Hello JP. I was about to start refactor of the predictor's and after pulling bzr I found that the LZW patch I sent on october is not on trunk. Is there any problem with the implementation I shall fix?
In the thread http://lists.gnu.org/archive/html/pdf-devel/2008-10/msg00115.html we were discussing about the cache > 5 bytes limitation. Since the streams implementation allow to return PDF_ENOUTPUT when finish_p is true, maybe that limitation could be fixed in the LZW filter.
