[Test] Optimize Newton test runtime

[StafaH]

Description

Reduces isaaclab_newton CI runtime while preserving distinct behavioral coverage.

• Tests actuator target-mode configuration directly instead of launching a simulation for pure configuration behavior.
• Preserves singleton and batched coverage at initialization boundaries, then uses representative batched layouts for downstream behavior.
• Consolidates redundant rigid-object and rigid-object-collection parameter matrices.
• Keeps the complete contact lifecycle shape matrix while narrowing the secondary horizontal-collision matrix to representative primitive and mesh cases.
• Combines contact-sensor metadata checks into one scene.
• Removes simulation-backed print/string smoke tests and replaces tautological self-comparisons with cross-view behavioral assertions.

Before and after measurement

Measured with the CI-style per-file test runner on the same workstation, using a new empty Warp cache for each revision. Both revisions use the repository-default Warp backward configuration:

Revision	Cases	Result	Wall time
upstream/develop (c4a275975)	1,180	0 failures, 0 errors	30m 44.38s
This PR (552b329a8)	907	0 failures, 0 errors	21m 41.56s

This saves 9m 02.82s (29.4%) in a cold-cache run. Both measurements executed all 35 Newton test files.

Measurement command:

TEST_FILTER_PATTERN=isaaclab_newton \
WARP_CACHE_PATH=<new-empty-cache> \
uv run --no-sync python -m pytest tools -q

Type of change

• Test and CI performance improvement (non-breaking)

Validation

• Full optimized Newton suite: 35 files, 907 cases, 0 failures, 0 errors.
• Full untouched baseline suite: 35 files, 1,180 cases, 0 failures, 0 errors.
• Focused changed suites all pass: articulation, rigid object, rigid object collection, contact sensor, frame transformer, IMU, joint wrench, and PVA.
• Ruff and Ruff format hooks pass for all changed Python files.
• git diff --check passes.
• uv run --no-sync isaaclab -f passes all applicable hooks; the repository-wide changelog checker reports pre-existing fragment immutability/missing-fragment issues in unrelated packages. This PR includes an isaaclab_newton .skip fragment.

Checklist

• I have read and followed the contribution guidelines.
• I have run the relevant tests and included the results above.
• I have formatted and linted the changed files.
• I have added the appropriate changelog fragment.

Screenshots

Not applicable.

[isaaclab-review-bot]

Isaac Lab Review Bot

The Newton test-runtime optimization reduces redundant simulation and parameter coverage, but the new suite-level Warp configuration leaks beyond the Newton tests in shared pytest processes.

• Design and architecture: The package conftest sets a process-global Warp option during collection without restoring or isolating it. Directory scoping of conftest discovery does not scope the resulting Warp state, so other collected packages can inherit Newton-specific behavior.
• API: No production or public API surface changes. The test-only private helper import and optional scene configuration field do not affect users, and the package includes an appropriate .skip changelog fragment.
• Implementation: The narrowed matrices and cross-view assertions retain representative batched coverage, but wp.config.enable_backward = False must be confined to isolated Newton execution or safely saved and restored so broad pytest invocations remain order-independent.

Minor fixes needed. Posted 1 actionable finding inline.

Automated review; human maintainers own approval decisions.

[isaaclab-review-bot]

🟡 Warning · Implementation — Package conftest mutates global Warp config

wp.config.enable_backward is process-global, and pytest imports a directory's conftest.py during collection, before any test runs. A session that collects source/ therefore disables Warp backward codegen for every other package's tests as well, and the previous value is never restored, so behavior depends on which directories are collected. Scope this to Newton-only execution (save/restore around the suite, or set it in the isolated Newton runner).

[greptile-apps]

Greptile Summary

This PR reduces Newton test-suite runtime by narrowing redundant parameter matrices, replacing simulation-backed configuration checks with direct unit checks, consolidating sensor scenes, disabling unused Warp backward generation, and removing low-value smoke assertions.

• Preserves singleton and batched initialization coverage while using representative batched layouts downstream.
• Retains complete contact-lifecycle coverage and narrows secondary collision-shape combinations.
• Replaces tautological rigid-body assertions with cross-view behavioral comparisons.

Confidence Score: 5/5

The PR appears safe to merge because the changes are confined to test optimization and no concrete correctness or security defect remains.

Representative boundary coverage is retained, the new cross-view assertions strengthen behavior checks, and the global Warp setting is consistent with the repository’s lack of Warp-autodiff tests.

Important Files Changed

Filename	Overview
source/isaaclab_newton/test/conftest.py	Disables unused Warp backward-code generation for Newton tests at collection time, consistent with existing repository test configuration.
source/isaaclab_newton/test/assets/test_articulation.py	Converts target-mode checks to direct configuration tests and substantially reduces redundant downstream batch-size combinations.
source/isaaclab_newton/test/assets/test_rigid_object.py	Narrows batch matrices and replaces self-comparisons with meaningful root/body cross-view assertions.
source/isaaclab_newton/test/assets/test_rigid_object_collection.py	Consolidates environment/object-count matrices while retaining representative singleton and batched initialization boundaries.
source/isaaclab_newton/test/sensors/test_contact_sensor.py	Narrows secondary collision-shape coverage and consolidates three distinct metadata configurations into one multi-sensor scene.

_{Reviews (1): Last reviewed commit: "Optimize Newton test coverage" | Re-trigger Greptile}

[AntoineRichard]

LGTM