[vtk-developers] Correct handling of information propagation

[Brad King]

John Biddiscombe (CSCS) wrote:
> Ken Martin wrote:
>> If bounds are not provided then
>> you must go with some default extent.
> If there's no information because the filters haven't generated data 
> yet, (ie no bounds available), and a default value is not feasible (just 
> can't be guessed sensibly), is it safe to update the inputs during the 
> request information call and get the correct sizes? This is what I used 
> to do, but I am not sure if the new pipeline will like it.

One of the design principles of the new pipeline is that a filter should 
be able to satisfy a request sent to its ProcessRequest method without 
being aware of the executive or the pipeline at all.  It should work 
only with the information passed to this method's arguments.  Therefore 
it is invalid to try to update the inputs during REQUEST_INFORMATION. 
Something will produce an error message and the update will fail.

What you have here is a filter that can't compute its whole extent until 
the REQUEST_DATA pass (Execute).  Currently we deal with this by 
producing a fake whole extent (something like 0 0 0 0 0 0) during 
REQUEST_INFORMATION and then correcting it during REQUEST_DATA.

-Brad