Emily Boudreaux
cb20faeb3d
The test suite is extensive. Note that LLMs were used to aid in the development of the test suite. All LLM generated code was manually reviewed.
1.2 KiB
1.2 KiB
GENERATIVE AI DISCLOSURE
-
Generative AI has been used to develop the test suite of this code base. Specifically ChatGPT 6 astra max.
-
There are two primary ways this was used
- Porting tests from a previous version of this code base to the current version. Specific instructions were given and code was manually reviewed to ensure correctness. This porting was required as significant parts of the interface have changed; however, the goal is to maintain the same numeric behavior.
- Developing new tests based on preexisting tests (e.g a test ensuring volume is consistent under an affine transformation was written and then we asked GenAI to write a number of other tests with a number of different affine transformations)
-
The backend consolidation added AI-assisted regressions comparing element coordinate extraction and assembly against native MFEM operations, and exercising the shared Newton algorithm with standard-library vectors without an MFEM dependency.
-
The TOP-export regressions were developed with AI assistance: analytic polytrope and manufactured geometry checks, physical-unit invariance, deformed/nonconforming FE sampling, axis handling, state preservation, publication failures, and an optional check with the actual TOP reader.