meshio4

[gdmcbain]

Fixes #1092.

(This is mostly to see how big a change is required and to make use of the continuous integration to test against more versions of Python etc. than I have installed.)

[sdrave]

@renefritze, somehow this PR did not trigger a CI build. Maybe related to the recent Gitlab update?

[renefritze]

I hope the general CI issue is fixed now.

CI checks if dependencies.py and the requirements* files match. If you execute ./dependencies.py (or just make) and commit the update, then CI should continue.

[renefritze]

Sorry I didn't think of that right away, but I'll have to update the CI images so that the pypi mirror serves the newer meshio.

[renefritze]

Sorry for the spam, @gdmcbain, the bot isn't quite doing what it's supposed to and needs some reprogramming still.
Meanwhile, docker images are updating here.

[renefritze]

There's now an actual pytest error happening: https://zivgitlab.uni-muenster.de/pymor/pymor/-/jobs/423244#L966

[gdmcbain]

There's now an actual pytest error happening: https://zivgitlab.uni-muenster.de/pymor/pymor/-/jobs/423244#L966

Yes. I see that it's in

https://github.com/nschloe/meshio/blob/d467cb8a49e218b7bee899f59def3c3f898b3133/meshio/gmsh/_gmsh41.py#L117

I'm not getting the error here though with

python src/pymordemos/elliptic_unstructured.py 6 16 1e-1

Perhaps as without an explicit installation of Gmsh, pyMOR's picking up /usr/bin/gmsh which is 4.4.1 for Ubuntu 20.04.1 LTS. If (as usual) I do

pip install gmsh-sdk

I get 4.6.0, and no error with that either.

At https://zivgitlab.uni-muenster.de/pymor/pymor/-/jobs/423244#L981, it looks like Gmsh 4.1.3. This was a shortlived buggy version of Gmsh, as noted in nschloe/meshio#797:

That's a bug indeed, but mind you not many people take interest in this particular version of gmsh. It uses an "intermediate" format that was fixed on version after. You might want to upgrade.

Any chance of upgrading from Gmsh 4.1.3?

[sdrave]

I get 4.6.0, and no error with that either.

At https://zivgitlab.uni-muenster.de/pymor/pymor/-/jobs/423244#L981, it looks like Gmsh 4.1.3. This was a shortlived buggy version of Gmsh, as noted in nschloe/meshio#797:

That's a bug indeed, but mind you not many people take interest in this particular version of gmsh. It uses an "intermediate" format that was fixed on version after. You might want to upgrade.

Any chance of upgrading from Gmsh 4.1.3?

Debian stable, on which our CI images are based if I remember correctly, has 4.1.5+really4.1.3+ds1-1. On my Debian machine I get the same error locally. However, I would be fine with adding gmsh to our pymor[full] requirements (gmsh-sdk is deprecated). It has wheels for all relevant platforms.

[renefritze]

I agree, adding the python package that gets us the right gmsh would be a better solution than to get a newer gmsh binary into our debian stable based images.

I'll update the images again to make gmsh available.

[renefritze]

I realized I missed some stuff in my previous commit @gdmcbain, so I've added a fixup commit which we should squash before the merge.

[renefritze]

It seems we're now having trouble calling the gmsh binary/script installed from the gmsh python package. I can look into that after the pyMOR course is over.

[renefritze]

The CI images needed some more Xorg packages that had been installed as a gmsh apt package before. It'll take a bit to get those updates out and then I'll restart the gitlab pipelines.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@5d99181). Click here to learn what that means.
The diff coverage is 87.50%.

Impacted Files	Coverage Δ
src/pymor/discretizers/builtin/grids/gmsh.py	87.50% <87.50%> (ø)

[renefritze]

Tests are looking good now. If there's nothing else @gdmcbain @pymor/pymor-devs , I'll merge this in a day.

[gdmcbain]

No, that looks good. Thanks very much for following this through; there were a few more things required than I had envisaged.

[renefritze]

No, that looks good.

Great. This is now on track to merge. Thank you again for your work, @gdmcbain !