Preventing code rot

Preventing code rot

Recap

Computational reproducibility is hard!

• Shared reproducibility packages often fail
  (e.g., Eubank, 2016)
• Code is increasingly likely to break as time passes
  (it "rots")
  • This is true even if code is untouched

Preventing code rot

• Computing environments inevitably change
  • Installation or removal of software
  • Software updates
  • Removal or relocation of files
  • (Hardware updates)
• Complex computing environments exacerbate the problem

Preventing code rot

Preventing code rot

Preventing code rot

1. checkpoint by Microsoft
   • Requires a project-based workflow
   • Package database will gradually grow
2. groundhog
   • Package database will gradually grow
3. renv by RStudio
   • Most flexible and powerful
   • Least straight forward to use
   • No "forensic" applications

Preventing code rot

Preventing code rot

Dependencies are detected automatically

library("checkpoint")
checkpoint("2022-04-27", r_version = "4.1.2")
library("ggplot2")Copy Code

checkpoint("2022-04-27")
install.packages("ggplot2")Copy Code

Uses a date-specific directory outside of usual library

~/.checkpoint/...Copy Code

Preventing code rot

Preventing code rot

Virtual machines

Preventing code rot

Containers

(Piccolo & Frampton, 2016)

Preventing code rot

• Virtual machines (e.g., Vagrant)
  • Encapsulate an entire operating system
  • Require a lot of disk space
  • Take little longer to set up and boot
• Containers (e.g., Docker)
  • Widely used in software development
  • Depend on OS kernel, but can be run on all common operating systems

Preventing code rot

Preventing code rot

• Virtual machines and containers requires some expertise
• Online services simplify setup, collaboration, sharing, and archiving
  • PsychNotebooks
  • CodeOcean
  • RStudio.cloud

Preventing code rot

An unobtrusive Docker workflow for papaja

Preventing code rot

Summary

• Computing environments inevitably change
• Robust computational reproducibility requires a fixed computing environment
  1. prevents code rot
  2. facilitates sharing reproducible analyses

• Mind software licenses when sharing software environments