redesign ch10/msh2petsc.py

Question

redesign ch10/msh2petsc.py

bueler opened this issue 4 years ago · comments

It needs to work with different gmsh formats. Currently it expects format version 2.2, but that is at least at 4.1 now.

The script already reads a section at a time. Note that from the Gmsh format spec we have

To represent a simple mesh, the minimal sections that should be present in the file are $MeshFormat, $Nodes and $Elements. Nodes are assumed to be defined before elements.

But in Chapter 10 we also need $PhysicalNames. All other sections (e.g. $Entities) are already ignored by the design of msh2petsc.py.

The Gmsh doc site has a page for the legacy format which seems to be 2.2.

Plans:

Change check_mesh_format() to get_mesh_format() which also gets the version number. (Major,Minor? Or a legacy flag?)
Pass the version number into `read_SECTION()' functions, and have them act accordingly.

Ed Bueler · Answer 1 · Fri Jun 19 2020 05:30:13 GMT+0800 (China Standard Time)

Workaround for chapter 10 gmsh, until msh2petsc.py is improved:

Update to latest Gmsh.
Add -format msh22 whenever running gmsh.

Ed Bueler · Answer 2 · Mon Jun 22 2020 07:41:12 GMT+0800 (China Standard Time)

This is substantially done. (See branch update-to-gmsh41.) But there is an unwarranted line format assumption revealed by the koch/ example. So TODO: fix msh2petsc.py to remove unwarranted, and build regression to check.

Ed Bueler · Answer 3 · Mon Jun 22 2020 14:03:48 GMT+0800 (China Standard Time)

The koch example now works. As far as I can tell msh2petsc.py is correct. But there is a mysterious hang, at least on galago, for koch5 and higher.

Ed Bueler · Answer 4 · Tue Jun 23 2020 03:56:56 GMT+0800 (China Standard Time)

The mysterious hang is not present on thelio with latest petsc.

This issue is fixed in master.

Ed Bueler · Answer 5 · Tue Jun 23 2020 13:49:38 GMT+0800 (China Standard Time)

The "mysterious hang" on galago was fixed by upgrading the ubuntu 20.04 version of openblas. See OpenMathLib/OpenBLAS#2642