archetech / archon / 33277002911

92%

main: 87%

Pull #963

github

refactor(db): Replace both sqlite drivers with node:sqlite

Six manifests depended on `sqlite3`, split across two majors that nobody
chose -- each service took whatever `npm install` resolved on the day it
was added, either side of 6.0.1's release, and dependabot later moved one
straggler and left three. Herald selects `sqlite` too, but on
`better-sqlite3`, so the repo carried two different native sqlite drivers.

`node:sqlite` removes the question rather than answering it: there is no
driver left to pick a version for. The API the adapters need is all
present -- multi-statement `exec`, `BEGIN IMMEDIATE`, `changes` as a plain
number -- and every bind site was already a JSON string or a small number.

Herald's adapter was the proof this fits: it used better-sqlite3
synchronously through exactly prepare/exec/run/get/all/close, which is
node:sqlite's surface, so a public HTTP service has been serving from a
synchronous sqlite driver all along.

What this deletes: seven dependency entries, the sqlite3 prebuild seeding
and its `local_prebuilds` env var across 24 Dockerfiles, and -- for
Herald's tree -- node-gyp, prebuild-install and tar, the chain
SECURITY_AUDIT.md flags for a critical advisory. It also removes the
unseeded-prebuild defect Herald had: better-sqlite3 was fetched over the
network with no retry, into an image with no toolchain to fall back on.

The trade is that the SQLite engine now ships with Node instead of being
pinned in a lockfile. Every image and CI job pins one exact Node version,
so it can only change through a visible bump; tests/ci/sqlite-runtime.test.ts
holds that arrangement in place and floors the engine version.

The two published packages now declare `engines: node >=22.5.0`, which
they need for node:sqlite and previously declared not at all.

Closes #962

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GYGMfX3foBcVqP7AQNXjWj

3881 of 4464 branches covered (86.94%)

Branch coverage included in aggregate %.

48 of 49 new or added lines in 2 files covered. (97.96%)

8528 of 8953 relevant lines covered (95.25%)

717.55 hits per line