Extracting and cross-referencing figures in repo code and paper

[krassowski]

Just a general thought: I think what could be an interesting angle specific to academic papers is to focus on the figures as the scientific output. I.e., the main task of the tool would be to figure out what figures of the paper can be recreated using the repository and how. It would also be cool to not only take into account the code itself but also additional information that might be present in a README file, or even in the publication itself.

Originally posted by @mstimberg in #1 (comment)

Building upon this one I can envision a semi-automated tool which when matching a regular expression/substring from figure caption, for simplicity say "Figure 1" in the code will compare it with the article text and opens a dismissible suggestion in a sidebar:

It seems that [lines X-Y] in file Z.py reference Figure 1. [Description]. Is this correct?

Yes | Edit (modify exact location) | No (dismiss)

When confirmed by the user the data would be stored into a config file allowing to avoid repeating this on subsequent visits (if the config gets committed to that repository; we could be even make a bot create a PR on user request), and more importantly the figure would show in the flow visualisation of the repo, and when browsing code (on the sidebar) - assuming that we would add code browsing option, which would not be that difficult but probably outside of Sprint scope.

I also commented in the previous discussion that supplementary tables could get this treatment too.