Follow-up to #465, which built streaming Resolve ahead of the measurement the plan gated it on. Measure the cold nixpkgs-scale resolve-phase per-worker idle fraction before/after streaming, confirm it pays off, and tune MAX_RESOLVE_BATCH (or revert to batched) if it does not.
Minor cleanup to fold in while here: worker_pool/resolver.rs resolve_batch does stream.warnings.clone(); use std::mem::take instead (it returns stream afterward).
Follow-up to #465, which built streaming
Resolveahead of the measurement the plan gated it on. Measure the cold nixpkgs-scale resolve-phase per-worker idle fraction before/after streaming, confirm it pays off, and tuneMAX_RESOLVE_BATCH(or revert to batched) if it does not.Minor cleanup to fold in while here:
worker_pool/resolver.rsresolve_batchdoesstream.warnings.clone(); usestd::mem::takeinstead (it returnsstreamafterward).