This thread has been interesting reading.
I've been wondering whether I should be using DNG myself, but there is a sort of fundamental problem about DNG that has been bothering me. Maybe somebody here knows the answer.
The "meaning" of the numeric pixel values in a RAW file (i.e. the correspondence between the captured photons and the numeric values) are relative to the engineering properties of the particular sensor (and its supporting electronics). These engineering properties are not only different for each different camera, they are regarded as proprietary by (I think all) camera manufacturers. Anybody that makes any kind of a RAW converter has know a lot about these sensor properties for each camera. (If you are the camera manufacturer, you presumably know everything you need to know and can put that knowledge into either the in-camera firmware RAW converter or your own proprietary software RAW converter. Everybody else has to do some kind of reverse engineering in order to write a RAW converter.)
Now, lets look at the image data in a DNG file (lets say one where the RAW data is not embedded). I can think of two possibilities:
A) The numbers in the image data pixels are copied exactly from the raw file.
B) The numbers in the image data pixels have undergone a conversion to some kind of a universal grey-scale space.
If A), then DNG files still need to undergo RAW conversion in a camera-specific way, and are thus not really "universal". If B), then the DNG files are no longer camera-relative, but ARE relative to the quality of the knowledge that the convert-to-DNG software had about the camera (and, some might argue, are not really "RAW files" anymore).
Anybody know whether its "A" or "B" (or something else)?