Unify boundary conditions

[JoshuaLampert]

I would like to tackle #1510 to make it more attractive to create a new breaking release rather sooner than later, such that #2753 can be merged soon.
This PR allows to pass the boundary conditions as Dict or as NamedTuple and allows to not pass periodic boundary conditions to make it the default. For the first part I'm not sure if this is what we want or if we would prefer to allow only one of the possibilities – Dict or NamedTuple.
I used the help of Claude to generate this PR.

Closes #1510.

[github-actions]

Review checklist

This checklist is meant to assist creators of PRs (to let them know what reviewers will typically look for) and reviewers (to guide them in a structured review process). Items do not need to be checked explicitly for a PR to be eligible for merging.

Purpose and scope

• The PR has a single goal that is clear from the PR title and/or description.
• All code changes represent a single set of modifications that logically belong together.
• No more than 500 lines of code are changed or there is no obvious way to split the PR into multiple PRs.

Code quality

• The code can be understood easily.
• Newly introduced names for variables etc. are self-descriptive and consistent with existing naming conventions.
• There are no redundancies that can be removed by simple modularization/refactoring.
• There are no leftover debug statements or commented code sections.
• The code adheres to our conventions and style guide, and to the Julia guidelines.

Documentation

• New functions and types are documented with a docstring or top-level comment.
• Relevant publications are referenced in docstrings (see example for formatting).
• Inline comments are used to document longer or unusual code sections.
• Comments describe intent ("why?") and not just functionality ("what?").
• If the PR introduces a significant change or new feature, it is documented in NEWS.md with its PR number.

Testing

• The PR passes all tests.
• New or modified lines of code are covered by tests.
• New or modified tests run in less then 10 seconds.

Performance

• There are no type instabilities or memory allocations in performance-critical parts.
• If the PR intent is to improve performance, before/after time measurements are posted in the PR.

Verification

• The correctness of the code was verified using appropriate tests.
• If new equations/methods are added, a convergence test has been run and the results
  are posted in the PR.

Created with ❤️ by the Trixi.jl community.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.06%. Comparing base (21e5d0d) to head (8bf0f1e).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2761      +/-   ##
==========================================
+ Coverage   97.04%   97.06%   +0.02%     
==========================================
  Files         574      573       -1     
  Lines       44765    44823      +58     
==========================================
+ Hits        43439    43505      +66     
+ Misses       1326     1318       -8     

Flag	Coverage Δ
unittests	97.06% <ø> (+0.02%)	⬆️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JoshuaLampert]

The CI errors are really weird, but they seem to appear consistently on the CI runner. I cannot reproduce locally.

[JoshuaLampert]

Ah, it's also failing in other PRs, see #2754 and #2733. The last successful CI run was 6 hours ago (9:17 AM UTC+1). It seems like the error happens first in performance_specializations here:
https://github.com/trixi-framework/Trixi.jl/actions/runs/21359613034/job/61479451985?pr=2761#step:7:724, which also has the most information in the stacktrace (the other errors only say Package Trixi errored during testing (received signal: 6)). Seems to be some problem with LLVM. I found JuliaLang/julia#47176 and https://discourse.julialang.org/t/segmentation-fault-edit-signal-6-has-somebody-a-clue/66206, but this didn't really help me figured out what's happening. Do you have any idea, @vchuravy?
I think it first happened 5 hours ago (10:37 AM UTC+1) here. Since then it's failing consistently.

Edit: I looked at the diff of the used package versions between the last successful and the first failing CI run. Only SciMLBase.jl updated from v2.134.0 to v2.135.0 and DifferentiationInterface.jl updated from v0.7.14 to v0.7.15. Given that the error happens in examples/tree_2d_dgsem/elixir_euler_ec.jl:47, which is

Trixi.jl/examples/tree_2d_dgsem/elixir_euler_ec.jl

Lines 44 to 47 in 880e251

	callbacks = CallbackSet(summary_callback,
	analysis_callback, alive_callback,
	save_solution,
	stepsize_callback)

I am currently suspecting it has something to do with SciMLBase.jl.

Edit2: No, it's not SciMLBase.jl. It's also failing with the old version.
Edit3: Seems to work again...

[ranocha]

Thank you very much for tackling this long-standing issue!

In the very end, everything ends up being type-stable using tuples also for the unstructured BCs:

Trixi.jl/src/solvers/dgsem_unstructured/sort_boundary_conditions.jl

Lines 27 to 44 in 880e251

	function UnstructuredSortedBoundaryTypes(boundary_conditions::Dict, cache)
	# extract the unique boundary function routines from the dictionary
	boundary_condition_types = Tuple(unique(collect(values(boundary_conditions))))
	n_boundary_types = length(boundary_condition_types)
	boundary_indices = ntuple(_ -> [], n_boundary_types)
	# Initialize `boundary_symbol_indices` as an empty dictionary, filled later in `initialize!`
	boundary_symbol_indices = Dict{Symbol, Vector{Int}}()
	container = UnstructuredSortedBoundaryTypes{n_boundary_types,
	typeof(boundary_condition_types),
	Vector{Int}}(boundary_condition_types,
	boundary_indices,
	boundary_conditions,
	boundary_symbol_indices)
	return initialize!(container, cache)
	end

Thus, I strongly prefer using NamedTuples - and only them, no Dicts - to keep type stability in the whole computation. Or do I miss a good reason for using Dicts?

We also need to remember to add a NEWS.md entry when the design has been finalized.

[JoshuaLampert]

I agree that consistently using NamedTuples (and not using Dicts anymore) is a good idea. I did that. A couple of remarks and questions:

• The DGMulti solver used NamedTupless for the boundary conditions, but for some reason another type of syntax to construct them (of the type (; :symbol => boundary_condition) instead of (; name = boundary_condition). I also unified that, but this PR should not be breaking for DGMulti.
• DGMulti still uses Dicts for is_on_boundary. Is that fine or should we also use NamedTuples there?
• Did I understand your comment above correctly that UnstructuredSortedBoundaryTypes should also accept NamedTuples instead of Dicts, @ranocha? If yes, I'll also address that. Edit: Addressed.
• The failing Downstream tests are expected because this PR is breaking for TrixiAtmo.jl and TrixiShallowWater.jl. Once this PR is ready, I can create corresponding PRs in the downstream packages. Edit: I already created Unify boundary conditions TrixiAtmo.jl#141 and Unify boundary conditions TrixiShallowWater.jl#126.

[ranocha]

Great, thanks!

• Did I understand your comment above correctly that UnstructuredSortedBoundaryTypes should also accept NamedTuples instead of Dicts, @ranocha? If yes, I'll also address that.

Yes - if we keep working with NamedTuples (and Tuples etc.) but not Dicts in all steps, we should be able to avoid type instabilities completely.

[ranocha]

allows to not pass periodic boundary conditions to make it the default

This is something we might discuss in the next Trixi.jl meeting as well. I have observed that it can be a source of annoying-to-debug errors, that everything is periodic by default. Maybe it's time to change this and require passing periodic = true and the corresponding periodic BC?

[JoshuaLampert]

Thanks for the review! I have one open question above

DGMulti still uses Dicts for is_on_boundary. Is that fine or should we also use NamedTuples there?

before I move to the periodic boundary conditions.

I discussed with @jlchan that switching to NamedTuples is preferred. And it turned out StartUpDG.jl was already perfectly preferred for accepting NamedTuples. So this is done in 552beb3.

[jlchan]

Thanks for the review! I have one open question above

DGMulti still uses Dicts for is_on_boundary. Is that fine or should we also use NamedTuples there?

before I move to the periodic boundary conditions.

I discussed with @jlchan that switching to NamedTuples is preferred. And it turned out StartUpDG.jl was already perfectly preferred for accepting NamedTuples. So this is done in 552beb3.

Thanks for adding that @JoshuaLampert!

[JoshuaLampert]

This PR is finished from my point of view, @ranocha. It left it in draft mode so that it does not get accidentally merged.

[ranocha]

Thanks a lot!

[ranocha]

Please see also the comment #2761 (comment) above