Support User-Modifiable Template Export

[isosphere]

I want to be able to export annotations in a format that matches a specific structure of plain text output[^1]. If we support user-modifiable templates, users can fine-tune the exports from KoHighlight to match any plain text format they might want: plain text, CSV, markdown, HTML, JSON, TOML, etc.

[^1]: specifically, I want to generate markdown files that use properties as supported by logseq using book metadata

To that end, I propose that we identify an appropriate templating dependency and integrate it as an export choice in KoReader.

The user would not have control over what context variables are sent to the template; we'd just firehose them the metadata in a consistent variable format:

• Title
• Author
• Type
• Percent Read
• Rating
• Highlights (iterable)
  • Text
  • Comment
  • Datetime
  • Chapter
  • Page
• Modified
• Path
• Language
• Series
• Tags
• Pages
• Description

[isosphere]

Templating Dependency Selection

Considerations

• "Lightweight"
  • Not to complex to integrate with KOHighlights codebase
  • Limited technical debt
  • Limited additional dependencies
• Easy
  • Common
  • Simple
• Licensing
  • Permissive

Candidates

• mako
  • Active
  • Used by 202k projects on GitHub
  • 52 contributors
  • Limited dependencies
  • Plain-text oriented: can serve our purposes
  • Templates can inline Python - must trust templates
  • MIT License
• jinja
  • Active
  • 271 contributors
  • Limited dependencies
  • Seems more difficult to start with than mako, at a glance
  • I believe it's plain-text oriented and will serve our purposes
  • BSD-3-Clause license
• step
  • Has not been consistently active
  • 3 contributors
  • On release 0.0.4 - early days
  • Very simple
  • I think it would serve our purposes
  • BSD-3-Clause license
• cheetah3
  • Active
  • 32 contributors
  • Text-based
  • Weird syntax: most engines use tags, this is more like bash
  • MIT license
• pypage
  • Recently woke back up
  • 2 contributors
  • Release 2.0.8
  • Text-based
  • Eval-based: you must trust your templates, they execute Python
  • Apache 2.0 License

Selection

IMO the top candidates are Jinja and Mako. Jinja's syntax is more similar to Django's, and it does not allow embedded Python code which allows us to worry a bit less about malicious templates - even if they would only ever be user generated.

I think Jinja is a good choice.