@docture/testing

The shared test kit. Two jobs.

1. Conformance suites

What makes "swap pdfjs-dist for mupdf" a tested guarantee rather than an interface coincidence. Every loader and rasterizer package runs the same suite against its own class:

import { describeLoaderConformance } from "@docture/testing";
import { DocumentLoaderMuPdf } from "../src/index.js";

describeLoaderConformance({ name: "DocumentLoaderMuPdf", create: () => new DocumentLoaderMuPdf() });

It checks honest self-description, page numbering and dimensions, text/page consistency, the digital-vs-scanned verdict, multi-page documents, geometry inside the page box and sorted in reading order, on-demand rendering, an already-aborted signal, non-document bytes, and double disposal. Checks a library genuinely cannot satisfy are skipped from its declared capabilities, or explicitly via skip.

2. Synthetic fixtures

const pdf = await makeInvoicePdf({ seed: 11, lineItemCount: 45 });
pdf.bytes;      // a real PDF with a real text layer
pdf.expected;   // the ground truth that produced it

Generated rather than committed, because extraction is usually tested against documents that are confidential, which makes a corpus impossible to ship and impossible to review. A generated document is public, diffable, and its expected output cannot drift from the bytes, since the same values wrote both. seed fully determines the result, so a failure is reproducible on every machine.

textless: true produces a page of vector marks only, the stand-in for a scan.

makeRichDocumentPdf() produces a two-page conversion fixture with columns, typography, a list, table, link, code, and embedded raster image.

Also

inMemoryLoader, loadedDocument, fixedStrategy, failingStrategy are doubles for testing code that consumes docture without touching a PDF library or a model.