisk,

Sorry for the delay on this.
Do these changes compile in OpenSolaris? (applying the patch in revision 3057)
Could you send me the version and installed packages details of your OpenSolaris setup?

Yes, this builds and runs fine in OpenSolaris. However, it depends on important changes necessary for OpenCV itself to be portable, and for compatibility with Studio C++, see bugs #329, #333, #334, and #335. An additional patch is also necessary for Sun Studio if you use the Intel Integrated Performance Primitives).

These patches were made for the SFEopencv package recipe. If you build that (see http://sourceforge.net/apps/mediawiki/pkgbuild/index.php?title=Pkgbuild_on_OpenSolaris ), it should work out of the box.

I'm using OpenSolaris build 134 and Sun Studio 12u1 provided by pkg://opensolaris.org/developer/[email protected],5.11-0.111:20100306T002245Z. The headers should even work in builds prior to 77 (where sys/videodev2.h was missing an #include).

Trisk,

Will run tests on Linux boxes (at least SuSE + Ubuntu) with your patch on the latest trunk version.

Will let you know as soon as I finish them (24~24h)

Hi Trisk,

Could you please try applying the patch I uploaded on the latest SVN version and checkout if everything goes ok on your OpenSolaris box?

https://code.ros.org/trac/opencv/attachment/ticket/336/cap_v4l.cpp.trisk.patch

On line 2346, under linux, I get an invalid argument with some webcams (v4l1 devices?) - didn't pursue, just a nuisance; works fine.

When I resize (changing width/height after getting a number of frames), I can no longer work with the device - that problem is corrected in cap_libv4l.cpp (it's a workaround, not a solution).

Other than these, all your corrections/improvements seem to work perfectly under Ubuntu 10.04 LTS and openSUSE 11.2 (forcing usage of cap_v4l.cpp to hook with v4l).

I would still like to do further tests (under a lot of professional and academic workload).

Again, I plan to rewrite this the cap_libv4l.cpp library, hopefully starting July 12th 2010; It'll take opensolaris into account.

I plan for it to be able to deal with all of our current issues, make code more efficient and readable, gain stability/robustness and allow changes of most properties in runtime, at least width, height, frame rate and any properties that return "busy device"(EBUSY) while streaming.

Please let me know how your tests go.