shulihuazixuecongshu README

The project's README from github.com/tradecatlabs/shulihuazixuecongshu. As one markdown file: README.md.

数理化自学丛书(电子书重建版)

本仓库保存《数理化自学丛书》17 册的原始扫描 PDF、语义化 Markdown、扫描插图资源和可复现 EPUB3 构建工具。

电子书处理交流群

新建了一个电子书处理交流群,欢迎感兴趣的朋友加入,一起交流、学习和分享电子书及扫描版 PDF 的数字化处理经验:

Telegram: https://t.me/dzscl

群内会定期更新和发布扫描版 PDF 处理的数字工序、工具与相关镜像。也欢迎提供赞助 Token 或计算资源;赞助商名单会写入制作的作品,并在合适的位置进行介绍和推广。

赞助

本项目计算资源由 交易猫实验室(TradeCat Labs) 赞助。

CA: https://dexscreener.com/bsc/0x8a99b8d53eff6bc331af529af74ad267f3167777

收录书目

作者元数据统一为“数理化自学丛书编委会”,电子书语言统一为 zh-CN。

仓库结构

.
├── books/               # 电子书源文件区
│   ├── *.md             # 17 册规范正文源
│   └── assets/          # 4,875 个正文图片资源
├── raw/                 # 17 册原始扫描 PDF、导入哈希和资料说明
├── catalog.json         # 书目、顺序、语言和稳定 EPUB UUID
├── epub/                # EPUB 样式、Lua 过滤器及 Pandoc 本地化数据
├── scripts/             # 源文件审计与 EPUB3 构建工具
├── docs/                # 来源、维护与已知问题说明
├── reports/             # 动态审计报告(JSON 不纳入版本控制)
└── dist/                # 构建出的 EPUB(不纳入版本控制)

Markdown 与 books/assets/ 保持在同一源文件区,因此正文中的 assets/... 相对引用可直接解析;仓库根目录不放置单册电子书源文件。

构建

依赖:

原始 PDF 深度审计额外需要 PyPDF2 3.x、qpdf、Poppler(pdfinfo、pdfdetach、pdfsig、pdftotext)和 ExifTool。

完整审计并构建:

make all

仅审计源文件:

make audit

审计17份原始扫描 PDF 的哈希、结构、附件、主动内容和隐私元数据:

make pdf-audit

仅构建 EPUB:

make epub

验证已有 EPUB:

make verify

构建单册:

python3 scripts/build_epubs.py --book '代数(第一册)'

输出位于 dist/,报告位于 reports/。构建器会检查 ZIP、XML、元数据、内部链接、图片替代文本、MathML 数量以及 EPUB3 nav/EPUB2 NCX 同步状态。同一源文件、同一 Pandoc 版本和同一 SOURCE_DATE_EPOCH 下,输出字节可复现。

Release 发布策略

从 v2.0.0 起,GitHub Release 只发布以下三本学科合订版 EPUB,不再上传17本独立 EPUB:

17册 Markdown、原始 PDF、资源和单册构建能力继续保留在仓库中,作为合订本的可审计来源。既有 v1.0.0、v1.1.0 Release 保留,不追溯删除。

发布前隐私审计

make privacy
make pre-push

推送门禁覆盖 Git 候选文件、路径跨平台兼容性、PNG 完整性与元数据、生成 EPUB、常见秘密格式、本机路径以及 Git 历史身份信息。详见 docs/PRIVACY.md。

当前源资产状态

《立体几何》原扫描第 49、50、55、56 页仍为空白;Markdown 和 EPUB 已依据另一份扫描逐页核对恢复,raw/ 原字节未改动。详见 docs/KNOWN_ISSUES.md。

权利说明

本仓库不对原著文字、版式或扫描图像授予新的版权许可。使用或再分发前,请自行确认适用地区的版权状态和授权条件。重建文件主要用于文献保存、校勘和个人研究。

shulihuazixuecongshu (tradecatlabs/shulihuazixuecongshu)

Searching Database…

What is this?This repository preserves and rebuilds the 17-volume Chinese textbook series '数理化自学丛书' (Self-Study Series of Mathematics, Physics, and Chemistry), keeping the original scanned PDFs alongside semantic Markdown transcriptions, extracted image assets, and tools to reproducibly build EPUB3 editions.
Is this an ebook collection or a software tool?Both: it is an archive containing the 17 original scanned volumes and their Markdown transcriptions, plus a Python/Pandoc-based toolchain that audits the source files and builds verifiable, byte-reproducible EPUB3 files from them.
What do I need installed to build or audit this?Building requires Python 3.10+, Pandoc 3.1+, and optionally GNU Make; a full raw-PDF audit additionally needs PyPDF2 3.x, qpdf, Poppler utilities (pdfinfo, pdfdetach, pdfsig, pdftotext), and ExifTool.
What changed with the v2.0.0 release, and are all 17 books still accessible?Starting from v2.0.0, GitHub Releases only ship three combined-subject EPUBs (Math, Physics, Chemistry) instead of 17 individual ones, though all 17 volumes of Markdown, raw PDFs, and assets remain in the repository and can still be built individually.
What does the pre-release privacy audit check before publishing?A push gate scans for risky Git candidate files, cross-platform path compatibility, PNG integrity and metadata, the generated EPUBs, common secret formats, local file paths, and Git history identity leaks.
What subjects do the 17 volumes cover?The series spans Algebra (4 volumes), Plane Trigonometry, Plane Geometry (2 volumes), Analytic Geometry, Solid Geometry, Chemistry (4 volumes), and Physics (4 volumes) — 17 volumes in total.
What happened to the blank pages in the Solid Geometry scan?Pages 49, 50, 55, and 56 of the original Solid Geometry scan were blank; the Markdown and EPUB versions were restored page-by-page using a second scan copy, while the raw scanned bytes were left untouched.
How thoroughly is the EPUB output verified?The current build has 0 missing or orphaned image resources, 0 empty image alt-text entries, and 88,653 verified MathML elements across the 4,875 extracted images, with byte-reproducible output guaranteed for a given source, Pandoc version, and SOURCE_DATE_EPOCH.
Size
139.6 MB
Stars
928
Age
1mo
Commits · 90 Days