[vtkusers] segfault on cell picking, and solution

[David Gobbi]

When you write a polydata filter, the pipeline executive will automatically
initialize the output data before the RequestData() method is called.  So
writers of polydata filters will never have to call DeleteCells().  It's only
something that people who build polydata "by hand" have to worry about.
I don't know if it is documented anywhere.

 - David

On Fri, Mar 1, 2013 at 12:20 PM, Robert Dean <bob.dean at gmail.com> wrote:
> Hello,
>
> I was building the polydata from scratch, so no filters were involved.  Main
> reason for posting this was that I could not find a solution online, and
> thought it might be useful. Does the documentation about how to build
> filters include directions on how to use DeleteCells/BuildCells?
>
> -Bob
>
>
> On Fri, Mar 1, 2013 at 2:12 PM, David Gobbi <david.gobbi at gmail.com> wrote:
>>
>> Hi Robert,
>>
>> I'd like to get a feel for whether or not this might be is a bug.
>>
>> Were you building the polydata yourself (i.e. from scratch)
>> or was it produced by a filter?  And if the latter, what filter
>> was producing it?
>>
>> Going a bit further, was the polydata that you were picking
>> also being used as an input to a filter, i.e. were you picking
>> a polydata that was somewhere in the middle of a pipeline?
>> And if so, what filter?
>>
>> It seems to me that your application shouldn't have to call
>> DeleteCells() unless your application is also building the
>> polydata from scratch.  If your application is using VTK
>> sources, readers, or filters to produce the polydata then
>> really VTK should be calling DeleteCells()/BuildCells()
>> automatically, and if VTK isn't doing this, then there must
>> be a bug somewhere e.g. in one of the filters you are using.
>>
>>  - David
>>
>>
>> On Fri, Mar 1, 2013 at 10:35 AM, Robert Dean <bob.dean at gmail.com> wrote:
>> > Hello,
>> >
>> > For a long time I had an issue where my application would randomly crash
>> > while picking cells. Each time the stack dump was in the depths of  the
>> > vtk
>> > code.  What I tracked it down to was that when an actor is first
>> > clicked,
>> > vtk allocates an internal cell array to assist in mapping the click to
>> > cell
>> > index.  This array is not automatically resized on future clicks.
>> >
>> > Therefore when the actor is first clicked an array of N is allocated. If
>> > the
>> > array is then updated to contain N*2 cells, clicking cells N through
>> > 2N-1
>> > will access outside the cell index array, resulting in the segfault.
>> >
>> > As cell picking is a rare event in the application timing, the solution
>> > was
>> > to call polydata->DeleteCells() method on each update.
>> >
>> > Posting in case someone else runs into the same problem.   I am
>> > currently
>> > using vtk 5.10, so I do not know if a newer version addresses this
>> > issue.
>> >
>> > -Bob
>
>