[vtk-developers] VTK-5 - Filter with multiple outputs

[Ovidiu Toader]

> Your filter will need to override the ProcessRequest method and look for 
> a REQUEST_DATA_NOT_GENERATED. 
Thanks for this info! From your description it looks like this is what I am 
looking for. I've already noticed this key but wasn't sure how to use it.
After I found out that this particular request was the only one returning a 
Null from request->GetRequest() I decided to move away from it because it 
appears different from the other requests. 

>  Override the ComputePipelineMTime
> method and look at the request object for the output port that made the
> request.  Return the MTime from the object corresponding to that output
> port.
I tried this. My Algorithm overrides 
ComputePipelineMTime(vtkInformation *request)

However, all the calls to this method have a Null request. I thought that this 
request would be a good place to store info such as the output port which 
initiated the update but to my surprise it wasn't. 

I can provide more info about this request = 0 problem. When my algorithm's 
ComputePipelineMTime is called the backtrace looks like this:

starts with:

int vtkDemandDrivenPipeline::UpdatePipelineMTime() {
  ...
  this->ComputePipelineMTime(1,0,this->GetInputInformation());
  ...
}

followed by a variable number of calls like this:

unsigned long vtkDemandDrivenPipeline::ComputePipelineMTime(int 
forward,vtkInformation *request,vtkInformationVector **inInfoVec) {
...
   vtkInformation* info = inInfoVec[i]->GetInformationObject(j);
   // call ComputePipelineMTime on the input
   vtkExecutive* e;
   int producerPort;
   info->Get(vtkExecutive::PRODUCER(),e,producerPort);
   if(e)
     {
     unsigned long mtime = 
e->ComputePipelineMTime(1,request,e->GetInputInformation());
     if(mtime > this->PipelineMTime)
       {
       this->PipelineMTime = mtime;
       }
     }
...
}

which, eventually, lead to a call to my ComputePipelineMTime. The problem with 
this call sequence is that the same Null request initiated in 
vtkDemandDrivenPipeline::UpdatePipelineMTime  is propagated all the way to my 
ComputePipelineMTime. 

I don't feel like I have a good understanding of the implementation of this 
new pipeline so I refrain -- at least for now -- from making any suggestion 
about the way I see a possible fix to this problem. 

Thanks,
Ovidiu