0.20.2.

There is now more info on darktable's side: https://github.com/darktable-org/rawspeed/pull/250#issuecomment-1077723895

Thank you all for the answers. I read the binary with native C++ fstream (800x800 = 640000 values). then used __builtin_bswap16() to swap and obtain my bayer data which I was able to convert to RGB,
Many thanks

For byte swapping use something like ntohs() (actually this converts from big endian to your machine/host endianness, so it might be a noop if your machine is big endian as well) or, depending on your system/compiler, some incarnation of the bswap16() intrinsic/bultin function (e.g. __builtin_bswap16() for GCC, _byteswap_ushort() for MSVC, etc.)

Also, if your file is 2 bytes per pixel, uncompressed, you do not need LibRaw to read source data values. Your values are already here (in unsigned short format, probably you may need big/little-endian swap; I know nothing about your images and your camera)

Thank you Alex for your answer. I tried with open_bayer as I think the raw file has no metadata. With the code bellow I could see that readb as 1280000 values instead of 640000 for an image 800x8000. The final result with dcraw_process is not satisfactory. I wonder If and how I could retrieve the unpacked values and do the conversion myself.

FILE *in = fopen(fname, "rb");
	fseek(in, 0, SEEK_END);
	unsigned fsz = ftell(in);
	unsigned char *buffer = (unsigned char *)malloc(fsz);
	if (!buffer)
		return 2;
	fseek(in, 0, SEEK_SET);
	unsigned readb = fread(buffer, 1, fsz, in);
	if (readb != fsz)
		return 3;
	std::cout << "readb" << readb << std::endl;
	LibRaw rp;
	rp.imgdata.params.output_tiff = 1;
	int ret = rp.open_bayer(buffer, fsz, 800, 800, 0, 0, 0, 0, 0,
							LIBRAW_OPENBAYER_BGGR, 0, 0, 0);
	if (ret != LIBRAW_SUCCESS)
		return 4;
	if ((ret = rp.unpack()) != LIBRAW_SUCCESS)
		printf("Unpack error: %d\n", ret);
 
	if ((ret = rp.dcraw_process()) != LIBRAW_SUCCESS)
		printf("Processing error: %d\n", ret);
 
	char outfn[256];
	sprintf(outfn, "%s.tif", fname);
	if (LIBRAW_SUCCESS != (ret = rp.dcraw_ppm_tiff_writer(outfn)))
		printf("Cannot write %s: %s\n", outfn, libraw_strerror(ret));
	else
		printf("Created %s\n", outfn);
 
	ret = rp.dcraw_process();
	// Check for error using LIBRAW_SUCCESS. I never get an error here
 
	ret = rp.dcraw_ppm_tiff_writer(fname); 

Microsoft RAW extension is produced by Microsoft, we're not involved in this process in any way (Microsoft uses the public version of LibRaw).
So, we're completely not responsible for MS RAW Extension.

You may try to use our FastRawViewer instead.

So you're talking about Microsoft RAW Extension, right?

What software (that uses LibRaw for RAW support) do you use?

There is separate project on github: https://github.com/LibRaw/LibRaw-cmake

Note: we (LibRaw team) is not responsible for it and do not provide any support for it.

Sony ILCE-1 is fully supported in 202110 snapshot:
Supported camera list: https://www.libraw.org/supported-cameras-libraw-202110
Announce page: https://www.libraw.org/news/libraw-202110-snapshot

OK, understood.
DNG file metatata:
| | 1) ImageWidth = 3040
| | 2) ImageHeight = 2014
| | 21) DefaultCropOrigin = 15 7 (15/1 7/1)
| | 22) DefaultCropSize = 3008 2000 (3008/1 2000/1)
| | 27) ActiveArea = 0 0 2014 3040

By default, LibRaw extracts ActiveArea image (3040x2014)

Suppose we are discussing the latest public version (https://github.com/LibRaw/LibRaw, master branch):
DefaultCrop tags are parsed into imgdata.sizes.raw_inset_crops[0]
This crop can be applied via: LibRaw::adjust_to_raw_inset_crop(1)

BUT:
DefaultCrop top/left in this specific file is odd and will be rounded to nearest even margin, so image will be cropped to 3007x1999 starting at 16,8

There is no way to avoid this in current LibRaw (this change was implemented in LibRaw 0.20)

You may switch to LibRaw pre-0.20 (0.19 or so) with LIBRAW_PROCESSING_USE_DNG_DEFAULT_CROP flag set to params.raw_processing_options