Skip to content

CLAUDE.md

Latest commit

 

History

History
50 lines (34 loc) · 4.03 KB

CLAUDE.md

File metadata and controls

50 lines (34 loc) · 4.03 KB
Raw

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Build

pnpm build          # Build both Tampermonkey userscript + Chrome extension
pnpm build:tamper   # Build only Tampermonkey userscript → dist/tampermonkey/
pnpm build:ext      # Build only Chrome/Edge extension → dist/extension/

Package manager is pnpm@10. Node >=22.

Architecture

This is a dual-target project that downloads PPT courseware from 学习通 (chaoxing.com) as PDF files. It produces both a Tampermonkey userscript and a Chrome/Edge extension from the same core logic.

Build system

Two tsup configs produce two IIFE bundles. Both inject the full jspdf.umd.min.js source at build time via process.env.JSPDF_SOURCE — this is needed because the PDF generation runs inside a Web Worker constructed from a Blob URL, where normal module imports aren't available.

  • tsup.tampermonkey.config.ts — bundles src/index.ts, prepends UserScript== metadata banner read from package.json.tampermonkey
  • tsup.extension.config.ts — bundles src/extension/extension-entry.ts, generates manifest.json (Manifest V3), an inline background.js service worker, copies the icon, and zips everything

Runtime architecture

Entry point (src/index.ts): XuexitongPPTDownloader class. On load it:

  1. Finds PPT iframes via findPPTIframes() (recursive search for #panView, up to 5 levels deep, plus attribute-based fallback)
  2. Mounts a ButtonGroup (download + stop buttons) into each found iframe's document
  3. Uses WeakMap<HTMLIFrameElement, Document> to detect iframe navigations (new Document object → re-add buttons)
  4. Listens for SPA route changes (popstate/hashchange + monkey-patched history.pushState/replaceState) + MutationObserver + a 10-retry timer

Download pipeline (triggered by button click):

  1. ppt-extractor.ts — reads the data attribute for the filename, waits for nested #panView iframe to load, counts li[id^="anchor"] elements for page count, derives baseUrl from the first image URL
  2. pdf-generator.ts — downloads the first image to determine page dimensions, creates a Web Worker from the serialized pdfWorkerEntry function (+ injected jsPDF source), then concurrently downloads remaining images in batches
  3. image-downloader.ts — downloads via GM_xmlhttpRequest (bypasses CORS), converts PNG→JPEG via OffscreenCanvas/createImageBitmap, reports back ArrayBuffer objects transferred to the Worker with zero-copy
  4. pdf-worker.ts — receives INIT / ADD_PAGE / FINISH messages, calls jsPDF APIs, returns the final PDF Blob via postMessage
  5. Files are saved by creating a Blob URL and clicking a temporary <a> element

Extension polyfill (src/extension/extension-polyfill.ts): In the extension build, GM_xmlhttpRequest is polyfilled using chrome.runtime.sendMessage to a background service worker that does fetch() and returns data: URLs. GM_getValue/GM_setValue are stubbed. The extension entry (extension-entry.ts) imports the polyfill before importing the main logic.

Download cancellation (src/download-controller.ts): Cooperative abort via throwIfAborted() checkpoints in the download loop. The stop button sets the flag, the loop breaks, and a DownloadAbortedError is caught.

Key constraints

  • All image downloading MUST go through GM_xmlhttpRequest (or the extension polyfill). Direct fetch()/<img> won't work because PPT images are cross-origin (hosted on cldisk.com).
  • The pdfWorkerEntry function is serialized with .toString() and executed in a Worker — it cannot reference any outer-scope variables.
  • Zero-copy transfer: ArrayBuffer objects are transferred to the Worker via the transfer list (second argument to postMessage), so they become unusable on the main thread after sending.
  • Page limit: PDFs with >500 pages (configurable in src/config.ts) trigger a split/merge prompt to avoid RangeError: Invalid string length from jsPDF's string concatenation.