pdf-to-markdown/test/Files.test.ts
Johannes Zillmann 55ae236928 Improve header detection
- fix tests
- still run header detection based on heights even if TOC headlines have been identified
2024-03-28 11:39:34 -06:00

256 lines
8.7 KiB
TypeScript

import { toMatchFile } from 'jest-file-snapshot';
import * as pdfjs from 'pdfjs-dist/es5/build/pdf';
import * as fs from 'fs';
import * as path from 'path';
import * as http from 'http';
import * as https from 'https';
import PdfParser from 'src/PdfParser';
import PdfPipeline from 'src/PdfPipeline';
import { transformers } from 'src/index';
import Debugger from 'src/Debugger';
import Globals from 'src/Globals';
import Item from 'src/Item';
import { Change } from 'src/debug/ChangeIndex';
import StageResult from 'src/debug/StageResult';
import EvaluationIndex from 'src/debug/EvaluationIndex';
import RemoveRepetitiveItems from 'src/transformer/RemoveRepetitiveItems';
import DetectToc, { TOC_GLOBAL } from 'src/transformer/DetectToc';
import DetectHeaders from 'src/transformer/DetectHeaders';
import TOC from 'src/TOC';
import { getText } from 'src/support/items';
pdfjs.GlobalWorkerOptions.workerSrc = `pdfjs-dist/es5/build/pdf.worker.min.js`;
const parser = new PdfParser(pdfjs);
const pipeline = new PdfPipeline(parser, transformers);
const folder = './examples';
const files = fs.readdirSync(folder).filter((file) => file.endsWith('.pdf'));
const urls = [
'https://homepages.cwi.nl/~lex/files/dict.pdf',
'https://github.com/mozilla/pdf.js/raw/master/web/compressed.tracemonkey-pldi-09.pdf',
];
const downloadCache = 'node_modules/.cache/files';
expect.extend({ toMatchFile });
// Test is for debugging purpose
test('Debug', async () => {
// const data = fs.readFileSync(`${folder}/Adventures-Of-Sherlock-Holmes.pdf`);
const data = fs.readFileSync(`${folder}/ExamplePdf.pdf`);
await pipeline.parse(data, () => {});
});
describe.each(files)('Test %p', (file) => {
const data = fs.readFileSync(`${folder}/${file}`);
let debug: Debugger;
const printedGlobals = new Set<string>();
beforeAll(async () => {
debug = await pipeline.parse(data, () => {}).then((pc) => pc.debug());
});
test.each(transformers.map((t) => t.name).filter((name) => name !== 'Does nothing'))(
'stage %p',
(transformerName) => {
const stageResult = debug.stageResult(debug.stageNames.indexOf(transformerName));
const chunkedLines: string[][] = [[]];
let resultIndex = 0;
let collectedItems = 0;
stageResult.selectPages(true, true).forEach((page) => {
page.itemGroups.forEach((itemGroup) => {
const item = itemGroup.top;
const change = stageResult.changes.change(item);
if (change || stageResult.descriptor.debug?.showAll) {
chunkedLines[resultIndex].push(itemToString(debug.fontMap, item, change, stageResult.evaluations));
collectedItems++;
}
});
// we split results to multiple files to circumvent githubs 100MB limit
resultIndex = Math.floor(collectedItems / 450_000);
if (resultIndex === chunkedLines.length) {
chunkedLines.push([]);
}
});
// Global characteristics
chunkedLines[0].unshift(toHeader(stageResult, printedGlobals));
try {
chunkedLines.forEach((lines, idx) => {
const transformerResultAsString = lines.join('\n') || '{}';
expect(transformerResultAsString).toMatchFile(matchFilePath(file, transformerName, chunkedLines.length, idx));
});
} finally {
stageResult.globals.keys().forEach((globalKey) => {
printedGlobals.add(globalKey);
});
}
},
);
});
function matchFilePath(pdfFileName: string, transformerName: string, chunkCount = 1, chunkIndex = 0): string {
const pdfFileNameWithoutExtension = pdfFileName.substr(0, pdfFileName.length - 4);
const resultFileName = `${transformerName[0].toLowerCase() + transformerName.slice(1).replace(/\s/g, '')}`;
const fileIndex = chunkCount > 1 ? `.${chunkIndex}` : '';
return `${folder}/${pdfFileNameWithoutExtension}/${resultFileName}${fileIndex}.json`;
}
describe('Selective transforms on URL PDFs', () => {
const transformerNames = [new RemoveRepetitiveItems().name, new DetectToc().name, new DetectHeaders().name];
test.each(urls)('URL %p', async (url) => {
const { fileName, data } = download(url);
const debug = await pipeline.parse(data, () => {}).then((pc) => pc.debug());
const printedGlobals = new Set<string>();
transformerNames.forEach((transformerName) => {
const stageResult = debug.stageResult(debug.stageNames.indexOf(transformerName));
const pages = stageResult.selectPages(true, true);
const lines: string[] = [];
lines.push(toHeader(stageResult, printedGlobals));
pages.forEach((page) =>
page.itemGroups.forEach((itemGroup) => {
const item = itemGroup.top;
const change = stageResult.changes.change(item);
if (change) {
lines.push(itemToString(debug.fontMap, item, change, stageResult.evaluations));
}
}),
);
const transformerResultAsString = lines.join('\n') || '{}';
expect(transformerResultAsString).toMatchFile(matchFilePath(fileName, transformerName));
stageResult.globals.keys().forEach((globalKey) => {
printedGlobals.add(globalKey);
});
});
});
});
function toHeader(stageResult: StageResult, alreadyPrintedGlobals: Set<string>): string {
const groupedItemCount = stageResult
.selectPages(false, true)
.reduce((itemCount, page) => itemCount + page.itemGroups.length, 0);
return JSON.stringify(
{
pages: stageResult.pages.length,
items: stageResult.itemsUnpacked().length,
groupedItems: groupedItemCount,
changes: stageResult.changes.changeCount(),
schema: stageResult.schema,
globals: globalsToString(stageResult.globals, alreadyPrintedGlobals),
// messages: stageResults.messages,
},
null,
2,
);
}
function globalsToString(globals: Globals, alreadyPrintedGlobals: Set<string>): object {
return Array.from(globals.map)
.filter(([key]) => !alreadyPrintedGlobals.has(key))
.reduce((obj, [key, value]) => {
if (key === TOC_GLOBAL.key) {
const toc: TOC = value as TOC;
value = {
...toc,
tocHeadlineItems: toc.tocHeadlineItems.map((item) => ({
page: item.page,
text: getText(item),
})),
};
}
obj[key] = value;
return obj;
}, {});
}
function itemToString(
fontMap: Map<string, object>,
item: Item,
change: Change | undefined,
evaluationIndex: EvaluationIndex,
): string {
const fontName: string | Array<string> = item.data['fontName'];
let newFontName: string | Array<string> | undefined = undefined;
if (fontName) {
if (typeof fontName === 'string') {
newFontName = fontMap.get(fontName)?.['name'];
} else {
newFontName = fontName.map((name) => fontMap.get(name)?.['name']);
}
}
const changeType = change?.constructor.name || 'none';
const transform: undefined | number[] = item.data['transform'];
let newTransform: undefined | string[];
if (transform) {
newTransform = transform.map((num) => num.toFixed(2));
}
const object = {
page: item.page,
change: changeType,
score: evaluationIndex.evaluationScore(item),
types: item.data['types'],
...item.data,
fontName: newFontName,
height: item.data['height'].toFixed(2),
width: item.data['width'].toFixed(2),
transform: newTransform,
};
if (!evaluationIndex.hasScores()) {
delete object.score;
}
return JSON.stringify(object);
}
function download(url: string): { fileName: string; data: Buffer } {
const fileName = path.basename(new URL(url).pathname);
const localFilePath = `${downloadCache}/${fileName}`;
if (!fs.existsSync(localFilePath)) {
fs.mkdirSync(downloadCache, { recursive: true });
downloadToFile(url, localFilePath);
}
return {
fileName,
data: fs.readFileSync(localFilePath),
};
}
function downloadToFile(url: string, dest: string): Promise<void> {
const uri = new URL(url);
const pkg = url.toLowerCase().startsWith('https:') ? https : http;
return new Promise((resolve, reject) => {
pkg.get(uri.href).on('response', (res) => {
if (res.statusCode === 200) {
const file = fs.createWriteStream(dest, { flags: 'wx' });
res
.on('end', () => {
file.end();
resolve();
})
.on('error', (err) => {
file.destroy();
fs.unlink(dest, () => reject(err));
})
.pipe(file);
} else if (res.statusCode === 302 || res.statusCode === 301) {
// Recursively follow redirects, only a 200 will resolve.
if (res.headers.location) {
downloadToFile(res.headers.location, dest).then(() => resolve());
}
} else {
reject(new Error(`Download request failed, response status: ${res.statusCode} ${res.statusMessage}`));
}
});
});
}