Topics/cran_survey.md

In duncantl/CodeAnalysis: Tools for Static Analysis of R code

CRAN Survey

Motivation

• How commonly used are certain features of R? These could be any features we're interested in. E.g., assignment from a conditional.
• How complex is the code? E.g., number of calls inside a function
• How has the complexity evolved over time? Need a time series of code
• How does the evolution of packages on CRAN compare to other repositories like CPAN, CTAN, pip, etc?

Package Statistics Wishlist

Per function:

• depth of nesting:

  • control flow
  • calls
  • parentheses

• total number (or locations) of:

  • local aliases for global functions (e.g., options(), rm())
  • calls
  • calls, including calls made by calls (computed recursively)
  • ratio of calls inside packages compared to base, recommended, and 3rd-party packages.
  • literals
  • globals
  • loops, broken down by for, while, apply family
  • Depth of nested loops (computed recursively)
  • number of parameters (and whether default arguments are calls, other parameters, literals)
  • <<- assignments
  • calls to NSE/macro functions: eval(), substitute(), quote(), ...
  • calls to foreign functions: .C(), .Call()
  • use of registration of routines

• comments and number of characters

  • namespacing: :: and :::, also use NAMESPACE to identify calls to functions in other packages that don't use ::
  • conditional assignments
  • calls to parallelism functions, such as from parallel
  • Reference objects such as environments, closures, and formulas. Do people use these directly?

• Whether a function can be parallelized?

• Whether a function always returns the same type?

Per package:

• number of

  • exported objects
  • private objects
  • object-oriented classes and methods (S3, S4, RefClasses)

• What are all the dependencies of this package?

• Which of the non exported functions can be called by exported functions? (recursively - identifies a type of dead code)
• Which exported functions are used in the vignette?
• Which packages use Roxygen? Is there more or less documentation compared to Rd files?
• Usage of specific libraries over time, ie. Rcpp and magrittr %>%.
• Usage of plotting libraries over time, ie. base, grid, lattice, ggplot
• Presence of C/C++/Fortran/Java/other language code
• "Amount" of R code versus foreign language code
• Metadata from DESCRIPTION (e.g., author, date, versions)
• Associate author names to git blame
• Usage of system C libraries

Ideas for data sources:

• Stack Overflow
• R user mailing lists

duncantl/CodeAnalysis documentation built on Feb. 21, 2024, 10:49 p.m.