Repository navigation
bench: cover python FFI costs for core #776 P16 - #17
Conversation
Exercise prepared execute, cursor batch sizes, set round trips and host callbacks with a deterministic fixture and untimed test mode. Co-Authored-By: Codex <noreply@openai.com>
…s #776) Apply verified read-only review findings; all cases rechecked once with untimed test mode. Co-Authored-By: Codex <noreply@openai.com>
|
Phase-A untimed evidence after review fixes: target/bench-venv/bin/python bench/nightly.py --testEach ID ran once in this check. The same cases also passed against the P16 core source using a temporary Cargo patch, without changing the release dependency pin. No local/remote timing, SSH or benchmark workflow dispatch was performed. Read-only review: Route: core #832/#833 |
Adds #776 P16 G15 representatives through the actual python runtime, with no core behavior or production dependency changes.
New IDs:
binding/python/{prepared_execute/1k,cursor_batch_1/1k,cursor_batch_256/1k,set_roundtrip/2,predicate/1k,aggregate/1k}.Setup/preparation/registration happen outside measurement; the fixed 1k-fact fixture checks point/set results and drain/callback counts and sums.
--testruns each ID exactly once without timers or metrics files. Host decoding and cursor close are included in the round trip.Route: core #832 and identical main companion #833,
bench / binding_pythonmatrix job, entire fixed shortlist. Dedicatedbinding-python-ubuntu24-v1Bencher testbed; 5–15 job-minutes plus build/runtime setup provisionally reserved, unverified. Land binding code before those core routes. Phase B can selectbinding-ref=bench/776-p16-bindingsbefore merging.Validation: local release build and untimed
--testpassed, also with a temporary Cargo patch to the P16 core checkout. No timed local run, remote benchmark-machine access, or Actions benchmark dispatch. Normal CI and the read-only Codex review are tracked in follow-up comments. Manual Java/Android/Swift/WASI and other follow-up dispositions live in the core guide. No CHANGELOG change, matching the core benchmark PR precedent.Refs #776; source issue: project-minigraf/minigraf#776.
🤖 Generated with Codex, orchestrated by Claude Code