Looking at the code more closely:

• opencv/modules/videoio/src/cap_v4l.cpp

  Lines 1781 to 1784 in f87e1ef

  	case V4L2_PIX_FMT_SBGGR8:
  	bayer2rgb24(imageSize.width, imageSize.height,
  	start, (unsigned char*)frame.imageData);
  	return;

• opencv/modules/videoio/src/cap_v4l.cpp

  Lines 1344 to 1350 in f87e1ef

  	/* B */
  	if ( (i > WIDTH) && ((i % WIDTH) > 0) ) {
  	*scanpt++ = (*(rawpt-WIDTH-1)+*(rawpt-WIDTH+1)+
  	*(rawpt+WIDTH-1)+*(rawpt+WIDTH+1))/4; /* R */
  	*scanpt++ = (*(rawpt-1)+*(rawpt+1)+
  	*(rawpt+WIDTH)+*(rawpt-WIDTH))/4; /* G */
  	*scanpt++ = *rawpt; /* B */

These lines of code do BGGR --> RGB. You can see it in the case of even i and j indices: B output value is directly the value of *scanpt++ = *rawpt;:

But with V4L2_PIX_FMT_SGBRG8:

• opencv/modules/videoio/src/cap_v4l.cpp

  Lines 1795 to 1798 in f87e1ef

  	case V4L2_PIX_FMT_SGBRG8:
  	sgbrg2rgb24(imageSize.width, imageSize.height,
  	start, (unsigned char*)frame.imageData);
  	return;

• opencv/modules/videoio/src/cap_v4l.cpp

  Lines 1421 to 1429 in f87e1ef

  	if ( (i/WIDTH) % 2 == 0 ) //even row
  	{
  	if ( (i % 2) == 0 ) //even pixel
  	{
  	if ( (i > WIDTH) && ((i % WIDTH) > 0) )
  	{
  	*scanpt++ = (*(rawpt-1)+*(rawpt+1))/2; /* R */
  	*scanpt++ = *(rawpt); /* G */
  	*scanpt++ = (*(rawpt-WIDTH) + *(rawpt+WIDTH))/2; /* B */

Here we have instead GBRG --> BGR:

With even i and j, the output B component is the average from left and right pixels.

To sum-up:

• V4L2_PIX_FMT_SBGGR8 and V4L2_PIX_FMT_SN9C10X do BGGR --> RGB
• while V4L2_PIX_FMT_SGBRG8 do GBRG --> BGR
• and V4L2_PIX_FMT_SGRBG8 do GRBG --> BGR

@catree go ahead with other cases.

@asmorkalov

Now we have the following cases which use cvtColor() function:

• V4L2_PIX_FMT_SN9C10X --> BGR (breaking change, before it was to RGB)
• V4L2_PIX_FMT_SRGGB8 (added) --> BGR
• V4L2_PIX_FMT_SBGGR8, V4L2_PIX_FMT_SGBRG8 and V4L2_PIX_FMT_SGRBG8 --> BGR

I'm wondering whether these modes can be added/enabled to our virtual V4L device test:

I have updated the test file.

Could you please also add something like this to the test:

diff --git a/modules/videoio/test/test_v4l2.cpp b/modules/videoio/test/test_v4l2.cpp
index 12d2a7210e..5bb1a1f5a3 100644
--- a/modules/videoio/test/test_v4l2.cpp
+++ b/modules/videoio/test/test_v4l2.cpp
@@ -70,6 +70,7 @@ TEST_P(videoio_v4l2, formats)
     const string device = devs[0];
     const Size sz(640, 480);
     const Format_Channels_Depth params = GetParam();
+    const Size esz(sz.width * params.mul_width, sz.height * params.mul_height);
 
     {
         // Case with RAW output
@@ -83,7 +84,17 @@ TEST_P(videoio_v4l2, formats)
             Mat img;
             EXPECT_TRUE(cap.grab());
             EXPECT_TRUE(cap.retrieve(img));
-            EXPECT_EQ(Size(sz.width * params.mul_width, sz.height * params.mul_height), img.size());
+            if (params.pixel_format == V4L2_PIX_FMT_SRGGB8 ||
+                params.pixel_format == V4L2_PIX_FMT_SBGGR8 ||
+                params.pixel_format == V4L2_PIX_FMT_SGBRG8 ||
+                params.pixel_format == V4L2_PIX_FMT_SGRBG8)
+            {
+                EXPECT_EQ((size_t)esz.area(), img.total());
+            }
+            else
+            {
+                EXPECT_EQ(esz, img.size());
+            }
             EXPECT_EQ(params.channels, img.channels());
             EXPECT_EQ(params.depth, img.depth());
         }

Currently V4L2 returns a single line in Bayer formats.