I spent a lot of time trying to build a "zero-config" plot capture system for both R and Python. It turns out the two languages have fundamentally different philosophies on how pixels get to the screen which make this easy in Python and super hard in R.

I wrote a deep dive comparing the display architectures in both languages, including some admittedly hacky ways to find figure objects through stack inspection. Hope it helps someone avoid our mistakes!