Tighten relative tolerance for the testing

[andrewwinters5000]

No description provided.

[codecov]

Codecov Report

Merging #341 (2f25faa) into dev (5bef936) will decrease coverage by 0.02%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##              dev     #341      +/-   ##
==========================================
- Coverage   88.43%   88.40%   -0.03%     
==========================================
  Files          76       76              
  Lines        8886     8886              
==========================================
- Hits         7858     7856       -2     
- Misses       1028     1030       +2     

Flag	Coverage Δ
unittests	88.40% <ø> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/callbacks_step/save_solution_dg.jl	90.14% <0.00%> (-2.82%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5bef936...2f25faa. Read the comment docs.

[ranocha]

Did you look for CFL magic in the examples where you needed to increase the tolerance?

[andrewwinters5000]

Did you look for CFL magic in the examples where you needed to increase the tolerance?

I plan too for the 1D blast wave because it seems very fishy that switching the shock indicator causes tests to not satisfy rtol=1e-8. It is the same weirdness that all tests with the strict tolerance pass locally but not running on github.

For the parallel I need help from @sloede because I am having troubling running the MPI tests locally.

[sloede]

I confirmed for the 1D case that it's related to CFL magic. I propose to set the CFL to 0.5 in the elixir and to 0.2 for the two additional tests with density and pressure, then regenerate the references and try again with the regular tolerances. According to my tests, that should probably work (if not, maybe you need 0.4 in the elixir, but the 0.2 is conservative).

Could you do that please, @andrewwinters5000? I will look at the MPI stuff and see if there's an easy way to implement parallel Kahan summation, otherwise I'll just leave the larger tolerances for now.

[edit: reference changed]

[andrewwinters5000]

Could you do that please, @andrewwinters5000?

Yes, I will do this now

[sloede]

OK, I don't think it is easy enough to implement it in a proper way in parallel that would justify postponing the release of v0.3. Thus if there are no objections from @ranocha, I think this can be merged.