[Paraview-developers] Proposed ParaView CVS hierarchy

[Berk Geveci]

> > The new hierarchy seems fine to me, what I don't quite understand is why the
> > insistence that people recompiled another version of VTK for ParaView. 

Although I tend to compile and use VTK separately (and will be grumpy
for a while for not being able to do that), here is why I think this is
the best way to go:

1. In my experience, most ParaView end users end up compiling ParaView.
Either the binaries we provide do not work for them or they have to
compile against their own libraries (MPI is a good example. We just
can't distribute binaries for every MPI distribution out there, mangled
mesa is another one). Also, in my experience, when they compile VTK and
ParaView separately, things go wrong most of the time. They either do
not set the right options or the choose incompatible options (or they
link against wrong libraries etc etc).

2. Because of 1, we have to have a complete distribution. Once we have a
complete distribution, maintaining both an complete and a partial
distribution become troublesome.

3. It is easier and better to test the complete distribution of ParaView
and be sure that what the users (and developers) get is exactly what we
tested.

4. Developers (including me) make almost as many mistake as end users
when setting up. The complete distribution configure VTK, ParaView and
utility libraries appropriately so that the person configuring has to
deal with fewer options and has less change to make mistakes.

5. It is perfectly OK to use the VTK that is in ParaView for your own
VTK only development. You can build and run tests separately in the VTK
build tree. However, when you change VTK, you know that ParaView will be
re-compiled and re-linked if necessary when you run make. Many times, I
spent hours debugging ParaView to later find that it was broken because
I made a change to VTK but did not do a make in the ParaView tree.

-Berk