The goal of functiondepends is to allow for tidy exploration of unstructured codebase without evaluation of code.

Installation

One can install functiondepends from CRAN:

install.packages("functiondepends")

or development version from GitHub:

# install.packages("devtools")
devtools::install_github("jakubsob/functiondepends")

Examples

library(functiondepends)
# Create environment for loaded functions 
envir <- new.env()
# Search recursively current directory
functions <- find_functions(".", envir = envir, recursive = TRUE)

functions
#> # A tibble: 5 × 3
#>   Path  Function          SourceFile         
#>   <chr> <chr>             <chr>              
#> 1 R     find_dependencies find-dependencies.R
#> 2 R     is_function       find-functions.R   
#> 3 R     get_function_name find-functions.R   
#> 4 R     is_assign         find-functions.R   
#> 5 R     find_functions    find-functions.R

Search for dependencies of function find_functions within parsed functions:

dependency <- find_dependencies("find_functions", envir = envir, in_envir = TRUE)
dependency
#> # A tibble: 2 × 5
#>   Source            SourceRep SourceNamespace Target         TargetInDegree
#>   <chr>                 <int> <chr>           <chr>                   <int>
#> 1 get_function_name         1 user-defined    find_functions              2
#> 2 is_function               1 user-defined    find_functions              2

Note that SourceNamespace column has value user-defined as the functions are searched within source of the package.

Search for all dependencies of find_functions function:

library(ggplot2)
library(dplyr)
dependency <- find_dependencies("find_functions", envir = envir, in_envir = FALSE)
dependency %>%
  slice_max(SourceRep, n = 10) %>%
  mutate(Source = reorder(Source, SourceRep)) %>%
  ggplot(aes(x = Source, y = SourceRep, fill = SourceNamespace)) +
  geom_col() +
  coord_flip() +
  labs(caption = "Top 10 most repeated calls in 'find_functions'.")

Note that name df is often used to store object of type data.frame. df is also a name of F distribution density function from stats package. If you suspect that given function ought not to use a specific package, see the source code of function to check the context. To do so, one can execute find_dependencies function with add_info argument set to TRUE.

library(tidyr)
dependency <- find_dependencies("find_functions", envir = envir, in_envir = FALSE, add_info = TRUE)
dependency %>%
  filter(SourceNamespace == "stats") %>%
  select(Source, SourcePosition, SourceContext) %>%
  unnest(c(SourcePosition, SourceContext))
#> # A tibble: 6 × 3
#>   Source SourcePosition SourceContext                                           
#>   <chr>           <dbl> <chr>                                                   
#> 1 df                 10 "    df <- purrr::map_dfr(sourceFiles, function(file) {"
#> 2 df                 19 "    source_name <- basename(df$Path)"                  
#> 3 df                 21 "    df <- df %>% dplyr::mutate(Path = stringr::str_rem…
#> 4 df                 23 "        paths <- stringr::str_split(df$Path, \"/|\\\\\…
#> 5 df                 25 "        df <- tidyr::separate(df, \"Path\", into = pas…
#> 6 df                 27 "    df %>% dplyr::mutate(SourceFile = source_name)"

One can see that indeed df is not a call to function stats::df.

dependency <- find_dependencies(unique(functions$Function), envir = envir, in_envir = FALSE)
dependency %>%
  distinct(Target, TargetInDegree) %>%
  mutate(Target = reorder(Target, TargetInDegree)) %>%
  ggplot(aes(x = Target, y = TargetInDegree)) +
  geom_col() +
  coord_flip() +
  labs(caption = "Functions with most function calls.")

dependency <- find_dependencies(unique(functions$Function), envir = envir, in_envir = FALSE)
dependency %>%
  group_by(SourceNamespace) %>%
  tally(name = "Count") %>%
  slice_max(Count, n = 10) %>%
  mutate(SourceNamespace = reorder(SourceNamespace, Count)) %>%
  ggplot(aes(x = SourceNamespace, y = Count)) +
  geom_col() +
  coord_flip() +
  labs(caption = "Top 10 used namespaces.")

See which user-defined functions depend most on other user-defined functions within searched codebase.

dependency <- find_dependencies(unique(functions$Function), envir = envir, in_envir = TRUE)
dependency %>%
  distinct(Target, TargetInDegree) %>%
  arrange(-TargetInDegree)
#> # A tibble: 5 × 2
#>   Target            TargetInDegree
#>   <chr>                      <dbl>
#> 1 find_functions                 2
#> 2 is_function                    1
#> 3 find_dependencies              0
#> 4 get_function_name              0
#> 5 is_assign                      0

library(igraph)
edges <- dependency %>%
  select(Source, Target) %>%
  na.omit()
vertices <- unique(c(dependency$Source, dependency$Target))
vertices <- vertices[!is.na(vertices)]
g <- graph_from_data_frame(d = edges, vertices = vertices)
deg <- degree(g, mode = "in")
V(g)$size <- deg * 10 + 5
V(g)$label.cex <- (degree(g, mode = "in", normalized = TRUE) + 1)
plot(
  g,
  vertex.color = "grey",
  edge.color = "grey",
  edge.arrow.size = .4,
  main = "Functions dependency graph"
)

dependency <- find_dependencies(unique(functions$Function), envir = envir, in_envir = FALSE)
edges <- dependency %>%
  select(Source, Target) %>%
  na.omit()
vertices <- unique(c(edges$Source, edges$Target))
g <- graph_from_data_frame(edges)
deg <- degree(g, mode = "in")
V(g)$size <- deg
V(g)$label.cex <- (degree(g, mode = "in", normalized = TRUE) + 1) / 1.8
plot(
  g,
  vertex.color = "grey",
  edge.color = "grey",
  edge.arrow.size = .4,
  main = "Full functions dependency graph"
)