>the XSLT need to smart enough to only keep a small portion of DOM >tree in memory, instead of keeping the whole tree in memory.
See the DTM documentation for instructions on how to turn on incremental parsing. That will partly address your concern, by loading only as much of the source document as is needed. See the SQL extension, which does try to load only as much as is needed into Xalan's storage. And see the archives of this list for a discussion of "pruning", which is the other half of this puzzle. I don't know of any existing XSLT processor which has an effective pruning algorithm. We definitely want to do this, and in fact I have an initial draft of some support code for it which is unfortunately not in condition where it can be checked in (too far out of synch with CVS). We're not ignoring the issue. "We're making mistakes as quickly as we can." And remember that this is an open-source project; if you've got a bright idea on how to address your needs, please feel free to contribute code suggestions.
