Supprocom.NativeAllocationManagement gives C# code explicit ownership of native
storage. NativePool<T> reuses typed slabs. NativeRegion provides one lexical
heterogeneous lifetime. NativeArena provides reusable heterogeneous storage.
NativeBuilder<T> writes growable unmanaged sequences without managed intermediate
arrays. NativeTransfer<T> moves heap-storable ownership across thread boundaries.
NativeWorkspace<T> reuses one fixed typed block in a worker hot loop.
Each bounded operation checks the state required by its allocator contract. The bundled Roslyn analyzer checks ownership and bounded-view rules in consumer source. The package targets .NET 10.
Install the released 0.2.3 package. This main
checkout contains unreleased 0.3.0 development, including budgets, prepared
execution, typed layouts and unique/shared/weak ownership. Use the matching
release-tag documentation for 0.2.3 applications. See the migration guide
when moving to 0.3.0; the value-based NativeTransfer<T> requires rebuilding
consumers.
Build the source with .NET SDK 10.0.302 and C# 13. The repository pins the SDK without feature-band roll-forward.
See demo process/GC observations for their exact units, sources, lifetimes and availability; cold telemetry is not an allocator budget.
The getting-started guide contains installation instructions, complete examples, lifecycle rules, analyzer diagnostics, and cleanup requirements.
The guide covers typed pools, fixed workspaces, lexical regions, reusable arenas, growable builders, cross-thread transfers, scoped recycling, statistics, and trimming. The demo kernel-metrics contract distinguishes actual cgroup sources and units from native budgets and unavailable observations.
Historical voxel runs estimated a 50 to 85 percent performance improvement over expert safe C# in non-memory-constrained control profiles.
Very memory-constrained profiles estimate 90 to 130 percent. Selected higher-turnover runs exceed 150 percent.
Both implementations process equal inputs and outputs. The expert safe C# path uses pooling, exact sizing, bounded retention, and proactive memory admission.
The matrix treats constrained-memory qualification as information. It verifies equal binary limits, no swap, and cumulative demand. It does not require garbage collection or a resident-memory threshold.
Results depend on workload, memory limits and runtime configuration. The voxel pipeline guide contains the method and runnable commands. For known-size and growing output, use the builder benchmark. Compare elapsed time and total managed/native allocation for your workload.
This project uses the GNU Affero General Public License, version 3 only. See LICENSE for the complete terms. See NOTICE for the project notice and source offer.
Use the complete diagnostic index, verified cleanup fixes, and managed-first allocator-selection guide. Every shipped rule has its own help page. Compiler analysis remains required; IDE fixes do not disable ownership checks or change a memory budget.