Introduce global test tolerance

[davidrudlstorfer]

Closes #192

This PR introduces a global testing tolerance and removes all hard coded tolerances from all tests (which use the assert_results_equal functionality)

In another PR the array comparison should be also centralized, but this is #184

I needed to update 2 reference files so everything works locally, let's see how the pipelines are doing

[Copilot]

Pull Request Overview

This PR centralizes tolerance settings for result comparisons by introducing global defaults and removes individual tolerance parameters from all tests.

• Introduce RELATIVE_TOLERANCE and ABSOLUTE_TOLERANCE constants in conftest_result_comparison.py.
• Update assert_results_equal and compare_vtk_files to use these defaults.
• Remove hard-coded rtol/atol arguments and the ASSERT_RESULTS_EQUAL_TOL constant from all test files.

Reviewed Changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/conftest_result_comparison.py	Added global tolerance constants and updated fixtures
tests/test_mesh_creation_functions.py	Removed explicit rtol arguments from assertions
tests/test_four_c_simulation.py	Ditto
tests/test_create_cubit_input.py	Removed local tolerance constant and explicit args
tests/test_cosserat_curve.py	Removed explicit rtol/atol from assertions
tests/reference-files/test_four_c_beam_to_solid.4C.yaml	Reformatted beam triads floats to full precision

Comments suppressed due to low confidence (3)

tests/conftest_result_comparison.py:41

• [nitpick] Add a module-level docstring or comment explaining that RELATIVE_TOLERANCE and ABSOLUTE_TOLERANCE are the defaults used by assert_results_equal so contributors understand why they exist and how changing them affects all tests.

RELATIVE_TOLERANCE = 1e-13

tests/reference-files/test_four_c_beam_to_solid.4C.yaml:134

• [nitpick] Consider standardizing float formatting for readability (e.g., using 0.0 instead of -0.0 and rounding to a consistent number of decimal places) to keep reference files concise.

  - 9 BEAM3R HERM2LINE3 28 30 29 MAT 1 TRIADS -0.0 -1.5707963267948968 -0.0 -0.0 -1.5707963267948968 -0.0 -0.0 -1.5707963267948968 -0.0

tests/conftest_result_comparison.py:40

• Add a note or link here (e.g., in the project’s CONTRIBUTING or changelog) referencing issue Possibly rethink / discuss handling of tolerance in new testing framework #192 so that the rationale for introducing global tolerances is recorded for future maintainers.

# GLOBAL DEFAULT TEST TOLERANCES

[davidrudlstorfer]

@isteinbrecher very interesting! The pipeline fails on macOS (especially ARM based OSs). Should we increase the tolerance for this one test or the global test tolerance? I am curious on your opinion:)

[isteinbrecher]

Yes, that file makes the issues (I created it on macOS). I had a quick look and could not figure out what is going wrong, where the tolerances can be increased - I suspect different versions of the scipy interpolate functionality, but I am not sure. Anyway, this should not concern us, let's simply put both tolerances for this test to 1e-12, then everything should pass. I would then also not adapt this result file.

[isteinbrecher]

Nice, I marked this file as ported to double precision in #356