Patch Notes 0.1.1 and 0.1.2

Scope

• Bug fixes
• Internal refactors
• Performance and allocation improvements
• Platform-specific robustness
• Build and packaging cleanup

Existing projects built against 0.1.0 should continue to compile and run without modification. However, the new internal API around ::Await() is not included in the old main.cpp. Add that single line manually or simply create a new project from the 0.1.1 template and move your existing classes, files and resources over.

Shutdown and Lifecycle Safety

Engine shutdown has been made deterministic and race-safe across audio, graphics, node networks, and coroutines.

• Audio callbacks now terminate in a defined order
• Node processing and coroutine execution are guarded against late execution during teardown
• Graphics resources are released in a strictly ordered sequence
• Cross-domain coordination is prevented from executing after shutdown begins

These changes address rare but real issues observed during live coding, repeated start/stop cycles, and application exit under load.

macOS Stability Improvements

Several fixes address macOS-specific lifecycle and threading constraints:

• All GLFW and Cocoa interactions are dispatched on the main thread
• Lila live coding server and Clang interpreter execution are enforced on the main thread
• Vulkan and MoltenVK initialization and teardown paths are hardened
• Temporary Homebrew runtime environment handling is formalized
• Window teardown and resizing edge cases are stabilized

These changes improve reliability during live coding and repeated execution on macOS.

Performance and Allocation Cleanup

Internal refactors reduce unnecessary work and improve predictability:

• Persistent NodeContext reuse removes per-frame context allocation
• GPU sync and output nodes now use persistent graphics contexts
• Event loop dispatch is eliminated for NodeNetwork nodes where unnecessary
• Buffer management and stochastic generation are tightened

No observable behavior changes are intended.

Execution and Dispatch

• Introduction of a new parallel dispatch executor
• Cleaner separation between execution coordination and domain logic
• Improved scheduling behavior under mixed audio and graphics workloads

Build and Packaging

• Windows build script and versioning fixes
• Debian and Ubuntu packaging cleanup
• Minor cross-platform build improvements

Availability

• Platforms: Windows, macOS, Linux
• Distribution: GitHub releases, Launchpad PPA, COPR, AUR, Homebrew
• Version: 0.1.1

No migration steps are required.

Breaking Change: Explicit Temporal Buffer Semantics (std::span)

Affects:

• Polynomial lambdas (recursive and feedforward modes)
• Logic sequential mode lambdas
• Related context accessors, tests, and tutorials

What changed

• Lambda parameters previously typed as std::deque<T> are now std::span<T>
• Internal history storage is now implemented as explicit ring buffers
• Lambdas receive a linearized temporal view rather than a mutable container

Why

• Removes implicit allocation and push/pop semantics
• Makes temporal ordering explicit and deterministic
• Decouples user logic from internal storage representation
• Enables zero-allocation, real-time–safe execution
• Aligns CPU and GPU contexts under a unified data contract

This is a deliberate semantic correction, not a cosmetic refactor.

Logic & Polynomial Internals

• Introduced explicit ring buffers with linearized views
• Removed implicit buffer construction paths
• Unified context propagation across CPU and GPU execution
• Preserved observable behavior while making time semantics explicit

These changes make recursive systems, filters, and sequential logic predictable and inspectable.

Stochastic Generator Performance & Determinism

• Added fast xorshift-based uniform RNG for hot paths
• Cached normal and exponential distributions
• Introduced lazy rebuilds via dirty flags on parameter changes
• Eliminated per-sample std::distribution reconstruction
• Made set_normal_spread() safely invalidate cached distributions

These changes significantly reduce overhead in stochastic-heavy graphs while preserving statistical correctness.

Shader Pipeline Fix: Push Constant Size Calculation

Fixed

• Pipelines could be created with zero-byte push constant ranges when push constants were set via set_push_constant_data() instead of staging buffers

Resolution

• Pipeline initialization now:
  1. Uses staging buffer size when available
  2. Falls back to the processor’s internal push constant data size
• Ensures correct allocation regardless of data source

Affected

• RenderProcessor
• ComputeProcessor

This fixes a silent but critical Vulkan correctness issue.

ShaderProcessor Consistency

• set_push_constant_data() now updates push constant size when required
• Prevents mismatches between stored data and pipeline layout
• Explicitly enforces Vulkan push constant limits

Build & Portability

• Added external GLSL → SPIR-V compilation fallback when Shaderc is not enabled
• Improves portability and minimal build configurations

Documentation

• README updated to reflect correct shutdown and main-thread flow using MayaFlux::Await()