[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: PickData Method on IDLgrWindow Object
You don't mention if things worked differntly in IDL 5.3.
A detailed description of pickdata from RSI follows.
A while back, I was having a lot of difficulty understanding pickdata
and select, since pickdata appeared to ignore the object (model) that
you pass it. I didn't save the full thread of discussion with RSI but
the key concept that was throwing me (quoted from the RSI response) was:
"It is true that PickData DOES NOT check to see if the x,y is within the
specified object. But that is precisely what Select is for."
This might be your problem, as I was getting what appeared to be
incorrect values from pickdata as well. If you want to see strange
things, overlay a surface above an image plane and then look
at pickdata results when the surface is a mesh vs solid. There was no
easy way to know with pickdata if you saw through the surface to the
image plane or hit a grid line on the mesh surface.
Hope this helps.
Wed, 2 Feb 2000 14:57:55 -0700
RSI E-mail Support <firstname.lastname@example.org>
Using Select first and then Pickdata is NOT a kluge. It is that way by
design, which means both IDL and the external tools which IDL uses to
graphics (such as OpenGL). The external tools have their own
which I am sure you can understand.
The Select operation (which means the OpenGL Select function call) is
particularly expensive, since it performs an "invisible" rendering of an
entire scene, creating a list of objects that are in the selection box.
The IDL Object Graphics designers chose, wisely, to separate the Select
Pickdata operations, so that users would not be forced to pay the
the Select operation, when it is not really needed. The case where there
only one object in a model would never require a Select operation; in
case, you are ready to use Pickdata. It turns out to be fairly common
complex scenes to have several models, where some of the models contain
object. For this reason alone, it is important that the Pickdata method
perform a Select operation "under-the-hood."
I have included, below, some comments from one of our Objects Graphics
designers. I think you will find them very informative.
Technical Support Engineer
Research Systems Incorporated
Please include my name on the Subject line of email replies. Thanks!
---------- Comments from one of our Object Graphics developers:
First, let me explain briefly how Select and PickData work.
Select() performs an "invisible" rendering of the entire scene,
hit testing within the selection box for each object in the scene.
that are "drawn" into the selection box are returned in the object list.
PickData() simply reads the Z-buffer at the specified x,y to see if
at all got drawn at that location. If something was drawn there, the
following happens: PickData obtains the transform associated with the
object you specified in the call to PickData. It does not matter what
Select did the last time you called Select. (Select and PickData are
completely unrelated) It also does not matter where the x,y is.
just obtains that object's user->screen transform and computes the
inverse-transpose of that transform. It then multiplies the specified
by this inverse transform to obtain the corresponding data space
of the specified x,y. All it does is multiply the specified x,y by an
object's inverse transform if SOMETHING was drawn at that x,y. If x,y
on some other object, the returned user space point will be *in terms
*relative to* the object specified in the call to PickData. For example
suppose you have a sphere centered at (0,0,0) with radius=2 and a sphere
located at (5,5,5) with radius=2. If you call PickData with an x,y
in the middle of the second sphere with the first sphere used as the
reference, you will get a returned user data coordinate of (5,5,?),
of the expected (0,0,?) because you used the first sphere's transform.
though the pick was made beyond the bounds of the first sphere.
We specify a "graphic" object, instead of a model, in PickData, because
graphic object in a given model can have its own unique transform, which
be different from the model's transform. Why? Each graphic object has
additional transform associated with it called the COORD_CONV transform,
specified by the XCOORD_CONV, YCOORD_CONV, and ZCOORD_CONV properties.
these properties are not specified, and left as their default values,
the graphic object's transform WILL be the same as the transform of the
model that contains that graphic object. But, in general, they are not
same. So, the suggestion of passing a model object in PickData instead
graphic object does not work out well. Besides, the problem at hand
still exist if we used a model as the reference for the inverse
apply because there may be more than one model rendered at a given x,y.
There are several approaches to addressing the problem of associating
correct model with PickData.
1) Invoke the Select method every time before calling PickData. Use one
the objects returned by Select as the reference object for PickData. If
there are multiple objects returned by Select, you need to deal with
depending on the application. If there are no objects, there is no hit
no reason to call PickData. If you do call PickData, the result will be
terms of the object returned by Select.
If you want to call PickData again, for another x,y, then you really
call Select again first, to make sure that you get the right object
associated with the new x,y. For example, if you are tracking the mouse
cursor and updating a "read-out" with object data values, then you
need to call Select and PickData for every mouse movement event. This
seem expensive, and it might be for very complex models. But for most
cases, Select is fast enough to remain interactive.
It is true that PickData DOES NOT check to see if the x,y is within the
specified object. But that is precisely what Select is for. Therefore,
is not a kludge to call Select first; it is by design. The only thing
PickData knows about the supplied object is its transform - PickData has
idea where the object actually is on the screen.
Perhaps the "Pick" part of the PickData method was poorly named and is
partially responsible for the confusion. PickData does not perform a
as the PHIGS picking functions do. Perhaps you are thinking that
is a full-fledged PHIGS pick - it is not. It just performs a vertex
multiply by a certain inverse transform, if the screen was dirty there.
So, why not just have PickData perform a "Select" inside of PickData???
Probably because the designers of this function thought that it was
to give the user the choice and control over performing these two steps
always doing them both. The obvious example is the case of a single
- you don't need to call Select before PickData.
I think that this approach is the best, but here are some others.
2) Create (I don't mean draw one) a bounding box for each object. Call
PickData all you want without calling Select and see which bounding box
data coordinate returned by PickData is inside of. This only works well
where the bounding box encloses the object with little or no extra space
where objects don't interfere with each other. You would also have to
careful about using COORD_CONV and which object you use for the
object in PickData.
3) Create a "mask" of the object you want PickData to be sensitive to in
IDLgrBuffer by rendering just the one object into the buffer. If there
non-background color in the buffer at the specified x,y, then you have a
and can call PickData to get the values.
--------------------- end of comments ------------------------
Richard Tyc wrote:
> Has anyone noticed any differences with the Pickdata method in the IDL 5.3.1
> I have a fairly complex window with one main model (oModel) which has
> several other Models added to it each of which having many objects (such as
> Polylines, texture-mapped polygons etc.) and each having a different
> Using the following statement on the overall model (allowing the user to
> pick anything on screen) :
> pick = sState.oWindow->PickData(sState.oView,$
> sState.oModel, $
> The data point, dataxyz, never seems to look right. Even when I select a
> point on a IDLgrAxis object (which belongs directly to oModel), I never get
> the expected data point (ie. picking on the x axis should return a point
> with the range [ 0.0-1.0, 0 , 0 ] but there always seems to be some
> component of y
> and z ?? It does however return correctly if I have selected the line (axis)
> or hit the background.
> Anyone know whats going on or have similar experience ?