Arrow IPC

Return a read query’s result as Arrow IPC stream bytes with query_arrow, gated behind the plugin’s arrow Cargo feature, and decode it in the webview with apache-arrow.

With the plugin’s arrow Cargo feature enabled, query_arrow returns a read query’s result as Arrow IPC stream bytes — a compact, columnar, zero-JSON-parse transfer that decodes directly into an Arrow table in the webview.

query_arrow requires the arrow feature. Without it the command returns backend error: the \arrow` feature is not enabled`.

Enable the feature

# src-tauri/Cargo.toml
better-duck-tauri = { version = "0.1.0-beta.8", features = ["backend-core", "arrow"] }

The feature pulls in arrow-rs and better-duck-core/arrow, and compiles the query_arrow command in.

queryArrow

import { tableFromIPC } from 'apache-arrow'

const buf = await db.queryArrow('SELECT * FROM sales WHERE region = $1', ['eu'])
const table = tableFromIPC(new Uint8Array(buf))
console.log(table.numRows, table.schema.fields.map((f) => f.name))

• Signature: queryArrow(query: string, values?: unknown[]): Promise<ArrayBuffer>.
• values bind positionally, like select.
• Resolves to an ArrayBuffer of Arrow IPC stream bytes; decode with apache-arrow’s tableFromIPC.

How it works

Server-side the backend runs the query through better-duck-core’s Arrow C Data Interface export, then serializes each exported record batch with arrow-rs’s StreamWriter. The command returns the raw bytes via tauri::ipc::Response, so the webview receives the response body directly as an ArrayBuffer rather than a JSON-encoded array.

Prefer queryArrow over select when the result is wide or numeric and the frontend already uses Arrow (or feeds a charting / dataframe library). It avoids the DuckDB → JSON → JS round-trip and preserves exact numeric types that JSON would stringify.

Policy and permissions

• query_arrow is a read; it passes through the deny_risky_statements leading-keyword filter.
• Permission: better-duck-tauri:allow-query-arrow (not in the default set).

Like select, query_arrow forwards arbitrary read SQL, which in DuckDB can reach the filesystem through table functions. Treat it as filesystem-read-capable and constrain it the same way (read-only, connection allow-list).

See also

• Streaming — chunked JSON delivery for progressive rendering.
• Arrow & Polars — the core driver’s Arrow export this builds on.
• Feature flags — the arrow feature across the workspace.