GridTools::partition_triangulation() fails on refined grid

[Matteo Malvestiti]

Good afternoon everybody,

I'm experiencing a quite confusing error.

The demo example is straightforward:

1) Create a simple Triangulation<2>, we're using GridGenerator::hyper_ball()

2) Partition it with GridTools::partition_triangulation(...)

3) Create TriangulationDescription::Description

4) Create parallel::fullydistributed::Triangulation

We use dealii 9.6.0-rc1 from the M1 mac .dmg installer.

We also test on the Docker image with deal.II 9.5.0.

The aforementioned code works in both scenarios.

BUT, as soon as I introduce a simple global refinement on the serial triangulation, like:

1.5) tria_base.refine_global(5);

deali.II 9.5.0: Works fine

deal.II 9.6.0: ERROR - or actually goes in a deadlock during the call to

tria_pft.create_triangulation(description);

-------

I attach the simple demo script as a .cpp file.

Does anybody know something about this?

Kind regards,

Matteo Malvestiti

[Wolfgang Bangerth]

On 2/10/25 08:05, Matteo Malvestiti wrote:
>
> *BUT*, as soon as I introduce a simple global refinement on the serial

> triangulation, like:
> 1.5) tria_base.refine_global(5);
>
> deali.II 9.5.0: Works fine
> deal.II 9.6.0: ERROR - or actually goes in a deadlock during the call to

> /tria_pft.create_triangulation(description);/
> /
> /

> -------
> I attach the simple demo script as a .cpp file.
>
> Does anybody know something about this?

Matteo -- can you help debug this a bit further? Does the problem
already happen with two MPI processes, or how many do you need? Do you
actually need 5 global refinements, or what is the minimum so that the
problem happens?

Lastly, let's assume that you can get away with just two MPI processes.
Then it's not so difficult to run the two processes in a debugger. If
the program ends in a deadlock, can you obtain a backtrace for where the
program is stuck? (If you've never run a parallel program in a debugger,
check out lecture 41.25 at
https://www.math.colostate.edu/~bangerth/videos.html .)

Thanks already for producing a fairly minimal program -- this is exactly
what is necessary to debug the issue!

Best
Wolfgang

[Matteo Malvestiti]

Dear Prof. Bangerth,

I will gladly do all the debugging you've suggested and report back to the community. 

I'm right now very close to a deadline so I will need to postpone it for a couple of weeks, but I'll work on it then.

Thanks a lot for the support,

Matteo Malvestiti