Added a parameter to switch between fraction of total error and fraction of cells

[lkaratun]

Added a parameter to switch between fraction of error and fration of cells for the mesh refinement/coarsening routine

[gassmoeller]

The implementation looks good, but I would prefer a parameter name like "Refine fraction of cells" or "Refine fraction of cells instead of error". "Use ..." seems to suggest to use a different input parameter than normal. @bangerth, @tjhei what is your opinion?

[tjhei]

add a space after "of" at the end of the string.

[tjhei]

"Refine fraction of cells"

This has the disadvantage that is only mentions "refine" and not "coarsening", maybe "Adapt by fraction of cells"?

[lkaratun]

I changed the name to "Adapt by fraction of cells"

[tjhei]

/run-tests

[gassmoeller]

Please remove one or both (your decision) of these empty lines

[gassmoeller]

Hi Lev, sorry to come up with more comments after already reviewing it at the hackathon (there was simply too little time there). Can you please change the minor things that I have commented on, and add an entry into /doc/modules/changes.h to make sure you get credit for this change in the next ASPECT release mail. Just copy one of the 'New: ...' entries in there and describe the new feature in a sentence or two. And is there a test that uses the new functionality? If not, please create a very simple one.

[lkaratun]

Hi Rene, no worries! I cleaned up the blank lines and added the decription to changes.h. I'm not sure about the tests though - I don't know how to check if any of them use the functionality (I would guess they don't), and I don't really know how to make one. Are there any guidelines for that? Or what would be a good place to start?

[gassmoeller]

We should write a tutorial about that in the manual at some point, but as a bottom line: Take one of the existing parameter files in the tests/ directory, modify it to use the new feature, check that the new feature does what it is supposed to do, and then just add the parameter file to the tests directory. You will then need to add another folder to that directory that is named like the parameter file, and add the model output files that proof that the feature is working (in your case likely just the file log.txt, and statistics. rename log.txt to screen-output for historical reasons). The test and output files should be as small and quick to run as possible. Let me know if you get stuck at some point.

[gassmoeller]

Hi Lev,
thanks for updating the test output. You will still need to rebase this branch to a recent master branch of ASPECT. Maybe you have done that with your local master branch, but that one is likely out-of-date (this also explains the PETSC errors, @tjhei just recently fixed that again on the master branch). So the procedure would be:

1. Switch to your local master branch (git checkout master)
2. Update your local master branch (e.g. git pull upstream master)
3. Switch back to your local branch
4. Rebase your local branch to the newly updated master branch (git rebase master). While at it, you can also squash your commits into a single commit by doing git rebase -i master and then select to squash all but the first commit (change 'pick' into 'squash' for every commit but the first in the list of commits). You will likely get a conflict in this stage that you must fix manually and then do a 'git rebase --continue'
5. Reupload as before with git push -f origin branch_name.

[gassmoeller]

The one failing test is due to changes on the master branch that happened since you opened this pull request. Just apply the changes.diff file from the tester one more time (or do the changes by hand) and this should be good to go.

[lkaratun]

Thanks Rene!

[gassmoeller]

Ready to merge. Nice! Thanks for contributing 👍