Fix find active cell around point. #10174
Conversation
luca-heltai
force-pushed
the
fix-find-cell-around-point
branch
from
bc3ba0d
to
31a7d56
Compare
|
This test currently fails. I'm trying to figure out a fix. |
luca-heltai
force-pushed
the
fix-find-cell-around-point
branch
from
31a7d56
to
82a2e4b
Compare
|
The fix is dirty: I did not understand what is wrong with the actual algorithm. This algorithm was developed by @gassmoeller and myself during a deal.II developer workshop, and it seems to work most of the times, but sometimes it fails. I have not understood why, so in the fix I revert to the other algorithm when things are failing. The other algorithm is less smart and much slower, but seems to do the job correctly. |
|
/rebuild |
|
Hm, but then we have a regression-of-sorts. That's disappointing as well. We should really try to find out what the actual problem is... |
I agree. But there are two bugs now in the way of step-70. I'd like to try and fix the other one as well, and see if we manage to get going. At least now the behavior is correct. We should keep the issue open, even if we decide to merge this. |
|
In order to assess how to move forward: What's the other bug you run into? |
|
@bangerth Please re-review and accept if appropriate. I would like to get this into rc2. |
tamiko
changed the title
|
@bangerth let us move forward here. We have time later to understand why the slow path is needed. |
kronbichler
dismissed
bangerth’s stale review
The new commit fixes it and we need to move forward.
|
@luca-heltai -- can you open a PR that reminds us that we need to investigate this? |
Working on step-70, we hit this bug which makes running particle based applications impossible on large clusters.
I'm not sure yet hat is wrong. For the moment, this is a failing test. The (correct) output is generated using line 46.
@gassmoeller @blaisb
This is one of the two bugs we uncovered with step-70.
My educated guess is that the version that uses a
Cacheobject also does things differently, like passing a tree of locally owned vertices, and therefore it fails to find the points when they are on artificial cells.I'm going to bed... tomorrow I'll try to see if I can solve this.