Large files and streaming

import { toXlsxWorkbookStream } from "@sheetwrite/core";
import "@sheetwrite/xlsx/register";

declare const destination: WritableStream<Uint8Array>;
const controller = new AbortController();
await toXlsxWorkbookStream(store.exportSnapshot(), {
  signal: controller.signal,
  onWarning: (warning) => console.warn(warning.code, warning.message),
}).pipeTo(destination);

The output contains the same XLSX parts as buffered export. ZIP entry order and container bytes can differ.

Try it live: resource-limit failures. The lab demonstrates typed limits, not a whole-file memory guarantee.

import { fromXlsxWorkbookFile, toXlsxWorkbookStream } from "@sheetwrite/core";
import "@sheetwrite/xlsx/register";

const input = document.querySelector<HTMLInputElement>('input[type="file"]');
const file = input?.files?.[0];
if (!file) throw new Error("Choose an XLSX file first");
const options = {
  onWarning: (warning: import("@sheetwrite/core").XlsxWorkbookWarning) =>
    console.warn(warning.code, warning.message),
};
const snapshot = await fromXlsxWorkbookFile(file, options);
let outputBytes = 0;
await toXlsxWorkbookStream(snapshot, options).pipeTo(new WritableStream<Uint8Array>({
  write(chunk) { outputBytes += chunk.byteLength; },
}));
console.log(outputBytes); // Total depends on the selected workbook.

Streaming functions

CSV/TSV remain core-only; XLSX requires backend registration. Existing string and Uint8Array APIs remain available.

Both buffered and Blob XLSX import retain the ZIP directory and parse each worksheet incrementally from bounded inflated chunks. Neither builds a whole-worksheet string or XML tree. Snapshot cells are built as rows stream, with one cell representation rather than a second parsed record. The snapshot still holds every imported cell: streaming is not a lazy or constant-memory workbook store.

Resource limits

Every XLSX table/workbook path shares XlsxWorkbookOptions:

Defaults are exported as DEFAULT_XLSX_RESOURCE_LIMITS:

Hard resource bounds are checked before unsafe decompression or allocation. Exceeding one rejects with XlsxResourceError, which names the resource, limit and observed value. The kept-part budget instead drops over-budget graphs with warnings. Do not raise limits merely to suppress failures; budget for the resulting snapshot as well as the codec.

Memory evidence and limits

PR #186 measured seven interleaved fresh-process rounds per side with one CPU slot:

These are measured workloads with adjusted budgets, not default-limit promises. The later results supersede the earlier #179 import peak-risk observation; they do not establish a universal memory ceiling.

See also

• XLSX import
• XLSX export
• CSV and TSV
• XlsxWorkbookOptions
• XlsxResourceLimits