compare_branches.R

# compare the object-usage lints obtained before/after a given PR
# the PR is always compared against master

library(magrittr)
library(dplyr)
library(purrr)
library(tibble)

# In lintr directory

lint_all_packages <- function(pkgs, linter) {
  map(pkgs, function(x) {
    lint_dir(x, linters = linter, parse_settings = FALSE)
  }) %>%
  set_names(basename(pkgs))
}

format_lints <- function(x) {
  x %>%
    map(as_tibble) %>%
    bind_rows(.id = "package")
}

run_workflow <- function(pkg_dir, pr, linter_name) {
  stopifnot(dir.exists(pkg_dir))
  if(!file.exists("lintr.Rproj")) {
    "compare_branches.R should be run inside the lintr-package directory"
  }

  pkgs <- dir(pkg_dir, full.names=TRUE)

  # lints from current-master branch
  git2r::checkout(branch = "master")
  devtools::load_all()

  linter <- get(linter_name)()
  main_lints <- pkgs %>%
    lint_all_packages(linter) %>%
    format_lints()

  # load the new version of lintr
  usethis::pr_fetch(pr)
  devtools::load_all()
  linter <- get(linter_name)()

  pr_lints <- pkgs %>%
    lint_all_packages(linter) %>%
    format_lints()

  # revert back to main
  git2r::checkout(branch = "master")

  list(
    main = main_lints,
    pr = pr_lints
  )
}

###############################################################################

# TODO: handle both command line args and interactive runs
# TODO: handle the case when working directory is not the lintr directory
# TODO: convert to the original branch (if this was not master) 
#   - at the end of the workflow (currently this always converts back to
#   master)
#   - and if there is any error when running the workflow
# TODO: save data.frame of lints to file

###############################################################################

# User should define the following in an interactive session:
# - 'pr' the number of the pull request that is to be compared against master
# - 'linter_name' the linter whose behaviour should be compared before/after
# the PR-associated changes
# - 'pkg_dir' pathname for a directory that contains the repositories that are
# to be analysed;
#
# ... and then source this script

pr <- if (exists("pr")) pr else 709
linter_name <- if (exists("linter_name")) linter_name else "object_usage_linter"
pkg_dir <- if (exists("pkg_dir")) {
  pkg_dir
} else {
  file.path("~", "proj", "code_as_data", "data", "packages")
}

message(pr)
message(linter_name)
message("Any package repo found in this directory will be analysed:", pkg_dir)

lints <- run_workflow(pkg_dir, pr, linter_name)

At present, you have to

• prepare a directory (referred to as pkg_dir) that contains repositories for multiple R packages (each package being of standard form with DESCRIPTION etc at the top level);
• have your 'lintr' directory as the working directory
• run this interactively in R
• define the variables "pr", "linter_name" and "pkg_dir" before sourcing this script

This script is for use during development of 'lintr'.
It identifies the lints called by lintr on a given set of packages by both the master-branch of lintr, and a PR-branch (specified numerically).
These lints can then be compared to find any differences introduced by the PR-associated changes

Updated to handle linters = assignment_linter() syntax (rather than linters = assignment_linter).

Thanks for sharing this. There's a small refac error (pkgs_dir vs. pkg_dir).
I'm currently trying this out on lintr#709 with the top 100 cran packages from last month plus shiny, with devtools::load_all(x) added in line 13 just before lint_dir.

Fixed the typo. Thanks @AshesITR

A modified version of this script is being added to lintr: r-lib/lintr#760