Skip to content

Add cross-platform runtime and Windows/Linux shipping - #131

Draft
brxs wants to merge 89 commits into
mainfrom
codex/107-integration
Draft

Add cross-platform runtime and Windows/Linux shipping#131
brxs wants to merge 89 commits into
mainfrom
codex/107-integration

Conversation

@brxs

@brxs brxs commented Aug 9, 2026

Copy link
Copy Markdown
Collaborator

What changed

This branch combines the reviewed implementation package for running and shipping LSDJ on macOS, Linux, and Windows:

  • native cross-platform paths, environment isolation, process-tree supervision, and authenticated local service transport;
  • a pinned external PyTorch MRT2 dependency for managed Linux/Windows runtimes, with one shared two-deck worker and serialized CUDA ownership;
  • the official Stable Audio 3 TFLite baseline, plus a fail-closed optional Windows CUDA candidate that remains disabled for release;
  • hash-verified per-service runtime installation, recovery-safe promotion, provenance, LoRA management, cancellation, and diagnostics;
  • live audio-output health, route retention, reconnect, serialized device transitions, and atomic combined master/cue ring swaps;
  • Linux AppImage and Windows NSIS packaging, an exact macOS/Linux/Windows producer union, and platform diagnostics;
  • hardened Windows install/upgrade/downgrade/uninstall behavior, including owned-root validation, reparse-point defenses, fail-closed destructive purge, Unicode/space paths, and documented NSIS worker exit semantics;
  • fail-closed AppImage layout auditing, release CODEOWNERS coverage, and hosted tests for runtime/GPU/render/path safety contracts.

Architecture decisions

  • MRT2 remains an external, pinned, credited dependency; LSDJ does not vendor or maintain a fork.
  • Managed Linux/Windows builds resolve separate MRT2 and SA3 services. Bundled macOS behavior remains supported.
  • Release builds do not fall back to arbitrary system Python or an obsolete single-backend launcher.
  • SA3 CUDA is explicitly not release-qualified. MRT2 currently holds its GPU lease for the worker lifetime, so safe concurrent SA3 CUDA requires a future two-level broker and hardware qualification. The public release path stays on TFLite.

Validation

Local combined head:

  • frontend: 672 tests, lint, type-check, and production build;
  • Rust: 440 default tests and 452 managed-runtime tests; default and managed Clippy; managed release build;
  • backend shared groups: 178 portable-service tests, 26 model-free deck tests, and 157 model/runtime/GPU/render/path safety tests;
  • release and packaging: 40 contracts, Ruff formatting/lint, exact three-producer policy, AppImage adversarial audit, and Windows lifecycle contracts;
  • independent whole-branch architecture/security review: approved with no remaining code findings.

Hosted validation passed for exact head 01ccaca637accca82210e6ed86b72c24e1fd06c7 in CI run 31334707264:

  • shared checks on macOS, Ubuntu, and Windows;
  • managed-runtime tests, lint, and release build on Windows;
  • Ubuntu 22.04 AppImage build, adversarial audit, Unicode/space-path smoke, and evidence upload;
  • both Windows installers, fail-closed unsigned-release verification, install/upgrade/downgrade/uninstall lifecycle, and artifact upload.

The focused Windows source branch also passed its independent three-platform and installer-lifecycle run at exact head 39a480b2d512c6883ef2d3afa6d0f62f684434d4: CI run 31333418976.

Remaining external gates

Related to #107, #108, #109, #110, #111, #112, #113, and #114.

brxs added 30 commits August 8, 2026 14:12
@brxs brxs changed the title Add the cross-platform runtime foundation and portable model backends Add cross-platform runtime and Windows/Linux shipping Aug 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant