nakedpipe

Pipe into a sequence of calls without repeating the pipe symbol.

This is inspired by Stefan Bache and Hadley Wickham’s magrittr pipe and behaves mostly consistently.

nakedpipe calls are more compact, and are intended to be more readable, though it’s expected that they will look surprising to new users. The syntax allowed the development of many additional features that cannot be implemented as ergonomically with magrittr.

An instant translation addin between magrittr and nakedpipe is included.

It’s not yet on CRAN so you should install with :

remotes::install_github("moodymudskipper/nakedpipe")

General principles

A basic {nakedpipe} call looks a lot like a {magrittr} pipe chain, except that the piping symbol is not repeated, and that we surround the calls with {}

{magrittr} syntax :

library(magrittr)
cars %>%
  subset(speed < 6) %>%
  transform(time = dist/speed)
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

{nakedpipe} syntax :

library(nakedpipe)
cars %.% {
  subset(speed < 6)
  transform(time = dist/speed)
}
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

The dot insertion rules are identical to the ones used by {magrittr}, and likewise if we surround a step in {}, no dot will be inserted.

It plays well with left to right assignment:

cars %.% {
  subset(speed < 6)
  transform(time = dist/speed)
} -> res

Additional features include :

Side effects, using ~~, similar to magrittr::`%T>%
Temporary assignments and assignments to the calling environment
Shorthands for most common data manipulation operations, namely subset(), transform() and grouped transformations.
Conditional steps using if
Possibility to use {data.table} syntax for one step
Additional pipes to debug, assign in place, print or clock each step…

Side effects and assignments

Use ~~ for side effects:

cars %.% {
  subset(speed < 6)
  ~~ message("nrow:", nrow(.))
  transform(time = dist/speed)
}
#> nrow:2
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

This include assignments :

cars %.% {
  subset(speed < 6)
  ~~ cars_h <- . # or ~~ . -> cars_h
  transform(time = dist/speed)
}
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5
cars_h
#>   speed dist
#> 1     4    2
#> 2     4   10

To assign to a temp variable, use a dotted name:

cars %.% {
  ~~ .n <- 6
  subset(speed < .n)
  transform(time = dist/speed)
}
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5
exists(".n")
#> [1] FALSE

Data manipulation shorthands

For the very common subset() and transform() operations, shorthands are available, so that for our first example we could simply write:

cars %.% {
  speed < 6 # any call to < > <= >= == != %in% & | is interpreted as a subset call
  time = dist/speed # any call to = is interpreted as a transform call
}
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

Conditional steps

Use if for conditional step. if the condition is not TRUE and there is no else clause the data is unchanged:

cars %.% {
  subset(speed < 6)
  if(ncol(.) < 5) transform(time = dist/speed)
}
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

cars %.% {
  subset(speed < 6)
  if(ncol(.) > 5) transform(time = dist/speed)
}
#>   speed dist
#> 1     4    2
#> 2     4   10

Use data.table syntax

We can use data.table syntax for one step by using .dt[...], the output will be of the same class of the input (the temporary conversion to data.table is invisible):

cars %.% {
  speed < 8
  time = dist/speed
  .dt[, .(mmean_time = mean(time)), by = speed]
}
#>   speed mmean_time
#> 1     4   1.500000
#> 2     7   1.857143

We can chain data.table brackets too:

cars %.% {
  .dt[speed < 8][, time := dist/speed][,.(mmean_time = mean(time)), by = speed]
}
#>   speed mmean_time
#> 1     4   1.500000
#> 2     7   1.857143

Additional pipes

Assign in place using %<.%

cars_copy <- cars
cars_copy %<.% {
  head(2)
  ~~ message("nrow:", nrow(.))
  transform(time = dist/speed)
}
#> nrow:2
cars_copy
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

Clock each step using %L.%

cars %L.% {
  head(2)
  ~~ Sys.sleep(1)
  transform(time = dist/speed)
}
#> cars %L.% {
#>   head(2)
#>    user  system elapsed 
#>       0       0       0
#>   ~~Sys.sleep(1)
#>    user  system elapsed 
#>    0.00    0.00    1.02
#>   transform(time = dist/speed)
#>    user  system elapsed 
#>       0       0       0
#> }
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

print() the output of each step using %P.%

cars %P.% {
  head(2)
  transform(time = dist/speed)
}
#> cars %P.% {
#>   head(2)
#>   speed dist
#> 1     4    2
#> 2     4   10
#>   transform(time = dist/speed)
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5
#> }
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

View() the output of each step using %V.%

cars %V.% {
  head(2)
  transform(time = dist/speed)
}

%..% is faster at the cost of using explicit dots

cars %..% {
  head(.,2)
  transform(.,time = dist/speed)
}
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

It is better suited for programming and doesn’t support side effect notation but you can do :

cars %..% {
  head(.,2)
  {message("nrow:", nrow(.)); .}
  transform(.,time = dist/speed)
}
#> nrow:2
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

Create a function using %F.% on .

fun <- . %F.% {
  head(.,2)
  transform(.,time = dist/speed)
}
fun(cars)
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

Apply a sequence of calls on all elements using %lapply.%

replicate(2, cars, simplify = FALSE) %lapply.% {
  head(.,2)
  transform(.,time = dist/speed)
}
#> [[1]]
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5
#> 
#> [[2]]
#>   speed dist time
#> 1     4    2  0.5
#> 2     4   10  2.5

See ?"%.%" and ?"%lapply.%" to see all available pipes (including variants of the above).

Debugging

The %D.% pipe allows you to step through the calls one by one.

# Debug the pipe using `%D.%`
cars %D.% {
  head(2)
  transform(time = dist/speed)
}

You could also inster a browser() call as a side effect at a chosen step.

# Debug the pipe using `%D.%`
cars %D.% {
 head(2)
 ~~ browser()
 transform(time = dist/speed)
}

ggplot2

It’s a little known trick that you can use magrittr’s pipe with ggplot2 if you pipe to the + symbol. It is convenient if you want to use the ggplot object as the input of another function without intermediate variables of bracket overload :

library(ggplot2)
path <- tempfile()
cars %>%
  head() %>% 
  ggplot(aes(speed, dist)) %>%
  + geom_point() %>%
  + ggtitle("head(cars)") %>%
  saveRDS(path)

# rather than 
plt <- cars %>%
  head() %>% 
  ggplot(aes(speed, dist)) + 
  geom_point() +
  ggtitle("head(cars)")
saveRDS(plt, path)

The former case above shows operators on both sides, which looks a bit complicated, the latter requires a temporary variable and we must look at the end of the previous line to know what kind of piping was done.

In both cases additionally if I chose to comment out the ggtitle("head(cars)") line, I should also comment the last operator at the end of the previous line.

With nakedpipe we can write :

cars %.% {
  head()
  ggplot(aes(speed, dist))
  + geom_point()
  + ggtitle("head(cars)")
  saveRDS(path)
}

+ signs are neatly alligned, it’s obvious where the ggplot chain starts and ends, and trivial to pipe it to another instruction or to comment a line.

Conversion to magrittr syntax and back

We provide an addin to ease the conversion.

Running partial selection

It’s easy with standard pipes to run only the first steps of a pipe chain, by selecting them and running selected code. With {nakedpipe} the closing } is missing if we do the same. The addin “nakedpipe run incomplete call” allows one to run the selection after adding the closing }.

Benchmark

We’re a bit faster than {magrittr} 1.5, if you want to be even faster use %..% with explicit dots. Note that magrittr’s upcoming version is much faster than both, though keep in mind these are micro seconds and that the fastest solution is always not to use pipes at all. See below the benchmark using {magrittr} 2.0.

library(magrittr)
bench::mark(iterations = 10000,
  `%>%` = cars %>% 
    identity %>%
    identity() %>%
    identity(.) %>%
    {identity(.)},
  `%.%` = cars %.% {
    identity
    identity()
    identity(.)
    {identity(.)}
  },
  `%..%` = cars %..% {
    identity(.)
    identity(.)
    identity(.)
    {identity(.)}
   },
  `base` = {
    . <- cars
    . <- identity(.)
    . <- identity(.)
    . <- identity(.)
    . <- identity(.)
   }
)
#> # A tibble: 4 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 %>%           4.5us    5.1us   161773.        0B     16.2
#> 2 %.%         113.2us  124.3us     6258.      280B     22.6
#> 3 %..%         24.4us   27.2us    24060.        0B     14.4
#> 4 base          2.2us    2.4us   349727.        0B      0

Snippets

Runing setup_nakedpipe_snippets() will open RStudio’s snippet file so you can add our suggested snippets there. Follow the instructions and you’ll be able to type :

cars . # + 2 time the <tab> key

and display :

cars %.% {
  # with the cursor conveniently placed here
}

(or type .. to get the %..% equivalent)

Aknowledgements and similar efforts

{nakedpipe} is heavily inspired by {magrittr} and follows the same dot insertion rules.

The functions from *{dplyr}* and the tidyverse in general had a big influence on *{nakedpipe}*.

{data.table} is the package behind the .dt[...] syntax described above.

Alternative pipes are available on CRAN, at the time of writing and to my knowledge, in packages wrapr and pipeR. The latter includes a function pipeline() that allows piping a sequence of calls in a similar fashion as nakedpipe.

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

moodymudskipper / nakedpipe

Programming Languages