Technical Overview of Baidu Reading/Wenku Cross‑Platform Layout Engine

This article introduces the cross‑platform layout engine developed by Baidu Reading/Wenku, aiming to provide professional book‑level typesetting for mobile reading applications.

Background : Baidu Wenku is an online document‑sharing platform. Since 2013 Baidu launched the Baidu Reading and Baidu Wenku apps to meet mobile reading needs. The core of these apps is a custom layout engine that enables consistent, high‑quality rendering on both Android and iOS.

Technical Selection : The engine was designed with four main goals: (1) a single engine for Android and iOS to ensure identical display and improve development efficiency; (2) compliance with book‑printing layout standards for professional typesetting; (3) support for personalized and complex layout effects with rapid iteration; (4) a native‑experience engine with a minimal package size.

Architecture : The overall architecture consists of four modules (see Figure 1):

Control/Interface : receives external content, triggers layout, and returns layout commands to the rendering layer.

Parser : converts various source formats (epub, txt, docx, custom formats) into a unified abstract DOM tree.

Layout : traverses the DOM tree, applies layout rules, and produces a Layout Description Format (LDF) describing page composition.

LDF : a memory‑resident data structure that stores the layout result for fast paging, background pre‑layout, and interactive operations.

The abstract DOM tree (Figure 2) is a simplified version of the OOXML structure, featuring nodes such as CDoc , CBox , and CDoc that can be recursively nested to represent complex objects.

Main Techniques :

Basic Layout : paragraph layout (handling spacing, indentation, hanging), line layout (spacing, alignment), fragment layout (handling inline objects like images or formulas), and object layout (character and structural layout).

Reuse and Abstraction : all object types (text, formulas, pinyin, block structures) are abstracted into a common node model, enabling recursive processing and code reuse.

Relative Layout : sub‑objects are laid out using a local origin (0,0); after layout, a global coordinate correction aligns them within the page, allowing complex nested structures such as matrices of formulas.

Multi‑Direction Layout : supports horizontal, vertical, and reversed directions for different languages (e.g., Chinese horizontal, traditional vertical, Mongolian vertical, Uyghur reversed). Reversed layouts are achieved by mirroring coordinates after a normal layout pass (see Figure 4).

Summary and Outlook : Future work includes precise, content‑driven layout for specific document types (e.g., journals with distinct abstract and keyword styles), adaptive image placement for large‑screen versus small‑screen devices, and intelligent layout that understands both content and device characteristics. The Baidu Reading/Wenku engine demonstrates that a custom layout engine can combine general‑purpose techniques with domain‑specific optimizations to achieve high‑quality, cross‑platform document rendering.