[ooctipus]

Backports the three changes from the September 20–23 develop audit that are missing from release/3.0.0, alongside the existing cloning-guide backport:

• [Fix] Separate SimulationContext construction from lookup #7722 (4cf439dfeb): reject a second SimulationContext construction instead of silently discarding its configuration. Includes the original fixture cleanup and regression coverage.
• Fix DelayBuffer state mutation on invalid lag #7893 (22fcd71bc4): validate DelayBuffer lag values before mutating its live configuration.
• Make v3.0.0-EA the default documentation version #7947 (6e9ef17f59): make v3.0.0-EA the default documentation version in deployment and local multi-version builds.
• Remove stale sections from cloning guide #7129 (354168560f): clarify that replicate_physics=False skips the active physics context while other declared replication contexts still run.

Compatibility: repeated SimulationContext(...) calls now raise. Retrieve the live context with SimulationContext.instance(), or call clear_instance() before constructing a replacement. The original major changelog fragment includes this migration.

The runtime and test files match current develop. The docs-default backport adapts the release branch's existing redirect template, validates the target page, and preserves explicit overrides. The cloning-guide backport preserves the current release APIs; its obsolete sections were already removed.

Release audit

Checked all 35 PRs merged into develop from 2026-09-20 16:43:18 UTC through cff81e8473 (September 23), against release head 5d3e0b695e. 32 are already covered; the remaining three are included here. Git history, patch equivalence, GitHub's merged PR list, and remaining file differences were checked. Branch-specific version/changelog compilation and existing release-only changes are preserved.

Already covered PRs

26 have an equivalent patch in release: #7919, #7922, #7923, #7924, #7925, #6758, #7928, #7902, #7927, #7891, #7931, #7691, #7861, #7929, #7916, #7939, #4609, #7942, #7940, #7948, #7937, #7955, #7960, #7961, #7970, #7958.

Six were squash-merged or adapted: #7530 and #7899 in #7926; #7637 in #7933; #7860 in #7936; #7954 and #7950 in #7964. Their implementation differences were checked against the current release tree.

Validation

• 276 tests passed: SimulationContext (51, including CPU/CUDA), headless context construction (15), nine schema/fragment suites (165), four manager suites (38), and DelayBuffer (7).
• Both backported regressions reproduce with the original release implementations.
• Docs redirect recipe: local/deployment EA defaults, explicit override, and missing-target rejection passed with a stubbed Sphinx builder.
• All 22 changed files match the intended backport scope; uv run isaaclab -f and git diff --check passed.
• Fresh Sphinx validation (-b dummy -W --keep-going): passed without warnings, excluding static assets as the HTML builder does. Full HTML documentation CI is running.

Local runtime validation used PyTorch 2.12.0, CUDA 13.0, Isaac Sim 6.0.1.0, and an RTX 5090. Existing runtime/deprecation warnings occurred; the CI jobs validate the pinned environments.

The previous docs CI failure was an external connection reset fetching the Omniverse intersphinx inventory, producing one warning under -W. That URL returned HTTP 200 when rechecked. This update starts a fresh docs build.

[isaaclab-review-bot]

Isaac Lab Review Bot

This documentation-only backport clarifies the effect of replicate_physics=False in the cloning workflow. The addition is narrowly scoped and fits the existing global-only plan example and replication-order discussion.

• Design and architecture: The paragraph preserves the documented multi-context replication model: the active physics context is skipped while other declared contexts, including configured USD replication, may still execute. Its placement directly after the relevant example makes the distinction clear.
• API: No public API, configuration default, export, or type contract is changed. The text documents the existing cloner.replicate(plan, replicate_physics=False) usage already shown in the surrounding example.
• Implementation: Only prose was added. Existing code blocks, section structure, collision-filtering guidance, and references remain intact. The included change contains no source implementation to independently retrace beyond the documented context-dispatch flow.

No blocking issues. No inline issue met the actionable-evidence threshold; the assessment above records the review feedback.

Automated review; human maintainers own approval decisions.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge because the new documentation matches the replication dispatcher and tested scene behavior.

Summary

This documentation-only backport clarifies replication behavior when physics replication is disabled:

• The active physics replication context is skipped.
• Other declared contexts, including configured USD replication, continue running.
• The physics backend can parse the resulting per-environment prims independently.

_{Reviews (1) · Last reviewed commit: "Clarify cloning without physics replicat..."}

[ooctipus]

run-ci

[ooctipus]

run-ci