adapter/model-to-table

Purpose

Memoising mapper from Better-Auth model name to SurrealDB table name. The mapper function is invoked once per distinct model and the resolved name is validated against the SurrealQL identifier rule (/^[a-zA-Z_][a-zA-Z0-9_]*$/). Caching avoids re-running operator-supplied prefix logic on every call site while validation prevents accidental injection through a hand-rolled modelToTable that returns SurrealQL syntax.

Features

Identity-default — new ModelToTableCache() maps model→model verbatim.
Caches resolved name in a Map<string,string>.
Identifier-rule validation: invalid names raise INVALID_MODEL on first use of the offending model.
Decoupled from the adapter shell so unit tests can exercise mapping semantics in isolation.

Usage: Basic

import { ModelToTableCache } from "@suaveplan/better-auth-surrealdb/adapter";

const cache = new ModelToTableCache();
cache.resolve("user"); // "user"
cache.resolve("session"); // "session"

Usage: Advanced

import { ModelToTableCache } from "@suaveplan/better-auth-surrealdb/adapter";

let calls = 0;
const cache = new ModelToTableCache((m) => {
    calls++;
    return `bauth_${m}`;
});

cache.resolve("user"); // "bauth_user" — calls === 1
cache.resolve("user"); // "bauth_user" — calls still 1 (cached)
cache.resolve("session"); // "bauth_session" — calls === 2

API Reference

class ModelToTableCache
- constructor(resolver?: ModelToTableResolver) — defaults to identity.
- resolve(model: string): string — cached + validated.
type ModelToTableResolver = (model: string) => string.

Implementation Notes

The cache is a plain Map because invalidation is unnecessary: model names are stable Better-Auth contracts and the adapter is constructed per process. Throwing on first use of an invalid name (rather than at adapter construction) preserves the "lazy" semantics so callers wiring a multi-tenant prefix function can return different names per request without paying the validation cost upfront. The identifier check uses the same regex as every other validator in this package (/^[a-zA-Z_][a-zA-Z0-9_]*$/) so operators get consistent feedback when a name shape is wrong — whether the offending input is a model alias, a SurrealDB table name, a link-field target, or a select projection entry.

The class field is protected rather than private so subclasses can introspect the cache during debugging without paying the indirection cost of an additional accessor method — useful when wiring per-tenant audit trails that need to log every resolved table name on first use.

2.6 KiB Raw Blame History