R build status codecov Project Status: Active

Extract summary statistics of R package structure and functionality. Also includes a function to extract statistics of all R packages from a local CRAN mirror. Not all statistics of course, but a good go at balancing insightful statistics while ensuring computational feasibility.

What statistics?

Statistics are derived from these primary sources:

  1. Numbers of lines of code, documentation, and white space (both between and within lines) in each directory and language
  2. Summaries of package DESCRIPTION file and related package meta-statistics
  3. Summaries of all objects created via package code across multiple languages and all directories containing source code (./R, ./src, and ./inst/include).
  4. A function call network derived from function definitions obtained from ctags, and references (“calls”) to those obtained from gtags. This network roughly connects every object making a call (as from) with every object being called (to).
  5. An additional function call network connecting calls within R functions to all functions from other R packages.

The primary function, pkgstats(), returns a list of these various components, including full data.frame objects for the final three components described above. The statistical properties of this list can be aggregated by the pkgstats_summary() function, which returns a data.frame with a single row of summary statistics. See below for further details.

Installation

The easiest way to install this package is via the associated r-universe. As shown there, simply enable the universe with

options(repos = c(
    ropenscireviewtools = "https://ropensci-review-tools.r-universe.dev",
    CRAN = "https://cloud.r-project.org"))

And then install the usual way with,

install.packages("pkgstats")

Alternatively, the package can be installed by running one of the following lines:

remotes::install_github ("ropensci-review-tools/pkgstats")
pak::pkg_install ("ropensci-review-tools/pkgstats")

The package can then loaded for use with

Installation on Linux systems

This package requires the system libraries ctags-universal and GNU global, both of which are automatically installed along with the package on both Windows and MacOS systems. Most Linux distributions do not include a sufficiently up-to-date version of ctags-universal, and so it must be compiled from source. This can be done by running a single function, ctags_install(), which will install both ctags-universal and GNU global.

The pkgstats package includes a function to ensure your local installations of universal-ctags and global work correctly. Please ensure you see the following prior to proceeding:

Note that GNU global can be linked at installation to the Universal Ctags plug-in parser to expand the default 5 languages to 30. This makes no difference to pkgstats results, as gtags output is only used to trace function call networks, which is only possible for compiled languages able to dynamically share pointers to the same objects. This is possible with the default parser regardless. The wealth of extra information obtained from linking global to the Universal Ctags parser is ultimately discarded anyway, yet parsing may take considerably longer. If this is the case, “default” behaviour may be recovered by first running the following command:

Sys.unsetenv (c ("GTAGSCONF", "GTAGSLABEL"))

See information on how to install the plugin for more details.

Demonstration

The following code demonstrates the output of the main function, pkgstats, applied to the relatively simple magrittr package. The system.time call also shows that these statistics are extracted quite quickly.

tarball <- "magrittr_2.0.1.tar.gz"
u <- paste0 ("https://cran.r-project.org/src/contrib/",
             tarball)
f <- file.path (tempdir (), tarball)
download.file (u, f)
system.time (
    p <- pkgstats (f)
    )
names (p)

The result is a list of various data extracted from the code. All except for objects and network represent summary data:

p [!names (p) %in% c ("objects", "network", "external_calls")]

The first item, loc, contains the following Lines-Of-Code and related statistics, separated into distinct combinations of computer language and directory:

  1. nfiles = Numbers of files in each directory and language.
  2. nlines = Total numbers of lines in all files.
  3. nlines = Total numbers of lines of code.
  4. ndoc = Total numbers of documentation or comment lines.
  5. nempty = Total numbers of empty of blank lines.
  6. nspaces = Total numbers of white spaces in all code lines, excluding leading indentation spaces.
  7. nchars = Total numbers of non-white-space characters in all code lines.
  8. nexpr = Median numbers of nested expressions in all lines which have any expressions (see below).
  9. ntabs = Number of lines of code with initial tab indentation.
  10. indentation = Number of spaces by which code is indented (with -1 denoting tab-indentation).

Numbers of nested expressions are counted as numbers of brackets of any type nested on a single line. The following line has one nested bracket:

x <- myfn ()

while the following has four:

x <- function () { return (myfn ()) }

Code with fewer nested expressions per line is generally easier to read, and this metric is provided as one indication of the general readability of code. A second relative indication may be extracted by converting numbers of spaces and characters to a measure of relative numbers of white spaces, noting that the nchars value quantifies total characters including white spaces.

index <- which (p$loc$dir %in% c ("R", "src")) # consider source code only
sum (p$loc$nspaces [index]) / sum (p$loc$nchars [index])

Finally, the ntabs statistic can be used to identify whether code uses tab characters as indentation, otherwise the indentation statistics indicate median numbers of white spaces by which code is indented. The objects, network, and external_calls items returned by the pkgstats() function are described further below.

The pkgstats_summary() function

A summary of the pkgstats data can be obtained by submitting the object returned from pkgstats() to the pkgstats_summary() function:

This function reduces the result of the pkgstats() function to a single line with 91 entries, represented as a data.frame with one row and that number of columns. This format is intended to enable summary statistics from multiple packages to be aggregated by simply binding rows together. While 91 statistics might seem like overkill, the pkgstats_summary() function aims to return as many usable raw statistics as possible in order to flexibly allow higher-level statistics to be derived through combination and aggregation. These 91 statistics can be roughly grouped into the following categories (not shown in the order in which they actually appear), with variable names in parentheses after each description.

Package Summaries

  • name (package)
  • Package version (version)
  • Package date, as modification time of DESCRIPTION file where not explicitly stated (date)
  • License (license)
  • Languages, as a single comma-separated character value (languages), and excluding R itself.
  • List of translations where package includes translations files, given as list of (spoken) language codes (translations).

Information from DESCRIPTION file

  • Package URL(s) (url)
  • URL for BugReports (bugs)
  • Number of contributors with role of author (desc_n_aut), contributor (desc_n_ctb), funder (desc_n_fnd), reviewer (desc_n_rev), thesis advisor (ths), and translator (trl, relating to translation between computer and not spoken languages).
  • Comma-separated character entries for all depends, imports, suggests, and linking_to packages.

Numbers of entries in each the of the last two kinds of items can be obtained from by a simple strsplit call, like this:

length (strsplit (s$suggests, ", ") [[1]])

Numbers of files and associated data

  • Number of vignettes (num_vignettes)
  • Number of demos (num_demos)
  • Number of data files (num_data_files)
  • Total size of all package data (data_size_total)
  • Median size of package data files (data_size_median)
  • Numbers of files in main sub-directories (files_R, files_src, files_inst, files_vignettes, files_tests), where numbers are recursively counted in all sub-directories, and where inst only counts files in the inst/include sub-directory.

Statistics on lines of code

  • Total lines of code in each sub-directory (loc_R, loc_src, loc_ins, loc_vignettes, loc_tests).
  • Total numbers of blank lines in each sub-directory (blank_lines_R, blank_lines_src, blank_lines_inst, blank_lines_vignette, blank_lines_tests).
  • Total numbers of comment lines in each sub-directory (comment_lines_R, comment_lines_src, comment_lines_inst, comment_lines_vignettes, comment_lines_tests).
  • Measures of relative white space in each sub-directory (rel_space_R, rel_space_src, rel_space_inst, rel_space_vignettes, rel_space_tests), as well as an overall measure for the R/, src/, and inst/ directories (rel_space).
  • The number of spaces used to indent code (indentation), with values of -1 indicating indentation with tab characters.
  • The median number of nested expression per line of code, counting only those lines which have any expressions (nexpr).

Statistics on individual objects (including functions)

These statistics all refer to “functions”, but actually represent more general “objects,” such as global variables or class definitions (generally from languages other than R), as detailed below.

  • Numbers of functions in R (n_fns_r)
  • Numbers of exported and non-exported R functions (n_fns_r_exported, n_fns_r_not_exported)
  • Number of functions (or objects) in other computer languages (n_fns_src), including functions in both src and inst/include directories.
  • Number of functions (or objects) per individual file in R and in all other (src) directories (n_fns_per_file_r, n_fns_per_file_src).
  • Median and mean numbers of parameters per exported R function (npars_exported_mn, npars_exported_md).
  • Mean and median lines of code per function in R and other languages, including distinction between exported and non-exported R functions (loc_per_fn_r_mn, loc_per_fn_r_md, loc_per_fn_r_exp_m, loc_per_fn_r_exp_md, loc_per_fn_r_not_exp_mn, loc_per_fn_r_not_exp_m, loc_per_fn_src_mn, loc_per_fn_src_md).
  • Equivalent mean and median numbers of documentation lines per function (doclines_per_fn_exp_mn, doclines_per_fn_exp_md, doclines_per_fn_not_exp_m, doclines_per_fn_not_exp_md, docchars_per_par_exp_mn, docchars_per_par_exp_m).

Network Statistics

The full structure of the network table is described below, with summary statistics including:

  • Number of edges, including distinction between languages (n_edges, n_edges_r, n_edges_src).
  • Number of distinct clusters in package network (n_clusters).
  • Mean and median centrality of all network edges, calculated from both directed and undirected representations of network (centrality_dir_mn, centrality_dir_md, centrality_undir_mn, centrality_undir_md).
  • Equivalent centrality values excluding edges with centrality of zero (centrality_dir_mn_no0, centrality_dir_md_no0, centrality_undir_mn_no0, centrality_undir_md_no).
  • Numbers of terminal edges (num_terminal_edges_dir, num_terminal_edges_undir).
  • Summary statistics on node degree (node_degree_mn, node_degree_md, node_degree_max)

External Call Statistics

The final column in the result of the pkgstats_summary() function summarises the external_calls object detailing all calls make to external packages (including to base and recommended packages). This summary is represented as a single character string:

s$external_calls

This is structured to allow numbers of calls to all packages to be readily extracted with code like the following:

calls <- do.call (rbind,
                  strsplit (strsplit (s$external_call, ",") [[1]], ":"))
calls <- data.frame (package = calls [, 1],
                     n_total = as.integer (calls [, 2]),
                     n_unique = as.integer (calls [, 3]))
print (calls)

The two numeric columns respectively show the total number of calls made to each package, and the total number of unique functions used within those packages. While this result is relatively uninformative for the magrittr package, which imports no other packages and relies only on base R functions, these results will generally provide detailed information on numbers of calls made and functions used.

The following sub-sections provide further detail on the objects, network, and external_call items, which could be used to extract additional statistics beyond those described here.

Objects

The objects item contains all code objects identified by the code-tagging library ctags. For R, those are primarily functions, but for other languages may be a variety of entities such as class or structure definitions, or sub-members thereof. Object tables look like this:

head (p$objects)

The magrittr package has a total of 191 objects, which the following lines provide some insight into.

table (p$objects$language)
table (p$objects$kind)
table (p$objects$kind [p$objects$language == "R"])
table (p$objects$kind [p$objects$language == "C"])
table (p$objects$kind [p$objects$language == "C++"])

Network

The network item details all relationships between objects, which generally reflects one object calling or otherwise depending on another object. Each row thus represents one edge of a “function call” network, with each entry in the from and to columns representing the network vertices or nodes.

head (p$network)
nrow (p$network)

The network table includes additional statistics on the centrality of each edge, measured as betweenness centrality assuming edges to be both directed (centrality_dir) and undirected (centrality_undir). More central edges reflect connections between objects that are more central to package functionality, and vice versa. The distinct components of the network are also represented by discrete cluster numbers, calculated both for directed and undirected versions of the network. Each distinct cluster number represents a distinct group of objects, internally related to other members of the same cluster, yet independent of all objects with different cluster numbers.

The network can be viewed as an interactive vis.js network through passing the result of pkgstats – here, p – to the plot_network() function.

External Calls

The external_calls item is structured similar to the network object, but identifies all calls to functions from external packages. However, unlike the netowrk and object data, which provide information on objects and relationships in all computer languages used within a package, the external_calls object maps calls within R code only, in order to provide insight into the use within a package of of functions from other packages, including R’s base and recommended packages. The object looks like this:

head (p$external_calls)

These data are converted to a summary form by the pkgstats_summary() function, which tabulates numbers of external calls and unique functions from each package. These data are presented as a single character string which can be easily converted to the corresponding numeric values using code like the following:

x <- strsplit (s$external_calls, ",") [[1]]
x <- do.call (rbind, strsplit (x, ":"))
x <- data.frame (pkg = x [, 1],
                 n_total = as.integer (x [, 2]),
                 n_unique = as.integer (x [, 3]))
x$n_total_rel <- round (x$n_total / sum (x$n_total), 3)
x$n_unique_rel <- round (x$n_unique / sum (x$n_unique), 3)
print (x)

Those data reveal, for example, that the magrittr package makes 22 individual calls to 12 unique functions from the “base” package.

Code of Conduct

Please note that this package is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.