Heads-up: SchemaBot's core will become database-agnostic (PostgreSQL support planned)

[Kiran01bm]

What

A heads-up for anyone following or building on SchemaBot: we intend to make the
codebase database-agnostic, so new database families — PostgreSQL first —
can be added as implementations rather than rewrites.

We're posting this for transparency: this is a direction we've decided on,
not an open question. There's no near-term behavior change, and existing
MySQL/Vitess deployments are unaffected — the point is to set expectations now.

Two independent dimensions

The MySQL coupling has two dimensions that are often conflated. They're tracked
together but ship separately:

• Target databases (the databases SchemaBot runs schema changes against).
  This is the user-visible goal. The engine layer is already pluggable (Spirit
  for MySQL, PlanetScale for Vitess, and a Postgres engine stub already exists);
  the remaining work is DDL parsing, dialect assumptions, and the Postgres
  online-DDL mechanism.
• SchemaBot's own state store (plans, applies, tasks, checks, leases, etc.).
  This lives behind a clean storage interface today with a single MySQL
  implementation. Making it dialect-agnostic is independent of target support
  and can trail it.

Guiding principles

• Second implementation before abstraction polish — the interfaces already
  exist; we won't redesign them until a real Postgres implementation exposes
  pressure points.
• Capability gates, not lowest-common-denominator — MySQL-only features stay
  available for MySQL and are gated by engine capability interfaces, not removed.
• No behavior change for existing MySQL/Vitess deployments at any merge
  point — every step is dormant or dialect-routed.

The hard part

The Postgres online-DDL mechanism is the long pole — it shapes the engine's
cutover/revert state machine and needs its own mini-design before implementation.
Advisory-lock semantics also differ enough from MySQL's named locks to warrant a
deliberate design rather than a direct translation.

Scope

• PostgreSQL is the first new target; the architecture is meant to generalize
  beyond it.
• MySQL and Vitess remain first-class, fully supported.