Yep, in-house PDF generators should be some sort of good middle ground, but I dunno if this 'weasyprint' is open source, is _lean_ open source? (no c++, java, etc).

When dealing with an ultra-complex file format which cannot be dodged, usually a good way to deal with it is to only use a very simple but coherent subset and enforce this usage with validation tools.

For instance, the web, noscript/basic (x)html (or you are jailed in the 2.5 web engines of the whatng cartel).

With PDF, I dunno much of the format (since I did not manage to download easily the specs), but when I have to print some text, I have a very small PDF generator for that (written ~25 years ago, so no utf-8 for me).

But what's important: such attempt must be sided with re-assessing the pertinence of the usage of the information systems, and yes, it will annoying and much less comfy and that MUST be acknowledged before even trying.

And big tech is not the only one trying hard to do vendor and developer lock-in.