NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
After struggling with this problem for a week, I think I know why I'm having
trouble. I don't, however, know how to fix it, so I want to troll this list
one more time. First, the problem statement:
I can run VisAD as an applet, and it works beautifully the first time I hit
the page. If I go to another page and then return to the page with the
VisAD applet, then my applet appears to hang, no matter how simple I make
it.
Here's my theory about what's going wrong:
Under IE 5.0, using the Java plug-in 1.3, the JVM loads my applet and
visad.jar the first time they're needed. The next time around, the browser
calls Class.newInstance() (which performs a shallow copy) to generate a new
version of my applet class, while keeping the instantiations of any classes
that were loaded from the visad.jar. I know that VisAD is still loaded,
since I can call ScalarMap.getScalarName("myscalar"), for example. The
instance variables in my applet are all new, though my static variables are
still the same. Now the trouble starts. That first instance of my class is
gone for good, so the garbage collector can start picking it to pieces, and
since there are no explicit references to the information stored in VisAD's
namespace, I think that the memory VisAD has allocated starts to be picked
apart as well. The classloader doesn't bother to reload visad.jar from
scratch, since it knows that it loaded visad.jar once already, and assumes
that I've kept explicit pointers to any memory I want saved. So now I'm in
deep trouble, as I stand on instances of VisAD classes that are dissolving
beneath me.
Note: I think this IS NOT a VisAD bug (I'd call it instead a browser and/or
Java plug-in bug). There may, however, be aspects of VisAD that would
allow me to get around the problem. If, for example, I could get a pointer
to the root of the data structure that VisAD uses internally, then I could
set a static variable within my program to point to it, and everything would
be protected from the GC, and I could use the variables I created the first
time through. Or, if there were a way for me to tell VisAD to start fresh
and give me all new variables, then, while loading would be slower, at least
everything would run reliably. Alternatively, if there were a way to tell
VisAD to unload everything it has allocated, then that would do it too.
Maybe subclassing the classLoader to make it load new versions instead of
referencing old versions would work, but I'm not certain. Don't bother
suggesting that I simply make every VisAD variable that I use static, since
I've tried and that approach doesn't work (perhaps implying that my solution
#1 would fail too?). I imagine that there are other approaches, but since I
don't know how VisAD stores its internal variables I don't have any other
ideas right now.
I would really like to use VisAD in our front end, and I feel as if I'm 95%
of the way there, with only this last big problem standing in my way. Is
there anybody who can shed some light on my troubles? Please contact me
offline if you're interested and want to see my source, or if I could
perform any work which might then be folded back into somebody else's
project (as well as my own).
Ben
balexander@xxxxxxxxxx
781-994-0428
visad archives: