I doubted if could solve those problems by using db2latex which seems had already stoped development for a year. So if I want to make it work, learning latex is the must. Instead, sending patch to undeveloped db2latex is totally a waste of time. I must say I just perfer Java to latex.

I just made it work. However there are some kinds of problem need to solve.

bookmark:
As hyperref is in conflicts with CJK package right now, I can't generate a bookmark for PDF .
index:
The index is always incorrect. It seems that is a bug of db2latex, because its generated tex file's \section and \subsection are wrong.
images:
Images are too big. I should find a way to control them.

Today I tried two way to generate a pdf file from a our project's chinese xml file.

First FOP way, it is very easy to add chinese fonts to fop to get it work, and generate a very good one from a simple xml file. But unfortunately its ability handling large document is quite bad. I got a fatal OutofMemory error and failed to generate one. Java always does not get along quite well with Debian. Sometimes I need to do some tweak to make JAI work. Maybe FOP is not mature enough to use in practice.

For Latex way, it really has a longer learning curves. After some unsuccessful attempts, I had gotten pdflatex working with sample GB and UTF-8 files. On the other hand, when xsltproc create a .tex file, its /usepackage and /begin fields are always useless for Chinese. All I have to do is to make sure they use {CJK} package and {UTF8} along with a right font. Although I see it finally works, I just wander if there is a brilliant stylesheet, with which can get a usable .tex file for Chinese. Then I can simply use pdflatex to get a pdf out of latex.