Improve VTK output file naming

[BenjaminRodenberg]

The way how treelog.userfile resolves conflicting filenames leads to problems with Paraview. I will provide more information below and I hope it's enough to understand the problems I'm facing.

How do I write output?

Usually a nutils script has the following layout:

# initialization
nutils.export.vtk(`out/solution`, ...)

for step in range(5):
    # computation
    nutils.export.vtk(`out/solution`, ...)

This results in the following output files:

$ ls out/
solution-1.vtk  solution-2.vtk  solution-3.vtk  solution-4.vtk  solution-5.vtk  solution.vtk

I like about the current approach that I only have to provide the prefix solution and nutils takes care of the rest. However, the way how treelog.userfile deals with conflicting filenames is problematic (see below).

What's the problem?

1. If I run the same script again the output is not overwritten, but the old output files are kept while the new output is "appended". I get

$ ls out/
solution-10.vtk  solution-11.vtk  solution-1.vtk  solution-2.vtk  solution-3.vtk  solution-4.vtk  solution-5.vtk  solution-6.vtk  solution-7.vtk  solution-8.vtk  solution-9.vtk  solution.vtk

2. The output is not working correctly with paraview, because the files solution-* are considered as one dataset and solution.vtk is ignored here:
   

What do I expect?

I'm working a lot with FEniCS. Here the following rules are applied, when writing output:

• when the filename solution is given, start with solution-0.vtk. I currently can get what I want, by manually renaming solution.vtk to solution-0.vtk:
  
• override existing files. Here, I currently have to manually delete old files before running a new simulation.

Further resources:

• Full example of nutils code that we are using + workaround for the problems described above: https://github.com/precice/tutorials/blob/5783ae56fbb10bb3ca4a5e8e9808ca248cc645b8/partitioned-heat-conduction/nutils/heat.py#L97-L166

  • The workaround forces us to manually compose the filename
  • Another workaround is to provide a dummy file solution.vtk such that the first file nutils write is not solution.vtk, but solution-1.vtk (see precice/tutorials@e7b4459).
  • Still leads to problems in paraview, if running the same case multiple times:
    

• How FEniCS uses output files in a similar context:

  • definition: https://github.com/precice/tutorials/blob/5783ae56fbb10bb3ca4a5e8e9808ca248cc645b8/partitioned-heat-conduction/fenics/heat.py#L157-L161
  • initial data: https://github.com/precice/tutorials/blob/5783ae56fbb10bb3ca4a5e8e9808ca248cc645b8/partitioned-heat-conduction/fenics/heat.py#L166-L168
  • further timesteps: https://github.com/precice/tutorials/blob/5783ae56fbb10bb3ca4a5e8e9808ca248cc645b8/partitioned-heat-conduction/fenics/heat.py#L229-L231

• Minimal example using treelog.userfile:
  nutils-export.txt