better-sqlite3-skilld

name	better-sqlite3-skilld
description	ALWAYS use when writing code importing "better-sqlite3". Consult for debugging, best practices, or modifying better-sqlite3, better sqlite3.
metadata	{"version":"12.11.1","generated_by":"Anthropic · Haiku 4.5","generated_at":"2026-06-15T00:00:00.000Z"}

WiseLibs/better-sqlite3 better-sqlite3@12.11.1

Tags: latest: 12.11.1

References: package.json • README • Docs • Issues • Discussions • Releases

Search

Use skilld search "query" -p better-sqlite3 instead of grepping .skilld/ directories. Run skilld search --guide -p better-sqlite3 for full syntax, filters, and operators.

API Changes

This section documents version-specific API changes — prioritize recent major/minor releases.

better-sqlite3 v12.11.1 maintains API stability. No breaking, deprecated, or renamed APIs were found in the v12.x release history. Recent releases focus on infrastructure updates: bundled SQLite version upgrades (v3.50.2 → v3.53.2), prebuilt binary support for new Node.js and Electron versions, and internal V8 compatibility fixes.

SQLite capability update: v12.10.0+ includes SQLite compiled with SQLITE_ENABLE_PERCENTILE, enabling window aggregate functions (PERCENT_RANK, CUME_DIST, NTILE) for analytical queries source.

Node.js support: v12.8.0+ requires Node.js v20 or later source.

No user-facing API migrations required for upgrading within v12.x.

Best Practices

• Use .transaction() for grouped operations rather than manual BEGIN/COMMIT — automatic rollback on error, nested savepoint support, and deferred/immediate/exclusive variants prevent deadlocks under concurrent load source

• Enable WAL mode on database open to dramatically improve concurrent read/write performance in multi-process or multi-thread scenarios — default journal mode has poor concurrent throughput source

• Create a single prepared statement per distinct dynamic column/table and reuse it — SQLite only binds data values, not identifiers, so prepared statements with dynamic columns must be predetermined and selected at runtime source

• Prefer .iterate() when processing large result sets to stream results one row at a time, avoiding memory bloat from loading entire result arrays into RAM source

• Use json_each() for dynamic IN clauses instead of generating parameter placeholders — avoids SQLite's hardcoded parameter limits and allows reusing prepared statements across variable-length lists source

• Apply INTEGER PRIMARY KEY AUTOINCREMENT to ID columns to reuse SQLite's native rowid and prevent ID reuse after deletion source

• Call .safeIntegers() on statements that return 64-bit integers to receive BigInt instead of lossy JavaScript numbers — database state is corrupted silently if large IDs overflow source

• Use .raw() mode with .columns() when retrieving very high row counts for performance — arrays are faster to construct than objects, and column metadata is recovered separately source

• Register user-defined functions deterministically where possible — set options.deterministic = true to let SQLite cache results and optimise execution source

• Store dates as millisecond timestamps (JavaScript convention) or ISO-8601 strings, not seconds — SQLite has no native date type and its date functions favour string formats; choose consistency with your application's time representation source

• Enable PRAGMA foreign_keys = ON explicitly — foreign key constraints are disabled by default in SQLite for backwards compatibility, even after declaration source

• Avoid mixing manual transactions with .transaction() — never run raw COMMIT or ROLLBACK inside a transaction function, as SQLite can rollback silently on ON CONFLICT or certain errors and subsequent statements will execute outside the transaction source

• Use .bind() to permanently bind parameters when executing the same statement many times with identical values — improves performance by skipping parameter rebinding on each execution source

• Configure timeout appropriately (default 5000ms) when opening the database in multi-process scenarios — prevents long hangs on SQLITE_BUSY errors when other processes hold locks source

• Use virtual tables (.table()) for computed, read-only result sets like filesystem views or regex matches — computed dynamically without storing, and support table-valued functions with parameters source

• Defer long-running queries to worker threads to avoid blocking the main event loop — use the pool pattern from the docs with message-passing to SQLite instances in dedicated threads source

• Call .close() on database handles before process exit to flush WAL checkpoints and release file locks — prevents data loss and allows clean reconnection on restart source