The customer reports and I've verified that instead of the expected output the attached PDF file produces random noise when read by Ghostscript. gs8.64, gs8.70, and head (r10102) all behave the same way. The command line I'm using for testing: bin/gs -sDEVICE=tiff24nc -o test.tif ./OriginalFile.pdf Note that this is an OCR'd PDF file, consisting of a page of text with an image on top. The problem appears to be with Ghostscript reading the image; if instead of TIFF you convert the attached file to PDF and then view it in Acrobat you can see the OCR'd text being drawn before being overlaid with the random noise image.
Created attachment 5413 [details] OriginalFile.pdf
Created attachment 5449 [details] a8-2i4.pdf Simplified sample file that has only JBIG2 image.
Customer reports that this is a critical issue for them.
Just an update. We appear to be decoding the symbol dictionary and text region correctly, but not the final generic region. Disabling the fourth segment gives proper results.
The problem is the file includes a refinement region with typical prediction, which we haven't implemented. Resolution of the bug just requires doing so. In the meantime, I've committed a work-around patch to skip the refinement data when we can't decode it as r10362. With this patch, rendering is not completely correct, but is much better as it removes the obscuring layer of noise.
*** Bug 690913 has been marked as a duplicate of this bug. ***
Created attachment 5705 [details] Bug690791.diff Attaching a backport patch against 8.64/8.70 which fixes the problem.
Changing customer bugs that have been resolved more than a year ago to closed.