baby steps

rustdata-parallelismfork-jointype-systemconcurrency

Abstraction: Fork-join data parallelism in Rust via type-system-enforced data-race freedom

Key points:

Proposes lightweight "jobs" distinct from Rust tasks: jobs share parent stack memory and have fixed fork-join lifetimes
Two primitives: parallel::execute(closures) runs N closures in parallel; parallel::divide(slice, closure) splits a mutable slice across parallel workers
Existing borrow checker rules (&mut T aliasing prohibition) naturally prevent most data races at compile time without runtime overhead
Closure bounds fn:Isolate() and fn:Share() further restrict what values closures may capture: Isolate allows &mut T (exclusive), Share excludes &mut (multiple concurrent readers)
Hierarchy: Isolate ⊃ Share ⊃ {Freeze, Send} — new bounds fit above existing built-in bounds
Follows "three layer cake" model: message-passing actors → fork-join jobs → SIMD primitives; implementable as a library with no runtime changes

Connections: Rust Language · Data Parallelism · Type System · Concurrency