Discussion
Currently, there's a few different packages in pyreading:
• cardcatalog — support for working with Calibre-style library databases
• charminder — replacement for chardet, which went heavily into slop
• gardenpath — an html5ever-based replacement for Calibre's original html5parser (which in turn takes an untracked ABI dependency on lxml)
• icy_you — ICU4X bindings for Python
• spellsnake — spellbook bindings for Python
(con'd)
• pydofo — Python bindings for the PoDoFo PDF handling library
• pyresupport_win — Windows-specific platform support for Arcalibre
These packages are a mix of pure-Python, Python/Rust, and Python/C++. As much as possible, all of them use static linkage rather than shared library dependencies, even at the cost of efficiency. This is mostly because Python packaging makes it very hard to have dependencies on native, non-Python libraries.
The Python/Rust packages use Maturin, PyO3, and Cargo together to statically link Rust code into the build Python packages, while the Python/C++ packages use Conan to manage static compile-time dependencies.
Hopefully most or all of the functionality in the pyreading repo should be useful outside of Arcalibre directly, so please feel free to give it a go even if you're not looking to contribute to Arcalibre itself!
To talk a little more about the pyreading packages, let me focus on gardenpath in particular. Calibre makes extensive use of XPath for navigating HTML and ePub format books, which is very neat, but requires building an XML tree from an HTML document.
The previous strategy was to rely on html5-parser, which maps data structures exposed by Google's Gumbo library to populate an LXML tree. Since LXML supports XPath, that all works and gets you XPath over HTML5 documents.
There's three problems with this approach, though.
First, Gumbo has been deprecated since 2016 and thus isn't kept up to date with new HTML5 features. This means that Calibre is limited in what it can do with more recent HTML.
Second, this approach requires an ABI dependency on lxml, but pip only allows taking API dependencies on other Python packages. This required unstable and undocumented hacks to force ABI compatability.
Third, html5-parser is solely maintained by Kovid Goyal.
Install bonfire.cafe
Get the full app experience
