@sqlsmith/vite-plugin

Vite plugin for SQLsmith - automatic SQL schema merging during development.

Installation

npm install -D @sqlsmith/vite-plugin

Supports Vite 4, 5, and 6; the integration suite runs against Vite 6.4.3.

Usage

// vite.config.ts
import { defineConfig } from 'vite'
import { sqlsmith } from '@sqlsmith/vite-plugin'

export default defineConfig({
  plugins: [
    sqlsmith({
      input: './src/schemas',     // Directory with SQL files
      output: './src/schema.sql', // Output merged file
      dialect: 'postgresql',      // Optional: SQL dialect
      defaultSchema: 'public',    // Optional: unqualified relation schema
      logLevel: 'info'            // silent | error | warn | info | debug
    })
  ]
})

Options

• input (required): Directory containing SQL files
• output (required): Output file path for merged schema
• dialect (optional): SQL dialect from core's SUPPORTED_DIALECTS registry (default: postgresql)
• defaultSchema (optional): Schema assigned to unqualified relation names (PostgreSQL default: public)
• watch (optional): Enable file watching (default: auto-detected based on dev/build mode)
• logLevel (optional): Shared SQLsmith log level: silent, error, warn, info, or debug (default: info)
• allowExternalReferences (optional): Keep references to relations outside the input and emit diagnostics instead of failing

silent suppresses SQLsmith logs only; schema generation and Vite error reporting remain active. The former normal/verbose aliases were removed in favor of the shared core levels.

Debug output and warning deduplication

At logLevel: 'debug' every generation additionally reports the discovered SQL files, the dependency graph and the recommended execution order — the same renderers the CLI info command uses.

At warn and info, diagnostics (external references, raw passthrough statements, cross-file raw references) are deduplicated across watch rebuilds:

• the first build prints every diagnostic in full;
• subsequent rebuilds print only new or changed diagnostics and collapse the rest into a single line: SQLsmith: N known warning(s) — set logLevel: 'debug' to see all;
• a diagnostic that disappears and later returns is printed in full again;
• a failed rebuild keeps the known set of the last successful generation, matching the preserved last-good output;
• debug disables deduplication and always prints everything.

The known set lives in plugin memory, so restarting the dev server prints all diagnostics in full again.

Runtime SET search_path statements are preserved but are not interpreted by dependency analysis; configure defaultSchema to match the effective schema.

Discovery and watch lifecycle

• Discovery is recursive and uses the same MergePlan pipeline as core.
• The input root and every discovered SQL file are registered with Vite.
• Create, update, and delete events trigger one full rediscovery and generation.
• A generated output inside the input directory is excluded from discovery and watching, preventing self-merge loops.
• Output is written atomically. Invalid SQL leaves the last valid output unchanged; production builds fail, while development reports through Vite.

Features

• ✅ File watching - Recursive regeneration on create, update, and delete
• ✅ Build integration - Schema validation during build process
• ✅ Atomic output - Invalid updates preserve the last valid schema
• ✅ Error handling - Build failures propagate; dev failures use Vite reporting
• ✅ Zero config - Works out of the box with sensible defaults

License

MIT