Created attachment 10490 [details] list_of_affected_files.txt (No "Transparency" component, so I've assigned this to Images so it goes to Michael) In gdevp14.c there is a call to pdf14_update_device_color_procs_push_c. The return value to this is not checked. If we add code to make it check, a la: code = pdf14_update_device_color_procs_push_c(dev, pdf14pct->params.group_color, pdf14pct->params.icc_hash, pis, pdf14pct->params.iccprofile); if (code < 0) return code; Then the cluster reports many failures. On at least some of the cases this is because pdf14pct->params.group_color is UNKNOWN. This returns a rangecheck. This causes both differences, and it seems, SEGVs.
Created attachment 10491 [details] ev3.pdf Drastically simplified file that shows the problem.
Created attachment 10492 [details] 0001-Pass-on-an-error-code-rather-than-ignore-it.patch Patch to show the problem.
This should be fixed with http://git.ghostscript.com/?p=ghostpdl.git;a=commit;h=17f0dfaf200163d7207f27f3cc34e0a1cf843203