PreprocessVertexTriangles 2D Unstructured Grid

[CharlesGueunet]

Dear TTK Team,

I am actually developing a TTK plugin and I have encountered a strange behavior.
My plugin needs to recover the triangles attached to a vertex of the mesh, so I call the preprocessVertexTriangles().

This function bug in the case of a 2D VTK Unstructured Grid (Segmentation Fault).

Steps to reproduce

1. Create a new plugin with the createTTKmodule.sh
2. Add triangulation_->preprocessVertexTriangles() in the setupTriangulation() line 117
3. Compile your plugin
4. Open a 2D VTU (surface) in paraview and apply your plugin

You should have a Seg Fault here.

Expected result

As far as I understand, it seems that this function should not be used in 2D. We should use preprocessVertexStar() instead.
But in the case of my plugin, I try to retrieve triangles for both 2D and 3D mesh.
Would it be possible to have preprocessVertexTriangles() to do the same pre-processing than preprocessVertexStar() in 2D, to have only one function call for all data sets ?
Thank you for your help

Charles

[julien-tierny]

hi Charles, thanks for the issue.

This function bug in the case of a 2D VTK Unstructured Grid (Segmentation Fault).

that's a problem indeed.

As far as I understand, it seems that this function should not be used in 2D.

absolutely, but at least it should not crash. I started a series of edits to fix this (see commit f629e75), such that triangles always mean triangles (even in 2D where the notion of cell was preferred so far for this) and that edges always mean edges (even in 1D where the notion of cell was preferred so far for this). technically, this adds one if statement for a bunch of low-level access functions. for instance, getVertexTriangle() now needs to check the dimension of the triangulation to understand what to return (an actual triangle in 3D, or a "cell" in 2D). this will have an effect on performance. but it seems that in your case, such an if statement is mandatory: either in your code or in the triangulation API. it seems it might be best to put that in the triangulation API like I did (that may make the API more intuitive to other users). can you check that my commit fixes your issue? I'll progressively take care of the other functions in the near future. cheers, -- Dr Julien Tierny CNRS Researcher Sorbonne Universite http://lip6.fr/Julien.Tierny

…

On Wednesday, February 14, 2018 4:54:47 PM CET Charles Gueunet wrote: Dear TTK Team, I am actually developing a TTK plugin and I have encountered a strange behavior. My plugin needs to recover the triangles attached to a vertex of the mesh, so I call the `preprocessVertexTriangles()`. This function bug in the case of a 2D VTK Unstructured Grid (Segmentation Fault). ## Steps to reproduce 1- Create a new plugin with the createTTKmodule.sh 2- Add `triangulation_->preprocessVertexTriangles()` in the `setupTriangulation()` line 117. 3- Compile your plugin 4- Open a 2D VTU (surface) in paraview and apply your plugin You should have a Seg Fault here. ## Expected result As far as I understand, it seems that this function should not be used in 2D. We should use `preprocessVertexStar()` instead. But in the case of my plugin, I try to retrieve *triangles* for both 2D and 3D mesh. Would it be possible to have `preprocessVertexTriangles()` to do the same pre-processing than `preprocessVertexStar()` in 2D, to have only one function call for all data sets ? Thank you for your help Charles

[CharlesGueunet]

Hi Julien, I started a series of edits to fix this (see commit f629e75), such that

triangles always mean triangles (even in 2D where the notion of cell was preferred so far for this) and that edges always mean edges (even in 1D where the notion of cell was preferred so far for this). technically, this adds one if statement for a bunch of low-level access functions. for instance, getVertexTriangle() now needs to check the dimension of the triangulation to understand what to return (an actual triangle in 3D, or a "cell" in 2D). this will have an effect on performance. but it seems that in your case, such an if statement is mandatory: either in your code or in the triangulation API. it seems it might be best to put that in the triangulation API like I did (that may make the API more intuitive to other users). can you check that my commit fixes your issue? I'll progressively take care of the other functions in the near future.

These changes have fixed the issue. Thank you for your fast answer. Charles

[julien-tierny]

Thanks!