On 10/01/2012 10:51 AM, Mathieu Malaterre wrote: > On Mon, Oct 1, 2012 at 10:40 AM, Roland Stigge <[email protected]> wrote: >> On 10/01/2012 01:03 AM, Asheesh Laroia wrote: >>> rofl.tif is all black, 888x458 pixels. (Roland, is that what you were >>> expecting?) >> >> I'm not sure what image contents Mathieu intended, but will need to >> check the backtrace. We somehow need to support / work around / catch >> the signed data case, maybe even just prevent it by not supporting it. > > Here is what Gimp says: > > ... > Warning: > The image you are loading has 16 bits per channel. GIMP can only > handle 8 bit, so it will be converted for you. Information will be > lost because of this conversion. > ... > > I do not have a signed 8bits dataset at hand to reproduce this issue, > so we have to deal with this 16 bits one. > Most viewer on debian are unfortunately truncating 16bits to 8bits for > display. > > The only way I know is: > > $ gdcm2vtk rofl.tif rofl.tif.dcm > $ gdcmviewer rofl.tif.dcm > > And you can even check with the original j2k file: > > $ gdcmimg bugreport.j2k bugreport.j2k.dcm > $ gdcmviewer bugreport.j2k.dcm > > Of course if you have access to a 16bits display program, feel free to use it. > >> We probably need to switch from Jasper to GeoJasper (better supported >> fork) at some point, but for wheezy this is too late. > > There is no difference in between Geojasper and Jasper (maybe some > minor code optimizations, and support for geoxml). > A better switch would be from Jasper to OpenJPEG.
Right. Probably the way to go. Unfortunately, too late for a library switch for the reverse dependencies for wheezy. :-( Looking at the error, it's not a segfault, whouldn't lead to a security issue. It's "just" unsupported functionality, even if it doesn't error out gracefully. So does it really justify RC status? Don't get me wrong: I don't want to hide issues, and we probably need to replace the abort() case with sth. more useable. Just don't have a fast solution for today. Thanks, Roland -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
