[bug-pcl] PCL5e with bitmap images causes segmentation fault in GhostPCL

[Thomas Urban]

Hi,


after analysing for several days I think there's a bug in GhostPCL incl. 
1.41.

I have prepared a page at

http://www.toxa.de/ghostpcl/

to provide you with backtraces from gdb as well as straces and debug 
output from pcl6 binary itself.


Symptoms:

On a SuSE-linux based server with CUPS installed there is a backend 
which tries to convert enqueued PCL jobs to PS for analysis. These jobs 
are created on Win32 system using HP PCL5e printer drivers.

As long as there is only text or vector-based illustrations and tables 
in it a job gets converted without any problem. As soon as the document 
contains at least one bitmap image the call to pcl6 binary for 
conversion fails on SIGSEGV.

We tested this with several documents. First we've been working with 
1.38, then upgraded to a freshly downloaded 1.41, but the pcl6 was still 
crashing.


Today I recompiled the source tree (downloaded freshly again) on a 
different host using SuSE 9.0. Then I tried to convert the job given as 
an example to you on that web page. Again pcl6 crashed at the same point.



The web page includes:

1) the job used to test on the second host. other job files are 
available as well, but I think this single file is okay for testing, 
isn't it?

2) two logs of gdb session observing the crash with backtrace and memory 
dumps of the section the device handle is located in.

3) command line, output and strace used on a debuggable version of pcl6 
compiled in debug mode as given in docs

4) finally there are two files with detailed information on used gcc and 
libraries linked by pcl6



Well all these files come from the server we initially tried to analyse 
the crash on.


I hope, these information can help you to help me out of this situation.



Thanks in advance and
with Best Regards,


Thomas Urban

-- 

Walter Friedländer-Straße 2
D-10249 Berlin

  Telefon: +49 (30) 4280 8731
  Telefax: +49 (30) 4679 3243

Internet: www.toxa.de