Use container for code development

Robust usage

Base R’s list operations are designed for interactive use, offering flexibility but often being overly permissive with user input. While convenient, this can lead to subtle bugs during development and requires additional, sometimes tedious, checks to ensure code robustness.

The {container} package addresses these challenges by providing operations that explicitly define the intent of each action. By prioritizing clarity and precision, {container} enables you to write leaner and more reliable code right from the start.

This vignette revisits some of the basic operations familiar from base R lists and demonstrates how {container} enhances them with strict validation and powerful additional features.

co <- container()

Add

Using base R lists notation, elements are usually added by name or concatenation.

co[["x"]] <- 1
co <- c(co, 2)

co
# [x = 1, 2]

The {container} package provides the add function to add elements.

co <- add(co, x = 3)  # same as c(co, x = 3)

co
# [x = 1, 2, x = 3]

For container objects there is not much of a difference between the two methods. Now, if for example you don’t want to allow duplicated names, you can use dict objects instead. These are a subclass of container and would throw an error in this case.

d <- dict(x = 1)

add(d, x = 3)
# Error: name 'x' exists already

For more details see the reference documentation or have a look at the Deque, Set, and Dict vignette. Lastly, note that the base append function also works with containers.

append(co, 1.5, after = 1)
# [x = 1, 1.5, 2, x = 3]

Replace

As demonstrated before, elements can be loosely replaced by index or name.

co[["x"]] <- 0
co[[2]] <- 12

co
# [x = 0, 12, x = 3]

Also, in contrast to base lists, the container will not allow to add elements at positions longer than the length of the object.

co[[4]] <- 3
# Error: index out of range (length = 3): 4

If the name does not exist, the element is appended as known from base lists.

co[["y"]] <- 5

co
# [x = 0, 12, x = 3, y = 5]

Strict replace

Let’s imagine you want to replace an element of a certain name, and therefore expect that the name exists already. In code development, this would require an additional check, for example:

name <- "z"
if (name %in% names(co)) {
  co[[name]] <- 10
} else {
  stop("Name '", name,  "' does not exist.")
}

Clearly this is a lot of boilerplate code for a simple operation, and it is easy to forget such checks. In addition, you end up with a lot of unit tests basically to check the checks. Last but not least, the intent of the code is not as clear as it could be.

This is where the {container} package comes in. If you want to make sure that something is replaced, {container} provides the function replace_at, which will only replace elements at names or positions that exist. The following statements are all equal and show the different possibilities on how to use replace_at.

replace_at(co, x = 10, y = 13)            # name = value pairs
# [x = 10, 12, x = 3, y = 13]

replace_at(co, c("x", "y"),  c(10, 13))   # names followed by values
# [x = 10, 12, x = 3, y = 13]

replace_at(co, c(1, 4),      c(10, 13))   # positions followed by values
# [x = 10, 12, x = 3, y = 13]

replace_at(co, list(1, "y"), c(10, 13))   # mixed indices followed by values
# [x = 10, 12, x = 3, y = 13]

Next, let’s see how invalid indices are signaled.

replace_at(co, z = 10)
# Error: names(s) not found: 'z'

replace_at(co, "z", 10)
# Error: names(s) not found: 'z'

replace_at(co, 5, 10)
# Error: index out of range (length = 4): 5

If you instead don’t mind that elements at new names will be added, set .add = TRUE. Invalid positional indices are still signaled.

co <- replace_at(co, z = 10, .add = TRUE)   # ok

co <- replace_at(co, 7, 10, .add = TRUE)
# Error: index out of range (length = 5): 7

co
# [x = 0, 12, x = 3, y = 5, z = 10]

Strict replace by value

It is also possible to replace elements by value, that is, you specify the value (not the index) that should be replaced. To see this, let’s replace 12 (located at the 2nd postion) by "foo" and then y = 5 (located at the 4th position) by 1:2.

co <- replace(co, old = 12, new = "foo")
co
# [x = 0, "foo", x = 3, y = 5, z = 10]

co <- replace(co, old = 5, new = 1:2)
co
# [x = 0, "foo", x = 3, y = (1L 2L), z = 10]

Implementing this “manually” would require even more additional code as before. As intended, if the value does not exist, an error is signaled.

replace(co, old = "non-existent-value", new = "my value")
# Error: old element ("non-existent-value") is not in Container

Again, the intend that you want to replace but don’t mind that the element is added can be specified:

replace(co, old = "non-existent-value", new = "my value", add = TRUE)
# [x = 0, "foo", x = 3, y = (1L 2L), z = 10, "my value"]

Extract

Let’s recap the standard extract operators.

co[[1]]
# [1] 0

co[["x"]]
# [1] 0

co[3:5]
# [x = 3, y = (1L 2L), z = 10]

co[c("x", "y", "z")]
# [x = 0, y = (1L 2L), z = 10]

Strict extract

The {container} functions to strictly select one or multiple elements are named at2 and at.1

at2(co, 1)
# [1] 0

at2(co, "x")
# [1] 0

at(co, 3:5)
# [x = 3, y = (1L 2L), z = 10]

at(co, c("x", "y", "z"))
# [x = 0, y = (1L 2L), z = 10]

As before you can specify mixed indices via lists.

indices <- list("x", 4, "z")

at(co, indices)
# [x = 0, y = (1L 2L), z = 10]

Accessing non-existent names or positions is signaled with an error as follows.

at2(co, 10)
# Error: index 10 exceeds length of Container, which is 5

at2(co, "a")
# Error: index 'a' not found

at(co, 3:6)
# Error: index 6 exceeds length of Container, which is 5

at(co, c("x", "a"))
# Error: index 'a' not found

Be reminded that with base lists non-existent indices just would have returned NULL values.

l <- list()

l[2:3]
# [[1]]
# NULL
# 
# [[2]]
# NULL

l[["a"]]
# NULL

If needed, the (less strict) list access can be mimicked with peek_at and peek_at2.

co
# [x = 0, "foo", x = 3, y = (1L 2L), z = 10]

peek_at(co, 10, 11)
# []

peek_at(co, 5:10)
# [z = 10]

peek_at2(co, "a")
# NULL

As you see, one important difference is that multiple access via peek_at by default instead of NULL values just returns nothing. However, both functions allow to specify a custom default value being returned if the index does not exist.

co
# [x = 0, "foo", x = 3, y = (1L 2L), z = 10]

peek_at2(co, "a", default = -1)
# [1] -1

peek_at(co, "z", "a", .default = -1)
# [z = 10, a = -1]

peek_at(co, 4:8, .default = NA)
# [y = (1L 2L), z = 10, NA, NA, NA]

Remove

To remove elements in lists, they are usually replaced by NULL.

l <- list(a = 1)

l
# $a
# [1] 1

l[["a"]] <- NULL

l
# named list()

With the container package this is done differently, as replacing by NULL will not delete the element but literally replace it by NULL.

co[["x"]] <- NULL

co
# [x = NULL, "foo", x = 3, y = (1L 2L), z = 10]

Instead, elements can be deleted by index (delete_at) or value (delete) as follows.

co
# [x = NULL, "foo", x = 3, y = (1L 2L), z = 10]

delete_at(co, 1, "y", "z")
# ["foo", x = 3]

delete(co, NULL, 1:2, 10)   # same but remove by value
# ["foo", x = 3]

As before, invalid indices or missing values are signaled.

co
# [x = NULL, "foo", x = 3, y = (1L 2L), z = 10]

delete_at(co, "a")
# Error: names(s) not found: 'a'

delete_at(co, 10)
# Error: index out of range (length = 5): 10

delete(co, 1:3)
# Error: (1L 2L 3L) is not in Container

If you need a less strict delete operation, use the discard functions, which delete all valid indices or values and ignore the rest.

co
# [x = NULL, "foo", x = 3, y = (1L 2L), z = 10]

discard_at(co, 1, "a")
# ["foo", x = 3, y = (1L 2L), z = 10]

discard_at(co, 1:100)
# []

discard(co, NULL, 1:2, 1:3, 1:4)    # discard by value
# ["foo", x = 3, z = 10]

Combine containers

The update function is used to combine/merge two containers.

c1 <- container(1, b = 2)
c2 <- container(   b = 0, c = 3)

update(c1, c2)
# [1, b = 0, c = 3]

update(c2, c1)
# [b = 2, c = 3, 1]

With the container package this function is also provided for base R lists.

l1 <- list(1, b = 2)
l2 <- list(   b = 0, c = 3)

update(l1, l2)
# [[1]]
# [1] 1
# 
# $b
# [1] 0
# 
# $c
# [1] 3

update(l2, l1)
# $b
# [1] 2
# 
# $c
# [1] 3
# 
# [[3]]
# [1] 1

Note that there is a similar function utils::modifyList, which, however, in contrast to update, does not “forward” unnamed elements.

modifyList(l1, l2)
# [[1]]
# [1] 1
# 
# $b
# [1] 0
# 
# $c
# [1] 3

modifyList(l2, l1)  # drops l1[[1]] = 1
# $b
# [1] 2
# 
# $c
# [1] 3

Also, while utils::modifyList modifies a list recursively by changing a subset of elements at each level, update just works on the first level.

l1 <- list(a = 1, b = list(c = "a", d = FALSE))

l2 <- list(e = 2, b = list(d = TRUE))

modifyList(l1, l2)  # modifies l1$b$d from FALSE to TRUE
# $a
# [1] 1
# 
# $b
# $b$c
# [1] "a"
# 
# $b$d
# [1] TRUE
# 
# 
# $e
# [1] 2

update(l1, l2)      # replaces l1$b by l2$b
# $a
# [1] 1
# 
# $b
# $b$d
# [1] TRUE
# 
# 
# $e
# [1] 2

Functional programming

The apply family and common higher-order functions both can be used with containers as usual.

co <- container(a = 1, b = 2, c = 3, d = 4)

sapply(co, function(x) x^2)
#  a  b  c  d 
#  1  4  9 16

Filter(co, f = function(x) x > 2)
# [c = 3, d = 4]

Reduce(co, f = sum)
# [1] 10

Summary

This vignette demonstrates how {container} enhances robust code development by providing:

  • Clear and intent-driven operations for adding, replacing, extracting, and removing elements, minimizing boilerplate and potential errors.
  • Strict validation methods like replace_at() and at() for safer and more precise modifications and access.
  • Flexible tools like peek_at() and discard() for handling invalid or non-existent indices gracefully.
  • Safe merging of containers and lists with update()
  • Full compatibility with functional programming tools like sapply(), Filter(), and Reduce() for streamlined workflows.

To see how some of the functions disussed here are applied with derived data structures, see:

  1. Resembling R base-internal .subset2 and .subset.↩︎