This is a heads up. I've been looking at the failing CMM jtreg tests, and I think I have a solution.
The problems are: 1. Some incorrect "golden" images: these have errors, presumably caused by bugs in whatever CMM was used to create them. 2. In the case of PYCC, some of the entries in the golden image correspond to colours that are impossible. I don't mean just out of gamut, I mean colours that correspond to no possible stimulus. This leads to somewhat unpredictable results with any CMS, since these non-colours are not representable in the profile connection space. I've created a GoldenImageFactory class that generates golden images for testing by using high accuracy floating-point arithmetic. This means that the tests are independent of the CMM being used. With regard to the PYCC tests, I think it doesn't make any sense for large parts of the test image to contain colours that could never occur in any PYCC file, so I have adjusted the tests only to contain possible colours. I've also noticed that the PYCC profile I provided causes clipping in highlights when converting PYCC images made by the profile in the proprietary JDK 6. I've solved this by creating a new PYCC profile with a lower L value for PYCC's 100% white and a soft knee in the gamma curve, as recommended in Kodak PCD 042, retrieved from http://www.kodak.com/global/en/professional/products/storage/pcd/techInfo/pcd-042.jhtml. This gives good visual results for conversions. A related problem with the PYCC profile I provided is that it was limited to the sRGB colour space. This new profile uses Lab as its profile connection space, and is well-defined over that whole space. I'll submit a webrev when I'm done. Andrew.
