[kgryte]

Description

What is the purpose of this pull request?

This pull request:

• migrates the tests for math/base/special/rempio2 from EPS-scaled relative tolerance comparisons to ULP-based assertions using @stdlib/assert/is-almost-same-value, per [RFC]: Migrate math/base/special packages from relative tolerance testing to ULP difference testing (tracking issue) #11352.
• removes the now unused @stdlib/math/base/special/abs and @stdlib/constants/float64/eps requires from test/test.js and test/test.native.js.

Details:

• Package converted: math/base/special/rempio2 (test/test.js and test/test.native.js; these are the only test files in the package).

• Final ULP bounds, identical across both test files:

  test	previous tolerance	final bound
  reconstruction, positive (randu()*100)	EPS * abs( x )	1 ULP
  reconstruction, tiny positive (randu()*1.0e-100)	EPS * abs( x )	exact
  reconstruction, negative (-100 * randu())	EPS * abs( x )	1 ULP
  reconstruction, tiny negative (-1.0e-100 * randu())	EPS * abs( x )	exact
  reconstruction, multiples of π/4	EPS * abs( x )	1 ULP

• Measured minimum: 1 ULP is the smallest integer bound under which the positive and negative reconstruction loops pass. Instrumenting ulpdiff( ( PIO2*n ) + ( y[0]+y[1] ), x ) directly over 1.2M random samples per range gives a maximum observed difference of exactly 1 ULP for the ±[0,100) ranges and 0 for both tiny ranges and for all 20 multiples of π/4. Lowering the bound to 0 produces 88–97 failing assertions per run.

• For the two tiny ranges, |x| < 2^-27, so the function returns n = 0 with y[0] = x and y[1] = 0 and reconstruction is bit-for-bit exact; those assertions therefore use exact equality, matching the float32 counterpart math/base/special/rempio2f. Exact equality (rather than a 0 ULP bound) is used deliberately, because randu() is documented as returning on the interval [0,1) and so x may be -0, which isAlmostSameValue( z, x, 0 ) would reject.

• The multiples-of-π/4 loop measures 0 ULP here, but is kept at 1 ULP to match the other reconstruction loops and rempio2f, and to leave headroom for compiler-dependent contraction in the C implementation on platforms other than the one measured.

• Determinism: the native add-on was compiled locally so that test/test.native.js actually executed rather than being skipped, and the suite was then run four times at the final bounds. Every run: 83291 assertions in test/test.js and 83291 in test/test.native.js, all passing.

• Linting is clean for both files under etc/eslint/.eslintrc.tests.js, and the only changed files are the two test files.

Related Issues

Does this pull request have any related issues?

This pull request has the following related issues:

• [RFC]: Migrate math/base/special packages from relative tolerance testing to ULP difference testing (tracking issue) #11352

Questions

Any questions for reviewers of this pull request?

The assertion messages in test/test.native.js were 'can recover input value' rather than 'returns expected value'; they have been normalized to 'returns expected value' to match the idiom in #11352 and in rempio2f. Happy to restore the original wording if reviewers would prefer to keep it.

Other

Any other information relevant to this pull request? This may include screenshots, references, and/or implementation notes.

The conversion mirrors the float32 counterpart in the same family, math/base/special/rempio2f, which has the same test structure and already uses t.strictEqual( isAlmostSameValue( z, x, 1 ), true, 'returns expected value' ); for the reconstruction loops and exact equality for the tiny ranges.

The for large/huge positive/negative input values tests contain no tolerance math and are unchanged.

Checklist

Please ensure the following tasks are completed before submitting this pull request.

• Read, understood, and followed the contributing guidelines.

AI Assistance

When authoring the changes proposed in this PR, did you use any kind of AI assistance?

• Yes
• No

If you answered "yes" above, how did you use AI assistance?

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

Disclosure

If you answered "yes" to using AI assistance, please provide a short disclosure indicating how you used AI assistance. This helps reviewers determine how much scrutiny to apply when reviewing your contribution. Example disclosures: "This PR was written primarily by Claude Code." or "I consulted ChatGPT to understand the codebase, but the proposed changes were fully authored manually by myself.".

This PR was authored by Claude Code, running unattended as a scheduled task, which selected the package, mirrored the conversion idiom from previously merged conversions, and empirically determined and verified the minimum ULP bounds.

---

@stdlib-js/reviewers

---

Generated by Claude Code

[stdlib-bot]

Coverage Report

Package	Statements	Branches	Functions	Lines
math/base/special/rempio2	$\\color{red}893/907$ $\\color{green}+98.46\\%$	$\\color{red}82/87$ $\\color{green}+94.25\\%$	$\\color{green}5/5$ $\\color{green}+100.00\\%$	$\\color{red}893/907$ $\\color{green}+98.46\\%$

The above coverage report was generated for the changes in this PR.