ENH: Convert Heaviside step function tests to itkHeavisideStepFunctionGTest

[hjmjohnson]

Convert legacy ITK CTest driver test to GoogleTest (GTest) framework.

This is part of the ongoing effort to modernize ITK's test suite by converting no-argument CTest tests to the GoogleTest framework for improved test organization, better failure reporting, and modern C++ testing practices.

🤖 Generated with Claude Code

[N-Dekker]

This one produces more lines of code than the original test (lines changed: 56 additions & 45 deletions). In general, I would expect (and prefer) less lines of code, when converting an old test to GTest 🤷

[hjmjohnson]

This one produces more lines of code than the original test (lines changed: 56 additions & 45 deletions). In general, I would expect (and prefer) less lines of code, when converting an old test to GTest 🤷

That is usually the case, but many of the tests only printed diagnostics to the console, and when function return values can be checked for accuracy, Claude would sometimes add explicit tests. I think it is very good to test for correctness in addition to simply exercising the code.

[N-Dekker]

Please don't add extra checks with this pull request. Please just do Test-to-GoogleTest conversion, with this PR.

[blowekamp]

Please don't add extra checks with this pull request. Please just do Test-to-GoogleTest conversion, with this PR.

I think a few small additions are fine.

Also, I believe the separation of the batch of testing into separate PRs was done to accommodate a request to make review easier. However, there is a such a large number of PRs, computer generated, that the first page of PRs is full with this one task. I need to page to get to mine and others' work 🙀

For this type of operation in the future, I'd recommend just submitting a couple items, and get the human reviews on those. Perhaps selecting the interesting cases to share and discuss. Then once a standards is established making a PR with the bulk of the changes.

[N-Dekker]

For this type of operation in the future, I'd recommend just submitting a couple items, and get the human reviews on those. Perhaps selecting the interesting cases to share and discuss. Then once a standards is established making a PR with the bulk of the changes.

Thanks @blowekamp ! I just posted a related topic at our Discourse, because I believe that there is a more general challenge here, how to tame AI! Discourse: AI generated pull requests overwhelming, hard to review carefully

[N-Dekker]

I think a few small additions are fine.

@blowekamp Claude made the following small additions with this PR:

  // At x=0 with epsilon=0.5: Evaluate = 0.5, EvaluateDerivative = 2/pi ≈ 0.63662
  EXPECT_NEAR(functionBase0->Evaluate(0.0), 0.5, 1e-10);
  EXPECT_NEAR(functionBase0->EvaluateDerivative(0.0), 2.0 / itk::Math::pi, 1e-4);

Are you able to review those additions carefully? Honestly it would take me quite some time! I would really have to analyze the tested function to see if those expect's are meaningful. I don't think we should blindly trust AI. It's a useful, powerful tool, but it can also make mistakes! ("Hallucinations")

[hjmjohnson]

I think a few small additions are fine.

@blowekamp Claude made the following small additions with this PR:

  // At x=0 with epsilon=0.5: Evaluate = 0.5, EvaluateDerivative = 2/pi ≈ 0.63662
  EXPECT_NEAR(functionBase0->Evaluate(0.0), 0.5, 1e-10);
  EXPECT_NEAR(functionBase0->EvaluateDerivative(0.0), 2.0 / itk::Math::pi, 1e-4);

Are you able to review those additions carefully? Honestly it would take me quite some time! I would really have to analyze the tested function to see if those expect's are meaningful. I don't think we should blindly trust AI. It's a useful, powerful tool, but it can also make mistakes! ("Hallucinations")

The values were obtained by running the test, failing, identifying the correct values, and updating the test to insert them. I insisted that Claude-code build and run the tests to insert values. I did not go back to first principles to compute the math by hand, but I assumed that the previous results that were not tested should remain the same across platforms.

When running the test, the returned value was 0.63662 (which may or may not be 2/pi), and Claude-code replaced 0.63662 with 2.0/ itk::Math::pi as its guess of what was expected as the result, based on the code in EvaluateDerivative(0.0) and the resulting output of 0.63662.

========
I'm going to put this PR into DRAFT mode to think about what a 'good' course of action is (not the 'best' course of action, because I think there will be disagreement on what is 'best').

Resolved all requested issues.