Known problem in variable handling

[Joseph Kesselman/CAM/Lotus <jo...@us.ibm.com>]

Just so folks know: Yes, there seems to be a confirmed problem with
variables -- specifically, the interaction between deferred evaluation of
variables and the new shared-DTM variable stack -- in the current release
of Xalan. The most common symptom will be that the content of a variable is
damaged after a call to another template. I'm still working on the problem,
since it's a moderately convoluted one... but I believe this accounts for
most of the recent variable-related bug reports.

(Summary of the problem: When the variable stack is popped on exit from a
template, it is supposed to discard the Result Tree Fragments for all
variables created within that template. The problem is that it's currently
discarding all RTFs created _during_ that template -- which may include an
RTF created by late resolution of a variable's value. The challenge is
finding a fix that neither requires giving up the use of shared DTMs nor
hamtrings the attempt to release and resuse memory.)