[vtk-developers] Charts TestScatterPlotColors should be failing...

[Marcus D. Hanwell]

Hi Eric,

On Thu, Dec 30, 2010 at 10:54 AM, Eric Monson <eric.monson at duke.edu> wrote:
> Hey Marcus,
>
> I got back to some of my charts stuff and noticed that the scatter plot (vtkPlotPoints) coloring doesn't work, at least from Python wrapping, when you try to select the color array with the (string) column name. (Git master from yesterday.) I think the Python test should be failing.

I didn't realize the Python tests worked that way, I assumed as the
test still passed that my changes were good. I have spotted the issue
now - I didn't go far enough in changing to use vtkStdString...I have
a fix I will push shortly.
>
> I'm still not very familiar with running VTK tests, but it looks to me like the images to compare against for these new color-by-array tests should be in VTKData, but they're not, so when the test is run it's just creating the wrong image and then thinking it passed because it's comparing against a bad result.

That isn't very good - in the C++ case it produces the image but
fails. It will only pass when there is a valid baseline image. Let me
see if I can fix that up, I think the parallel coordinate test may
suffer from the same issue.
>
> I'll attach an image with the CTest generated image on the left, and the correct image on the right. You see the shades-of-blue points are just black on the left, which is the default behavior when it can't find the column name ("Cosine" here) passed to it.

My image looks different here, but is no longer black. Same types of
blue, but with almost transparent in the center.
>
> I'll try to look at it a bit to see if I can figure out what's wrong, but the switch you made on 2010-12-24 to vtkStdString for ColorArrayName broke something.
>
It is the string name comparison that was failing - I should have it
pushed to VTK shortly.

Marcus