>XalanSourceTree
I presume you're talking about the C++ version, since that class isn't used
in the Java code...
>Is there any way of getting around this and limiting the amount of memory
>required?
Xalan-J recently redesigned its internal model (see the discussion of DTM).
That reduced memory consumption a bit (not as much as we'd like; we're also
looking at an alternative implementation which is more compact but has some
performance costs) and improved our ability to load models incrementally
(so if your stylesheet is only looking at the first half of your document,
the second half never gets loaded into memory). However, DTM does
complicate our attempts to evolve toward model pruning (search for "prune"
in the archives of this mailing list), which is going to be important in
allowing us to handle larger documents in less space... so we're still
trying to decide whether this is really The Right Answer or if a better
solution exists.
None of the DTM work has yet been replicated into the C++ code; the C++
team is very wisely waiting to see how this experiment works out. Also, C++
can do some things Java can't, so the optimum solution may wind up being
very different.
>Is there any way of getting around this and limiting the amount of memory
>required?
As a workaround: You may want to consider whether it's possible to break
that killer document into a group of smaller documents, style them
separately, and then combine the results. That only works if your styling
operations are relatively local... but if they aren't, you may really need
it all in memory at once.
