Space invasion: Using Levels Pick white point for all channels doesn't work
Using babl/GEGL/GIMP-2.99 updated this morning (Feb. 25, 2019), I opened a raw file using darktable. The raw file can be downloaded here:
In darktable, no processing was done other than the actual interpolation. In particular, the raw file wasn't white-balanced, so the multipliers were "1,1,1,1" and the resulting image looks blue-greenish. The reason for not white balancing the raw file was to make a sample file for testing the possibility of white balancing an image in GIMP after converting to a suitable RGB working space, in this case a working space based on the camera profile (white balancing in sRGB produces garbage in such cases, as explained here: https://ninedegreesbelow.com/photography/unbounded-srgb-color-correction-fails.html).
No matter what combination of precision conversions I tried - first for the conversion from GIMP sRGB to "CameraRGB", and then for the actual white balancing using Levels, I wasn't able to open the interpolated raw file (whether opened directly using darktable or else saved to disk as a 32f tiff or exr and then opened using darktable), convert it to the appropriate color space, and then use Levels to white balance the image. Results were identical for all the combinations (or at least appeared identical, the red channel value was very negative, on the order of -35), regardless of the various combinations of precision changes.
Here's a screenshot showing the properly white balanced image in my GIMP-CCE (bottom image), and the result in GIMP-2.99 (top image):
As an aside, the top two layers in GIMP-2.99 layer stack (on the left side of the screen shot) were to verify that white balancing by division still does work, which in fact it does. But white balancing by division is not something most users are going to even know how to do.
My apologies if I already filed a bug report on the inability to white balance post-space-invasion using Levels/Pick white point for all channels.
This is on Debian Sid, fwiw.