>Hello, all!
>
>I am trying to write a network that will toggle the Image module's
>renderMode property from the network (rather than from the actual Image
>window "Render Options" toggle button).  So I go into Image an unhide the
>renderMode box.  When I try to connect anything to it from the network
>(rather than "hardcoding" in a value), it runs just fine but if I save it
>and then try to reopen later, it won't open in the VPE...won't load to the
>point where it displays all the modules on the screen.  Still runs okay in
>script mode.
>
>For example, I unhide the renderMode box and connect the tab to a Compute
>module that always spits out 1.  I can run this network just fine, and it
>saves okay.  But if I then go to reopen the file (after saving it), it won't
>load in the VPE.
>
>Is this expected behavior, or am I doing something wrong?
>
>Thanks!
>  Jeremy Zoss

For what it's worth, I did a quick test on 3.1.4b on SGI Irix 6.2 and did
not get this problem (3 module test net). Execute on Change with Selector
toggling renderMode works multiple times, Save, open another net, open the
test net again, no problem, Executes on Change multiple times.

Please keep up the good work on testing the Image bug. I think it's real too.

BTW, just for yucks, what color(s) did you try in your test? I have had a
client report a bug that I believe is related to image cache problems (in
SGI 6.5 openDX 4.0.5 as downloaded, not locally built) in which a
ShowBoundary object colored red [1,0,0] creates a weird pattern when you go
back in the cache (Undo from later images), but if you change to any other
color including [.99,.01,.01], the (visible) problem goes away! (The
pattern is a red smear that goes from left to right from the edge of the
ShowBoundary). I'm thinking that the pointer indexing of the objects placed
in cache is off, which might explain the growing memory leak you've
reported. If the Undo referenced the wrong starting point in the Image
Cache, it might possibly retrieve an image that would appear messed up but
not completely random. I still can't decide if this means the placing or
the retrieving is off, though I'd suspect the placing is more likely to
lead to a growing memory. It could though be some counter that is being
incorrectly incremented and claiming it's using more memory but isn't
really, then the counter value is incorrectly being used to build the
retrieval offset value, thus returning a bogus image.

Another wild-eyed theory is based on the knowledge that in the past, there
has been some funny business going on under the covers when DX sees a
vector that can be integer, even if you thought it was float. For example,
I would think that any RGB vector should be type float and left that way.
However, there are (or once were) conditions where the vector [1,0,0] may
get internally recast to integer causing untold problems downstream if fed
to a routine that demands float (and doesn't reverse the cast). As I say,
wild-eyed, but I'm just throwing out a little paranoia that might help us
uncover the real culprit. I believe this internal recasting was once seen
as an "optimization". (:-)

I got as far as some of the frame buffer routines which are "Big Hair" C
code (run-on lines with mondo bitwise comparisons, offsets, etc.; easily
candidates for the Obtuse C Awards) before getting sidetracked by more
lucrative time-wasting activities.

What concerns me overall is what fundamental changes were in progress but
not sufficiently tested from the time of the relatively stable 3.1.4b and
the release of what would eventually have become DX 4.x which is the code
base for openDX? In other words, we have to keep in mind that IBM was not
planning to pull the plug on commercial development: in fact, I know it was
not the decision of the DX group themselves to end development. So we must
assume that in the fullness of time, many of these bugs would have been
revealed by internal testing or early user testing, and quashed before the
release of an official commercial 4.x. Instead, we have to remember that we
are now both the testers and the developers. But we are seeing the code
base for 4.x, not necessarily for 3.1.4b (although clearly similar or
identical in many places). So I ask, can anyone at IBM suggest what sorts
of things might have been changed that could have created the Image
problems we're reporting that weren't problems in 3.1.4b? Thanks!

Chris Pelkie
Vice President/Scientific Visualization Producer
Conceptual Reality Presentations, Inc.
30 West Meadow Drive
Ithaca, NY 14850
[EMAIL PROTECTED]
(607) 257-8335 or (607) 254-8794

Reply via email to