• Stars
    star
    37
  • Rank 661,121 (Top 15 %)
  • Language
    R
  • License
    Other
  • Created almost 2 years ago
  • Updated 6 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

R function spotter, a helper package for identifying the functions and/or packages from file(s) or a github repository.

funspotr

Lifecycle: experimental R-CMD-check CRAN status

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 repositories1.

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

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 project2). 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

Examples of funspotr built reference tables

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

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)
#> # A tibble: 10 Ă— 2
#>    funs         pkgs     
#>    <chr>        <chr>    
#>  1 library      base     
#>  2 require      base     
#>  3 as_tibble    tidyr    
#>  4 mutate       dplyr    
#>  5 as.character base     
#>  6 group_by     dplyr    
#>  7 nest         tidyr    
#>  8 map          purrr    
#>  9 lm           stats    
#> 10 made_up_fun  (unknown)
  • funs: functions in file
  • pkgs: best guess as to the package the functions came from
  • …3

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
#> # A tibble: 4 Ă— 3
#>   relative_paths                absolute_paths                      spotted     
#>   <chr>                         <chr>                               <list>      
#> 1 R/Rmd-to-R.R                  https://raw.githubusercontent.com/… <named list>
#> 2 R/feat-engineering-lags.R     https://raw.githubusercontent.com/… <named list>
#> 3 R/load-inspections-save-csv.R https://raw.githubusercontent.com/… <named list>
#> 4 R/types-of-splits.R           https://raw.githubusercontent.com/… <named list>
  • relative_paths : relative filepath
  • absolute_paths: absolute filepath (in this case URL to raw file on github)
  • spotted: purrr::safely() style list-column of results4 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"))
#> # A tibble: 8 Ă— 4
#>   funs               pkgs    relative_paths                absolute_paths       
#>   <chr>              <chr>   <chr>                         <chr>                
#> 1 here               here    R/Rmd-to-R.R                  https://raw.githubus…
#> 2 read_csv           readr   R/feat-engineering-lags.R     https://raw.githubus…
#> 3 initial_time_split rsample R/feat-engineering-lags.R     https://raw.githubus…
#> 4 training           rsample R/feat-engineering-lags.R     https://raw.githubus…
#> 5 testing            rsample R/feat-engineering-lags.R     https://raw.githubus…
#> 6 sliding_period     rsample R/feat-engineering-lags.R     https://raw.githubus…
#> 7 write_csv          readr   R/load-inspections-save-csv.R https://raw.githubus…
#> 8 here               here    R/load-inspections-save-csv.R https://raw.githubus…

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 files5.

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
#> # A tibble: 4 Ă— 2
#>   relative_paths                absolute_paths                                  
#>   <chr>                         <chr>                                           
#> 1 R/Rmd-to-R.R                  https://raw.githubusercontent.com/brshallo/feat…
#> 2 R/feat-engineering-lags.R     https://raw.githubusercontent.com/brshallo/feat…
#> 3 R/load-inspections-save-csv.R https://raw.githubusercontent.com/brshallo/feat…
#> 4 R/types-of-splits.R           https://raw.githubusercontent.com/brshallo/feat…

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()
#> # A tibble: 24 Ă— 4
#>    funs      pkgs      relative_paths      absolute_paths                       
#>    <chr>     <chr>     <chr>               <chr>                                
#>  1 purl      knitr     R/Rmd-to-R.R        https://raw.githubusercontent.com/br…
#>  2 here      here      R/Rmd-to-R.R        https://raw.githubusercontent.com/br…
#>  3 library   base      R/types-of-splits.R https://raw.githubusercontent.com/br…
#>  4 theme_set ggplot    R/types-of-splits.R https://raw.githubusercontent.com/br…
#>  5 theme_bw  ggplot    R/types-of-splits.R https://raw.githubusercontent.com/br…
#>  6 set.seed  base      R/types-of-splits.R https://raw.githubusercontent.com/br…
#>  7 tibble    dplyr     R/types-of-splits.R https://raw.githubusercontent.com/br…
#>  8 rep       base      R/types-of-splits.R https://raw.githubusercontent.com/br…
#>  9 today     lubridate R/types-of-splits.R https://raw.githubusercontent.com/br…
#> 10 days      lubridate R/types-of-splits.R https://raw.githubusercontent.com/br…
#> # â„ą 14 more rows

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)
#> # A tibble: 5 Ă— 2
#>   funs        pkgs     
#>   <chr>       <chr>    
#> 1 library     base     
#> 2 as_tibble   dplyr    
#> 3 print       base     
#> 4 made_up_fun (unknown)
#> 5 hello_world (unknown)

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 funspotr6.

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)
#> # A tibble: 5 Ă— 2
#>   funs   pkgs     
#>   <chr>  <chr>    
#> 1 source base     
#> 2 tibble tibble   
#> 3 mutate dplyr    
#> 4 today  lubridate
#> 5 days   lubridate

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)
#> # A tibble: 11 Ă— 2
#>    funs         pkgs     
#>    <chr>        <chr>    
#>  1 library      base     
#>  2 require      base     
#>  3 as_tibble    tidyr    
#>  4 mutate       dplyr    
#>  5 as.character base     
#>  6 group_by     dplyr    
#>  7 nest         tidyr    
#>  8 mutate       dplyr    
#>  9 map          purrr    
#> 10 lm           stats    
#> 11 made_up_fun  (unknown)

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:

---
title: This is a post
author: brshallo
date: '2022-02-11'
tags: ["`r funspotr::spot_tags()`"]
slug: this-is-a-post
---

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)7.

(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 out8. The steps taken by funspotr would also not be needed in many other programming languages9.
  • 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 in10. 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 space11.
  • 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 problems12.
    • 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.

Footnotes

  1. Prior posts (some of which used a now deprecated API):

    ↩
  2. 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). ↩

  3. Other arguments may produce additional columns. See spot_funs() reference page for details. ↩

  4. list-column output where each item is a list containing result and error. ↩

  5. Took inspiration from plot() method in cranly. ↩

  6. 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. ↩

  7. 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}. ↩

  8. 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. ↩

  9. 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. ↩

  10. i.e. in interactive R scripts or Rmd or qmd documents where you use library() or related calls within the script. ↩

  11. 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. ↩

  12. 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. ↩

More Repositories

1

tidyverse-gifs

Code for using `flair`, `xaringan`, `pagedown`, and `magick` to build gif of steps in a series of piped operations. Example pulled from "Many Models" chapter of "R for Data Science"
HTML
19
star
2

pwiser

Package for pairwise operations in {dplyr}.
R
11
star
3

animatrixr

Toy package for animating 2x2 matrix transformations
R
6
star
4

r4ds_solutions

Solutions and notes to exercises for R4DS, http://r4ds.had.co.nz/, and bookdown set-up for book
CSS
6
star
5

spin

Package for Simulating Prediction Intervals
R
5
star
6

feat-engineering-slides

Slides for study group on "Feature Engineering and Selection..."
R
4
star
7

piececor

Package and notes for calculating piecewise correlations.
R
4
star
8

rta_2019

Talk with Research Triangle Analysts on 2019-08-20
4
star
9

save-versions-from-git

Save all versions of a file on git to a new folder.
R
3
star
10

flipbooks

Folder to start saving flipbooks in
CSS
2
star
11

macro-shell-example

R
2
star
12

flipbookr-gifs-examples

Examples of building gifs from flipbooks via pagedown and magick.
HTML
2
star
13

funspotr-rstudioconf2022

Placeholder for slides and materials for Rstudio conf 2022 presentation
2
star
14

handling-uncertainty-predictions-slides

Slides for User2022 Elevator pitch
R
1
star
15

dplyr-1.0.0-example

Repository for creating image shared to promote this post: https://www.bryanshalloway.com/2020/06/25/using-across-to-build-functions-with-dplyr-with-notes-on-legacy-approaches/
HTML
1
star
16

spinach

Simulating Prediction INtervals And CHance -- placeholder for a tidymodels friendly package for a generalizable approach to simulating uncertainty in predictions. See {spin} for some initial notes on approach https://github.com/brshallo/spin
1
star