I'm trying to create a DNG with 32 bit floating point pixels. The reason is to create a file which is not demosaiced but contains a large dynamic range. It goes well for the major part, but it fails in the linearization step (BuildStage2Image) when I'm generating a thumbnail. Looking into the code it seems like the floating point case is not handled, so it is easy to understand why it does not work.
Is this a bug in the SDK, or does not the DNG standard support floating point pixels?
(I'm using the latest release at time of writing, version 1.4)
Edit: I patched the SDK to support floating point pixels (like 4 lines of code), so it is quite simple to do. But not surprisingly it cannot be interpreted even by Lightroom.