Prototype description (MVP)

[krassowski]

A quick write up to describe the process that should be possible to follow in an ideal MVP:

1. User should be able to visit a website without the need to login anywhere (low barrier of entry)
2. User should be able to provide an URL to a GitHub/Lab repository with a single paste and click
3. The code from the repository should be downloaded (onto the server OR into the users browser - needs decision) subject to reasonable size restrictions
4. Any R markdown and Jupyter notebooks would be converted to a file to allow for simplified processing using readily available tools (nbconvert/jupytext)
5. A directed acyclic graph (if possible, or network otherwise) would be created by static analysis of the code, searching for known expressions indicating file import/export across R and Python files; the toolset needed for this needs more exploration and discussion
6. The network data would be sent to client to create a visualisation; a an algorithm to place the nodes in potentially complex DAGs needs to be chosen; I previously experimented with dagreD3 longest-path/finetop and graphviz. Apparently there is a recent progress in making it better: https://twitter.com/keenanisalive/status/1422318437590872076
7. The user should be able to annotate the visualisation by linking it to the relevant sections of the publication, or adding other comments; the annotations should be exportable to a (JSON/yaml) file with a specific name (to be decided), which can be added to the repository to make these parameters apply by default
8. The user should be able to download a static snapshot (e.g SVG) and maybe even an interactive version of the visualisation

Please feel free to work on any of the points by: researching, bouncing ideas, and drafting protocols, code/pseudo-code. If you would like to, please feel welcome to open a new issue asking questions and let's start narrowing down the implementation details.

[mstimberg]

User should be able to provide an URL to a GitHub/Lab repository with a single paste and click

Random thought: maybe even nicer would be if the user can give the doi or link to a publication and the tools finds the link to the github repository in the text itself :)

[krassowski]

That should be doable with Entrez API (for scanning full-text when available) and doi.org application/vnd.citationstyles.csl+json API (for fetching abstracts and metadata) example query.