[Query] Dependencies

Generated from 'wiki-query.ts' on 2026-11-08, 09:20:49 UTC (v2.17.4), do not edit directly.

Returns all direct dependencies (in- and outputs) of a given R script
This query is requested with the type dependencies.
Run in the REPL: :query @dependencies (<crit>;...) <code | file://path>

This query extracts all dependencies from an R script, using a combination of a Call-Context Query and more advanced tracking in the Dataflow Graph. Loaded libraries are resolved against the signature database.

The query reports one list of entries per category, each entry a call and the value it reads or writes, like the file of a read.csv:

What the value of an entry tells, and the fields that say more:

For a script reading library(x), the following query returns the loaded library:

[ { "type": "dependencies" } ]

(This can be shortened to @dependencies when used with the REPL command :query).

Results (prettified and summarized):

Query: dependencies (3 ms)
   Libraries (1)
     x via library (node 3)

{"dependencies":{".meta":{},"library":[{"nodeId":3,"functionName":"library","value":"x"}],"remote":[],"source":[],"read":[],"write":[],"visualize":[],"transform":[],"test":[],"statistics":[]},".meta":{}}

The query also handles more complicated scripts and offers information on the loaded libraries, sourced files, data which is read and data which is written. For example, consider the following script:

source("sample.R")
foo <- loadNamespace("bar")

data <- read.csv("data.csv")

#' @importFrom ggplot2 ggplot geom_point aes
ggplot(data, aes(x=x, y=y)) + geom_point()

better::write.csv(data, "data2.csv")
print("hello world!")

The following query returns the dependencies of the script.

[ { "type": "dependencies" } ]

(This can be shortened to @dependencies when used with the REPL command :query).

{"dependencies":{".meta":{},"library":[{"nodeId":8,"functionName":"loadNamespace","value":"bar"},{"nodeId":32,"functionName":"::","value":"better"}],"remote":[],"source":[{"nodeId":3,"functionName":"source","value":"sample.R"}],"read":[{"nodeId":14,"functionName":"read.csv","assignedTo":"data","value":"data.csv"}],"write":[{"nodeId":41,"functionName":"print","value":"stdout"}],"visualize":[{"nodeId":28,"functionName":"ggplot","inputVariables":["data","x","y"],"parts":[30]},{"nodeId":30,"functionName":"geom_point","linkedIds":[28],"inputVariables":["data","x","y"]}],"transform":[],"test":[],"statistics":[]},".meta":{}}

Currently, the dependency extraction may fail as it is essentially a set of heuristics guessing the dependencies. We welcome any feedback on this (consider opening a new issue).

In the meantime we offer several properties to overwrite the default behavior (e.g., function names that should be collected)

[ { "type": "dependencies",   "ignoreDefaultFunctions": true,   "enabledCategories": [ "library" ],   "libraryFunctions": [ { "package": "base",     "name": "print",     "argIdx": 0,     "argName": "library",     "resolveValue": true } ] } ]

{"dependencies":{".meta":{},"library":[{"nodeId":41,"functionName":"print","value":"hello world!"}],"remote":[],"source":[],"read":[],"write":[],"visualize":[],"transform":[],"test":[],"statistics":[]},".meta":{}}

Here, resolveValue tells the dependency query to resolve the value of this argument in case it is not a constant.

By default the query reports only the dependencies the code names itself. Yet R attaches a handful of base packages to the search path on startup, so a bare sd(x) genuinely depends on stats without any library call saying so. Set assumedPackages to have those reported as well, as library entries marked implicit and carrying no nodeId (no single call stands for "R attached this"), with the calls that pulled the package in listed as linkedIds:

[ { "type": "dependencies",   "assumedPackages": true,   "enabledCategories": [ "library" ] } ]

Results (prettified and summarized):

Query: dependencies (1 ms)
   Libraries (2)
     base always attached by R, used at 8
     stats attached by R at startup, used at 10

{"dependencies":{".meta":{},"library":[{"functionName":"<attached>","value":"base","implicit":true,"alwaysAttached":true,"linkedIds":[8]},{"functionName":"<attached>","value":"stats","implicit":true,"linkedIds":[10]}],"remote":[],"source":[],"read":[],"write":[],"visualize":[],"transform":[],"test":[],"statistics":[]},".meta":{}}

sd(c(1, 2, 3))

base is reported alongside the others but additionally marked alwaysAttached. The other six are attached by convention and R_DEFAULT_PACKAGES (or options(defaultPackages=)) can drop any of them, whereas base is always there and cannot be detached -- so it is never something the script could have asked for, and never something to suggest a library call for.

To see what takes part in a dependency, ask the query to focus on it: .meta.focus then lists every node the dependency depends on and every node that depends on it, so a client can gray out the rest. In the REPL, the criteria go in parentheses, as with :query @dependencies (3@read.csv), which prints the lines taking part:

[ { "type": "dependencies",   "focus": [ "3@read.csv" ],   "enabledCategories": [ "read" ] } ]

{"dependencies":{".meta":{"focus":[10,8,0,2,1,11,6,18,20]},"library":[],"remote":[],"source":[],"read":[{"nodeId":10,"functionName":"read.csv","assignedTo":"d","value":"a.csv"}],"write":[],"visualize":[],"transform":[],"test":[],"statistics":[]},".meta":{}}

f <- "a.csv"
g <- 2
d <- read.csv(f)
h <- g + 1
print(d)

Beyond libraries, files and outputs, the query knows the visualize category (plots) and the transform category (Data Transformations): the calls that reshape, filter, join or aggregate data, like dplyr::filter, tidyr::pivot_longer, merge, aggregate or data.table::setDT. The category is read from the built-in configuration: every built-in labeled SemanticCallTag.Transforms belongs to it. Every such built-in names its package, so filter is only a transformation when it resolves to dplyr and never for the time-series stats::filter.

The read, write, visualize, and transform entries additionally state how data flows through the call, resolved with the dataflow graph instead of the syntax:

• inputVariables (for write, visualize, and transform) are the variables the result is computed from, through every call in between. What an argument's role in the signature excludes (a file, a flag, a callee) is no input, nor is what a call quotes or looks up in a data mask: df for df |> filter(x > 1), both a and b for left_join(a, b). An addon to a plot draws its data.
• assignedTo (for read, visualize, and transform) is the variable the result is stored in, following every call that hands it on (a pipe into a call, parentheses, a + adding to a plot) up to <-, = or ->: small for small <- filter(df, x > 1), and nothing for f(filter(df, x > 1)).
• Every plot additionally carries its title (from ggtitle, labs(title = ) or main =) and the file it is written to as savedTo (by ggsave, which names its plot or saves last_plot(), or by the graphics device open when it is drawn).

[ { "type": "dependencies",   "enabledCategories": [ "transform",    "visualize" ] } ]

Results (prettified and summarized):

Query: dependencies (3 ms)
   Visualizations (3)
      via ggplot (node 38)
      via geom_point (node 40, linked 38)
      via ggtitle (node 45, linked 38)
   Data Transformations (2)
      via filter (node 16)
      via select (node 24)

{"dependencies":{".meta":{},"library":[],"remote":[],"source":[],"read":[],"write":[],"visualize":[{"nodeId":38,"functionName":["ggplot","ggplot2",false],"inputVariables":["small"],"assignedTo":"p","parts":[40,45],"title":"Small"},{"nodeId":40,"functionName":["geom_point","ggplot2",false],"linkedIds":[38],"assignedTo":"p","inputVariables":["small"]},{"nodeId":45,"functionName":["ggtitle","ggplot2",false],"linkedIds":[38],"assignedTo":"p","inputVariables":["small"]}],"transform":[{"nodeId":16,"functionName":["filter","dplyr",false],"inputVariables":["measurements"]},{"nodeId":24,"functionName":["select","dplyr",false],"inputVariables":["measurements"],"assignedTo":"small"}],"test":[],"statistics":[]},".meta":{}}

library(dplyr)
library(ggplot2)
small <- measurements |> filter(x > 1) |> select(x, y)
p <- ggplot(small, aes(x, y)) + geom_point() + ggtitle("Small")
ggsave("small.png", plot = p)