brio - Basic R Input Output

Codecov test coverage R build status

Functions to handle basic input output, these functions always read and write UTF-8 files and provide more explicit control over line endings.

Reading files

library(brio)
#> 
#> Attaching package: 'brio'
#> The following objects are masked from 'package:base':
#> 
#>     readLines, writeLines
write_lines(c("abc", "123"), "my-file")

# Write with windows newlines
write_lines(c("abc", "123"), "my-file-2", eol = "\r\n")

file_line_endings("my-file")
#> [1] "\n"

file_line_endings("my-file-2")
#> [1] "\r\n"

read_lines("my-file")
#> [1] "abc" "123"

unlink(c("my-file", "my-file-2"))

Drop-ins

brio also has readLines() and writeLines() functions drop-in replacements for base::readLines() and base::writeLines(). These functions are thin wrappers around brio::read_lines() and brio::write_lines(), with deliberately fewer features than the base equivalents. If you want to convert a package to using brio you can add the following line and re-document.

#' @importFrom brio readLines writeLines

Benchmarks

Speed is not necessarily a goal of brio, but it does end up being a nice side effect.

gen_random <- function(characters, num_lines, min, max) {
  line_lengths <- sample.int(max - min, num_lines, replace = TRUE) + min
  vapply(line_lengths, function(len) paste(sample(characters, len, replace = TRUE), collapse = ""), character(1))
}

set.seed(42)

# generate 1000 random lines between 100-1000 characters long
data <- gen_random(letters, 1000, min = 100, max = 1000)

brio::write_lines(data, "benchmark")

Reading

Reading speeds are a decent amount faster with brio, mainly due to larger block sizes and avoidance of extra copies.

bench::mark(
  brio::read_lines("benchmark"),
  readr::read_lines("benchmark"),
  base::readLines("benchmark")
)
#> # A tibble: 3 x 6
#>   expression                          min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                     <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 brio::read_lines("benchmark")  686.09µs 715.66µs     1333.    8.05KB        0
#> 2 readr::read_lines("benchmark")   2.56ms   2.68ms      363.    9.07MB        0
#> 3 base::readLines("benchmark")     3.61ms   3.77ms      256.   31.39KB        0

Writing

Write speeds are basically the same regardless of method, though brio does avoid some extra memory allocations.

bench::mark(
  brio::write_lines(data, "benchmark"),
  readr::write_lines(data, "benchmark"),
  base::writeLines(data, "benchmark"),
  check = FALSE
)
#> # A tibble: 3 x 6
#>   expression                                 min   median `itr/sec` mem_alloc
#>   <bch:expr>                            <bch:tm> <bch:tm>     <dbl> <bch:byt>
#> 1 brio::write_lines(data, "benchmark")   812.1µs   1.18ms     835.         0B
#> 2 readr::write_lines(data, "benchmark")   15.1ms  17.82ms      55.4     589KB
#> 3 base::writeLines(data, "benchmark")    846.5µs    1.4ms     669.         0B
#> # … with 1 more variable: gc/sec <dbl>

unlink("benchmark")