bindings for dune-gdt (part 2: discretizers)

[ftalbrecht]

Together with part 1 (#1383), this supersedes #1381.

• ensure vector-valued interpolation of data functions is working
• move visualizers from bindings/dunegdt.py to discretizers/dunegdt/gui.py?
• move interpolation and data function handling out of discretizers to share code?
• ensure code is working with current time steppers
• disable shifting of Dirichlet-values to match Add StationaryModel.deaffinize #1410
• ensure IPDG discretizers work as well as CG ones
• adress issues raised in dunegdt bindings #969
• ensure visualization works in notebooks + offline (e.g. use Use pyvista for Qt based visualizations #1402, cherry-pick 4b13a94, delete dunegdt-discretizers-pyvista branch afterwards)
• ensure docs
• require newer dune wheels
• ensure passing tests
• ensure demo
• ensure rebase on main after bindings for dune-gdt (part 1: bindings) #1383 has been merged

[ftalbrecht]

CI depends on pymor/docker#81

[ftalbrecht]

Just in case this makes it into the release: what is your opinion regarding the Dirichlet data in this pr, @sdrave? I am currently shifting the model to solve in the H^1_0 subspace and adapt the visualizer accordingly. Should I simply return the solution with the correct Dirichlet data as the other CG discretizers do?

[sdrave]

Won't get this done before the release.

[renefritze]

I've rebased this on main just now, @ftalbrecht. The new visualization stuff will not land in 2021.2. Did you look into the dummy visualization/vtu download we discussed elsewhere?

[sdrave]

Also moving to next release.

[github-actions]

Unit Test Results

6 files  6 suites   1m 18s ⏱️
1 tests 0 ✔️ 0 💤 0 ❌ 1 🔥
6 runs  0 ✔️ 0 💤 0 ❌ 6 🔥

For more details on these errors, see this check.

Results for commit 8433553.

[ftalbrecht]

@TiKeil reports to have working k3d visualization with K3D==2.6.6.

[ftalbrecht]

I re-based on current main to get things rolling again.

[ftalbrecht]

Try installing "pymor[full]" and then gdt, see what gets installed?

In a fresh 3.9 virtualenv, doing

pip install -e '.[full]'
pip freeze | ack multidict

in this branch gives multidict==6.0.2, so this is fine. Installing

pip install dune-gdt[visualisation]

afterwards does not change this.

Similarly, in a fresh 3.9 virtualenv, running

pip install -e '.[full]' dune-gdt[visualisation]
pip freeze | ack multidict

yields the same. So pymor and dune-gdt seem to agree on the multidict version, and I deduce that its a problem of our CI setup.
Looking at the logs of the failure, the first line mentioning multidict is

Collecting multidict<5.0,>=4.5

which is already a bad version requirement (due to lacking wheels). This is triggered by the call

devpi install pymor[full]

which is evaluated to

/builds/pymor/pymor$ /usr/local/bin/pip install -U -i http://pymor__devpi:3141/root/public/+simple/ pymor[full]  [PIP_PRE=1,PIP_USE_WHEEL=1]

[ftalbrecht]

Rebasing on current main again to see if this helps in the CI. Needs a restart after image updates are through.

[ftalbrecht]

Current CI failures are due to pyside problems. Ideas, @renefritze?

[renefritze]

The check says an unexpected version of pyside is installed in the image. Is the docker image based on current main image? Does the visualization extra pull in a fixed version of pyside?

[TiKeil]

In the ipdg discretizer, the weighted_h1_semi_penalty-product does not seem to be symmetric and always incorporates Dirichlet terms.