YREA-SLS/docs/memory_management.md

88 lines
2.8 KiB
Markdown

## Appendix F: Memory Management (Future)
The language specification currently does not include heap memory management. This appendix documents potential future approaches.
**Current State**: All values are stack-allocated or embedded in data structures.
**Pointer Types**: The `ptr` type for raw pointers is planned as a future feature and will be part of the memory management system.
**Potential Approaches**:
**Option A: Manual Management**
```
// Allocate on heap
3.0 4.0 Point alloc // ( Point -- ptr<Point> )
// Dereference
ptr deref // ( ptr<T> -- T )
// Store through pointer
new_value ptr store // ( T ptr<T> -- )
// Free memory
ptr free // ( ptr<T> -- )
```
Pros: Full control, predictable, zero overhead
Cons: Error-prone, requires discipline, potential memory leaks
**Option B: Reference Counting**
```
// Create reference-counted value
3.0 4.0 Point rc // ( Point -- rc<Point> )
// Automatic reference counting
value dup // Increments count
drop // Decrements count, frees if zero
```
Pros: Automatic cleanup, relatively simple
Cons: Runtime overhead, cannot handle cycles, larger memory footprint
**Option C: Ownership System (Rust-like)**
```
// Linear types - each value has one owner
value // Move semantics by default
value dup // Error: cannot copy owned value
value ::clone call // Explicit clone required
```
Pros: Memory safe, zero overhead, prevents leaks
Cons: Complex type system, restricts stack operations, steep learning curve
**Option D: Arena/Region-Based**
```
// Create arena
::arena new // ( -- arena )
// Allocate in arena
arena 3.0 4.0 Point alloc_in // ( arena Point -- ptr<Point> )
// Free entire arena
arena free_arena // ( arena -- )
```
Pros: Fast allocation, simple bulk deallocation
Cons: Less granular control, memory held until arena freed
**Recommendation**: Start without heap allocation (current approach). When needed, implement Option A (manual) for simplicity, with Option D (arenas) added later for performance-critical code. The stack-based nature makes ownership tracking (Option C) particularly challenging.
**Type Parameter Enforcement Enhancement**:
**Current State**: Type parameters in generic functions are currently suggestions and are not enforced at parse time.
**Example**:
```
(T -- T) { dup * } ::square fn // Currently no error even without Multiplyable constraint
```
**Future Enhancement**: The compiler could enforce that type parameters actually constrain how operators and functions act, validated at parse time:
```
(Multiplyable -- Multiplyable) { dup * } ::square fn // Enforced constraint
```
This would provide stronger type safety but add complexity to the type checker.
---