mongo-go-driver Validation Report

Generated 2026-05-19 — SecantusDB 0.5.1b20 vs mongo-go-driver fd85a834c40e (vendor/mongo-go-driver/).

Run uv run python -m invoke validate-go to refresh. The pass rate is the analogue of the pymongo conformance gauge for the official Go driver — same shape, different wire-protocol pickiness. Type-strict bugs (int32 vs int64) that pymongo accepts silently fail loudly here.

Summary by package

Package	Passed	Failed	Skipped	Total	Pass rate
internal/integration	359	0	52	411	100.0%
internal/integration/unified	42	0	0	42	100.0%
Overall	401	0	52	453	100.0%

How this is generated

mongo-go-driver’s integration tests are run unmodified, against a standalone SecantusDB daemon. The submodule at vendor/mongo-go-driver/ is checked out at the pinned upstream tag with zero local edits. go_validation/runner.py spawns python -m secantus --host 127.0.0.1 --port 27018 --storage-path <tempdir> --noop-heartbeat-seconds 10 as a subprocess (a fresh tempfile.mkdtemp(prefix='secantus-go-gauge-') — never :memory:; on-disk WiredTiger keeps the checkpoint / journal code paths exercised), waits for its TCP listener, exports MONGODB_URI=mongodb://127.0.0.1:27018 (the env var internal/integtest.MongoDBURI and internal/integration/mtest read at setup), then runs go test -json -count=1 ./internal/integration/.... From the go-driver’s point of view it’s connecting to a real mongod over TCP — exactly like its CI does.

Integration-only. The pure-BSON unit tests under ./bson/... and ./mongo are out of scope for this gauge — they verify the driver’s own serialization logic without ever opening a TCP connection, and would inflate the pass count without proving anything about SecantusDB’s wire path. The pass rate above is a true measure of cross-driver compatibility with the language-canonical Go driver mongodump and mongorestore are built on.

Tests gated on topology (mtest.RequiresReplicaSet, mtest.RequiresSharded, etc.) self-skip when the server doesn’t match — those skips are honest gaps, not failures.