Summary
Remove the redundant range read in verified hydration: memory.hydrate_memory_detailed reads the byte range (src/hotmem/memory.py:244), then provenance.verify_range (src/hotmem/provenance.py:103) re-reads the same range to hash it.
Measured impact (spike #48, PR #84, B1)
Single-read hashing of the already-held bytes is ~1.16–1.65x faster than the current double-read path across range sizes (+16% at 100 MB). Both reads hit the page cache, so the cost is copy+syscalls, not I/O — still pure overhead.
Proposal
Hash the bytes already read in hydrate (or add a verify API that accepts in-memory bytes), keeping the same ProvenanceError semantics on mismatch.
Acceptance criteria
- Verified hydration reads each range once.
- Parity: digests identical to current path (spike B1 parity arms are reusable as fixtures).
- Benchmarks in
bench/native_spike/results.json show the improvement; no API/behavior change otherwise.
Context
Spike recommendation follow-up #1: bench/native_spike/README.md. Related: #48, #38, #40.
Summary
Remove the redundant range read in verified hydration:
memory.hydrate_memory_detailedreads the byte range (src/hotmem/memory.py:244), thenprovenance.verify_range(src/hotmem/provenance.py:103) re-reads the same range to hash it.Measured impact (spike #48, PR #84, B1)
Single-read hashing of the already-held bytes is ~1.16–1.65x faster than the current double-read path across range sizes (+16% at 100 MB). Both reads hit the page cache, so the cost is copy+syscalls, not I/O — still pure overhead.
Proposal
Hash the bytes already read in hydrate (or add a verify API that accepts in-memory bytes), keeping the same
ProvenanceErrorsemantics on mismatch.Acceptance criteria
bench/native_spike/results.jsonshow the improvement; no API/behavior change otherwise.Context
Spike recommendation follow-up #1:
bench/native_spike/README.md. Related: #48, #38, #40.