🚧 Note to Visitors: This repository houses a Interaction Combinators engine, which is currently a work in progress (WIP).
This Interaction Combinators engine introduces a few optimizations worth mentioning:
- Structured concurrency: Leverages the power of the Rayon crate to implement structured concurrency, allowing us to manage task parallelism with fine-grained control and robust error handling. This means our computations are not only fast but also resilient and maintainable.
- Zero-cost erase cells: Erase cells operate without additional allocation overhead, thanks to their unboxed representation.
- Optimized connections: Self auxiliary port connections are directly established without the need for variable allocation. This optimization also extends to reductions by directly evaluating both sides as a direct equation.
- Efficient cell handling: Cells exist transiently on the execution stack, only persisting in the store when explicitly assigned to a variable during reduction.
- Recycling of pointers: Rather than deallocating, we recycle pointers to variables and cells, minimizing memory churn.
- Immutable Cells: Cells are immutable and exist on the stack, while variables are mutable and passed by reference, ensuring thread safety during reduction.
- Local Statistics Gathering: Each thread collects its own statistics to minimize contention, contributing to global statistics post-execution.
- Depth-first Reduction: We prioritize depth of work within threads to avoid extraneous async task allocations and queuing, streamlining execution.
TODOs:
- Reusable definitions
- Parser
- Tests
- Implement labels
- Add examples
- Add benchmarks
Some known issues:
- When walking linked Var's during binds or connects, the engine ends up reading the Var from the Store a second time.
I would like to express my gratitude to Victor Taelin and the Higher Order Company (HoC) for their passionate work in the field of interaction nets and interaction combinators, which introduced me to these fascinating concepts.