Cleanup raycast tests

[StafaH]

Description

Cleanup of raycast tests. Functionally similar but substantially more concise, and less kernels (avoiding excessive kernel compilation). ray_intersect_geomis a dispatch function for the remaining ray functions and can be used across all tests.

Checklist

• New or existing tests cover these changes
• The documentation is up to date with these changes
• CHANGELOG.md has been updated (if user-facing change)

Test plan

uv run --extra dev -m newton.tests -k test_raycast

Summary by CodeRabbit

• Tests
  • Consolidated ray-intersection tests into a single, unified test harness for improved maintainability and consistency.
  • Replaced repetitive per-primitive test code with table-driven cases, including a new rotated-box scenario.
  • Simplified mesh validation by streamlining mesh-specific test paths.
  • Reduced redundant test imports and cleaned up duplicated coverage, lowering overall test complexity.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Pro

Run ID: 2eb90c08-2c4e-46e3-9362-79ed32179131

📥 Commits

Reviewing files that changed from the base of the PR and between 9e16b6a and 264f28b.

📒 Files selected for processing (1)

• newton/tests/test_raycast.py

🚧 Files skipped from review as they are similar to previous changes (1)

• newton/tests/test_raycast.py

---

📝 Walkthrough

Walkthrough

Removed per-primitive Warp test kernels and refactored raycast tests to use a single unified kernel_test_geom calling ray_intersect_geom (and kernel_test_mesh / ray_intersect_mesh for meshes); tests converted to table-driven cases and redundant helpers removed.

Changes

Cohort / File(s)	Summary
Test File newton/tests/test_raycast.py	Removed per-primitive Warp kernel wrappers and direct primitive launches; consolidated primitive and mesh ray intersection tests to use kernel_test_geom (for primitives) and kernel_test_mesh (for meshes) with table-driven case lists. Updated imports to ray_intersect_geom and ray_intersect_mesh only; added rotated-box case and expanded plane cases.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• Refactor raycast and remove redundant code #2482: Related refactor of raycast API that centralizes raycast logic and changed primitive ray intersection return conventions, which these test updates depend on.

Suggested reviewers

• eric-heiden

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'Cleanup raycast tests' directly and concisely describes the main purpose of the changeset: consolidating and refactoring raycast test code to reduce redundancy and kernel compilation.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!

[eric-heiden]

🟡 After this refactor, test_ray_intersect_mesh launches kernel_test_geom with GeoType.MESH, which is the same dispatcher path already covered by test_mesh_ray_intersect_via_geom at lines 308-333. That leaves no direct coverage of the ray_intersect_mesh() entry point in this file, even though renderer code calls it directly at newton/_src/sensors/warp_raytrace/raytrace.py:439-446 and 680-687.

This distinction matters because ray_intersect_geom() hardcodes enable_backface_culling=False and _DEFAULT_MESH_MAX_T when it forwards to the helper (newton/_src/geometry/raycast.py:739-742), so a regression in the direct helper plumbing would slip past both tests here. The current test name and docstring also still read as if this test targets ray_intersect_mesh() itself.

Would you consider either restoring kernel_test_mesh so this test calls the helper directly, or renaming it to reflect that it is another ray_intersect_geom case and adding a separate direct ray_intersect_mesh() smoke test?

[eric-heiden]

⚪ After the refactor, every per-type test (test_ray_intersect_sphere, test_ray_intersect_box, and so on) also launches kernel_test_geom, so this function overlaps with them substantially. The sphere, box, capsule, cylinder, cone, and ellipsoid cases here use origin (-2, 0, 0) and direction (1, 0, 0), which match the hit case in each per-type test modulo delta. The only genuinely unique case is plane with origin (0, 0, 5) and direction (0, 0, -1) expecting 5.0, since test_ray_intersect_plane only uses 3-4-5 triples and never a unit-length axial ray.

Since this PR is a cleanup, consolidation would match that intent. A few options:

• remove test_geom_ray_intersect entirely,
• trim it to just the plane case,
• or fold the plane case into test_ray_intersect_plane and delete this function.

If you keep it as-is, a short comment noting which coverage is unique would help future readers.