In brshallo/funspotr: Spot R Functions & Packages

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%",
  warning = FALSE,
  message = FALSE
)

The goal of funspotr (R function spotter) is to make it easy to identify which R functions and packages are used in files and projects. It was initially written to create reference tables of the functions and packages used in a few popular github repositories[^prior-posts].

There are roughly three types of functions in funspotr:

• list_files_*(): that identify files in a repository or related location
• spot_*(): that identify functions or packages in files
• other helpers that manipulate or plot outputs from the above functions

[^prior-posts]: Prior posts (some of which used a now deprecated API):
- Identifying R Functions & Packages Used in GitHub Repos (funspotr part 1)
- Identifying R Functions & Packages in Github Gists (funspotr part 2)
- Network Plots of Code Collections (funspotr part 3)

funspotr is set-up for parsing R, Rmarkdown or Quarto files. If you want to parse a Jupyter notebook you should first convert it to an appropriate file type. If you pass in a file type that is not recognized (e.g. a .txt file) funspotr will attempt to parse it as if it is a .R script.

funspotr is primarily designed for identifying the functions / packages in self-contained files or collections of self-contained files (e.g. a blogdown project^[Rather than, for example, targets workflows. Also, in some cases funspotr may not identify every function and/or package in a file (see [Limitations, problems, musings] or read the source code for details).]). Though see [Package dependencies in another file] for examples of using it in other contexts.

Installation

Install the latest stable version of funspotr from CRAN with:

install.packages("funspotr")

You can install the development version of funspotr from GitHub with:

# install.packages("devtools")
devtools::install_github("brshallo/funspotr")

Talks and posts

• slides and presentation from Rstudio Conf 2022 From summarizing projects to setting tags, uses of parsing R files
• Part 1 of 3 from a series on Identifying R functions and packages...

Examples of funspotr built reference tables

funspotr can be used to create reference tables of the functions and packages used in R projects.

• Julia Silge blog
• David Robinson screencasts
• R for Data Science book
• Bryan Shalloway blog
• brshallo gists

Spot functions in a file

The primary function in funspotr is spot_funs() which returns a dataframe showing the functions and associated packages used in a file.

library(funspotr)

file_lines <- "
library(dplyr)
require(tidyr)

as_tibble(mpg) %>% 
  mutate(class = as.character(class)) %>%
  group_by(class) %>%
  nest() %>%
  mutate(stats = purrr::map(data,
                            ~lm(cty ~ hwy, data = .x)))

made_up_fun()
"

file_output <- tempfile(fileext = ".R")
writeLines(file_lines, file_output)

spot_funs(file_path = file_output)

• funs: functions in file
• pkgs: best guess as to the package the functions came from
• ...^[Other arguments may produce additional columns. See spot_funs() reference page for details.]

Spot functions on all files in a project

funspotr has a few list_files_*() functions that return a dataframe of relative_paths and absolute_paths of all the R, Rmarkdown, or quarto files in a specified location (currently: github repo, gists, or local). These can be combined with a variant of spot_funs() that maps the function across each file path found, spot_funs_files():

library(dplyr)

# repo for an old presentation I gave
gh_ex <- list_files_github_repo(
  repo = "brshallo/feat-eng-lags-presentation", 
  branch = "main") %>% 
  spot_funs_files()

gh_ex

• relative_paths : relative filepath
• absolute_paths: absolute filepath (in this case URL to raw file on github)
• spotted: purrr::safely() style list-column of results^[list-column output where each item is a list containing result and error.] from mapping spot_funs() across absolute_paths.

These results may then be unnested with the helper funspotr::unnest_results() to provide a table of functions and packages by filepath. This can be manipulated like any other dataframe -- say we want to filter to only those files where here, readr or rsample packages are used.

gh_ex %>% 
  unnest_results() %>% 
  filter(pkgs %in% c("here", "readr", "rsample"))

The outputs from funspotr::unnest_results() can also be passed into funspotr::network_plot() to build a network visualization of the connections between functions/packages and files^[Took inspiration from plot() method in cranly.].

Previewing and customizing files to parse

You might only want to parse certain file types or a subset of the files in a repo.

preview_files <- list_files_github_repo(
  repo = "brshallo/feat-eng-lags-presentation", 
  branch = "main")

preview_files

Say we only want to parse the "types-of-splits.R" and "Rmd-to-R.R" files.

preview_files %>% 
  filter(stringr::str_detect(relative_paths, "types-of-splits|Rmd-to-R")) %>% 
  spot_funs_files() %>% 
  unnest_results()

Note that if you have a lot of files in a repo you may need to set-up sleep periods or clone the repo locally and then parse the files from there so as to stay within the limits of github API hits.

Other things

Files you didn't write

Functions created in the file as well as functions from unavailable packages (or packages that don't exist) will output as pkgs = "(unknown)".

file_lines_missing_pkgs <- "
library(dplyr)

as_tibble(mpg)

hello_world <- function() print('hello world')

madeuppkg::made_up_fun()

hello_world()
"

missing_pkgs_ex <- tempfile(fileext = ".R")
writeLines(file_lines_missing_pkgs, missing_pkgs_ex)

spot_funs(file_path = missing_pkgs_ex)

To spot which package a function is from you must have the package installed locally. Hence for files on others' github repos or that you created on a different machine, it is a good idea to start with funspotr::check_pkgs_availability() to see which packages you are missing and install the missing packages locally. If you don't want to edit your global library you may want to use renv or other environment management tools.

funspotr has an internal helper funspotr::install_missing_pkgs() for installing missing packages:

spot_pkgs(file_output) %>%
  check_pkgs_availability() %>%
  funspotr::install_missing_pkgs()

Alternatively, you may want to clone the repository locally and then use renv::dependencies() and only then start using funspotr^[renv is a more robust approach to finding and installing dependencies -- particularly in cases where you are missing many dependencies or don't want to alter the packages in your global library.].

Package dependencies in another file

spot_funs() is currently set-up for self-contained files. But spot_funs_custom() allows the user to explicitly specify pkgs where functions may come from. This is useful in cases where the packages loaded are not in the same location as the file_path (e.g. they are loaded via source(), or a DESCRIPTION file, or some other workflow). For example, below is a made-up example where the library() calls are made in a separate file and source()'d in.

# file where packages are loaded
file_libs <- "library(dplyr)
library(lubridate)"

file_libs_output <- tempfile(fileext = ".R")
writeLines(file_libs, file_libs_output)

# File of interest where things happen
file_run <- glue::glue(
"source('{ file_libs_output }')

tibble::tibble(days_from_today = 0:10) %>% 
    mutate(date = today() + days(days_from_today))
", 
file_libs_output = stringr::str_replace_all(file_libs_output, "\\\\", "/")
)

file_run_output <- tempfile(fileext = ".R")
writeLines(file_run, file_run_output)

# Identify packages using both files and then pass in explicitly to `spot_funs_custom()`
pkgs <- c(spot_pkgs(file_libs_output), 
          spot_pkgs(file_run_output, show_explicit_funs = TRUE))

spot_funs_custom(
  pkgs = pkgs,
  file_path = file_run_output)

Also see funspotr::spot_pkgs_from_description().

Show all function calls

Passing in show_each_use = TRUE to ... in spot_funs() or spot_funs_files() will return all instances of a function call rather than just once for each file.

Compared to the initial example, mutate() now shows-up at both rows 4 and 8:

spot_funs(file_path = file_output, show_each_use = TRUE)

Helper for blog tags

To automatically have your packages used as the tags for a blog post you can add an inline function funspotr::spot_tags() to a bullet in the tags or categories argument of your YAML header. For example:

cat(htmltools::includeText("inst/ex-rmd-tags.Rmd"))

How spot_funs() works

funspotr mimics the search space of each file prior to identifying pkgs/funs. At a high-level...

1. Create a new R instance using callr and clean-up the specified file using formatR
2. Load packages. Explicit calls (e.g. pkg::fun()) are loaded individually via import and are loaded last (putting them at the top of the search space)^[This heuristic is imperfect and means that a file with "library(dplyr); select(); MASS::select()" would view both select() calls as coming from {MASS} -- when what it should do is view the first was as coming from {dplyr} and the second from {MASS}.].

(steps 1 and 2 needed so that step 4 has the best chance of identifying the package a function comes from in the file.)

1. Pass file through utils::getParseData() and filter to just functions
2. Pass functions through utils::find() to identify associated package

Explainer slide from Rstudio Conf 2022 presentation:

Limitations, problems, musings

• funspotr is specific to R. If you try and pass in a file from a different language you will get a parsing error or the code commented out^[For example... If you pass in a .Rmd or .qmd that has a mix of R and python code chunks, the python chunks will simply be commented out. If you pass in a python script, you will almost certainly get a parsing error for that file.]. The steps taken by funspotr would also not be needed in many other programming languages^[In a language like python, where calls are more explicit (e.g. np.*), all of the stuff with recreating the search space would likely be unnecessary and you could more easily just identify packages/functions by parsing the text.].
• funspotr does not work perfectly at identifying functions or packages. One common example is it will not identify functions passed as arguments. For example it will not identify mean in this example: lapply(x, mean) . Similarly it will not identify functions within switch(). See #13.
• If a file contains R syntax that is not well defined it will not be parsed and will return an error. See formatR#further-notes (used by funspotr in parsing) for other common reasons for failure.
• knitr::read_chunk() and knitr::purl() in a file passed to funspotr will also frequently cause an error in parsing. See knitr#1753 & knitr#1938
• Please open an issue if you find other cases where parsing breaks :-) .
• As mentioned elsewhere, the default parsing of spot_funs() is primarily for cases where package dependencies are loaded in the same file that they are used in^[i.e. in interactive R scripts or Rmd or qmd documents where you use library() or related calls within the script.]. Scripts that are not self-contained typically should have the pkgs argument provided explicitly via spot_funs_custom().
• funspotr does not pay attention to when functions are reexported from elsewhere. For example, many tibble functions are reexported by dplyr and tidyr -- funspotr though will not know the "true" home of these functions it is simply looking at the top of the search space^[For example when reviewing David Robinson's Tidy Tuesday code I found that the meme package was used far more than I would have expected. Turns out it was just due to it reexporting the aes() function from ggplot.].
• Feel free to open an issue if you'd be interested in a simplifying function or vignette for mapping spot_funs() through other folder structures not yet mentioned.
• All the functions in "R/spot-pkgs.R" would probably be better handled by something like renv::dependencies() or a parsing based approach. The simple regex's I use have a variety of problems^[e.g. in this case lines <- "library(pkg)" the pkg would show-up as a dependency despite just being part of a quote rather than actually loaded. See #14 for disucssion of other approaches.].
  • There may be something to be learned from how R CMD check does function parsing. funspotr's current approach is comparatively slow and uses imperfect heuristics.
• Does not identify infix operators, e.g. +
• funspotr has lots of dependencies. It may have make sense to move some of the non-core functionality into a separate package (e.g. stuff concerning list_files*())
• Rather than running list_files_github_repo() it may make sense to instead clone the repo locally and then run list_files_wd() from the repo prior to running spot_funs_files() as this will limit the number of API hits to github.
• Currently it's possible to have github block you pretty soon due to hitting too many files (in which case you'll likely get a 403 or connection error). There are some things that could be done to reduce number of github API hits (e.g. above bullet, or using Sys.sleep(), ...).
• Throughout the code and package documentation I have "inspiration" bullets followed by a link pointing to places where I took stuff from stack overflow, github, or other packages. Also see the footnotes of the README.

brshallo/funspotr documentation built on Jan. 4, 2024, 10:08 a.m.