Replace obsolete wkhtmltopdf PDF engine

[mtmacdonald]

DocGen currently uses wkhtmltopdf to generate PDFs. This engine is obsolete.

We need to find an alternative modern engine for generating PDFs.

[mtmacdonald]

Here is a summary and explanation of what I found to help with deciding next steps.

wkhtmltopdf (existing obsolete engine)

The existing wkhtmltopdf engine really holds back Docgen project (outdated CSS engine, slow, hard-to-debug). It still "works", but we do need a replacement if DocGen is going to be a seriously maintained project, (and it seems there are no drop-in modern replacements that are open-source).

• wkhtmltopdf "prints" webpages to PDF (like a user would with File | Print in a browser)
• "headless browser" implementation
• based on Qt 4 and QtWebkit (obsolete)
• generic: works well for custom, user-defined content (global CSS styles, can print "real" web pages)
• it is old (2012-2015 tech), slow, insecure, and deprecated
• no modern CSS support (e.g. no FlexBox, Grids)
• hard to debug styles (new machines can't run ancient browsers with comparable CSS engines)
• there are no modern open-source replacements
  • Commercial (paid) tools
    • OneDoc
    • PrinceXML
  • someone tried to revive wkhtmltopdf, but looks inactive/unstable
  • closest alternatives I could find:
    • md-to-pdf
    • markdown-pdf

ReactPDF

• react-pdf uses a different, bottom-up approach to generate PDFs
• uses pdfkit to "construct" PDFs from low-level primitives (Node API)
• Exposes a React-friendly interface for generating PDFs, but you have to use their components and primitives (can't just "print" any generic React app, it's not as simple as re-using the existing website components to also build the PDFs, we'd need to make and maintain a parallel structure for the PDF and the website)
• Has some support for modern CSS, including FlexBox, but not Grids (uses YogaLayout - platform independent layout engine)
• Uses css-in-js approach (not global styles), which makes some things harder:
  • can't re-use our pre-existing SASS
  • can only use a print-friendly subset of CSS
  • custom user content and user-defined CSS can't be supported the same way as before
• For rendering Markdown (custom user content) to PDF, the best way I could find is react-pdf-html. It parses HTML with node-html-parser and csstree, then transforms the tree into ReactPDF primitives (it has a nice interface for overriding the default transformations)
• modern, maintained, great docs
• fast