[vtk-developers] Changing computation of PipelineMTime.

[Charles Law]

Hello,

In order to fix a problem I had with the pipeline re executing when it 
wasn't necessary, I had to change the computation of the pipeline modified 
time.  Although all the regression tests still pass, such a change warrants 
a warning on the developers list.

               filter --- mapper (piece 0 of 2)
            /
reader
            \
               filter --- mapper (piece 1 of 2)

The problem was that when a pipeline splits (the reader supplies two 
branches of the pipeline) and two mappers request different extents/pieces 
of the data, then the whole pipeline re executes multiple times on each 
render.  After branch 1 executes,  branch two executes and causes the 
reader to execute again to produce a different portion of the data.  The 
readers output gets a new "MTime" and "PipelineMTime".  Since the new 
PipelineMTime is later than the update time of the first mapper,  the next 
update to the first mapper causes every thing to execute again. ....

The only way I found to fix this problem is to remove the data's MTime from 
the computation of the PipelineMTime.


The implications of this:

If you muck around with a data object that "belongs" to a source, then the 
pipeline will not re execute for you.

If the data object has no source, then It behaves as before.

If this causes a problem for you, you can always do something like:

reader->Update();
// Get a pointer to the data object and make sure it does not destruct when 
released by the source.
vtkDataObject *data = reader->GetOutput();
data->Register(NULL);
// Orphan the data object so it has no source.
reader->SetOutput(NULL);

// Then do anything you want to the data.
data->GetPointData()->SetScalars(myScalars);

...




Charles.