Sequential pipe operator?

[ggrothendieck]

Consider this example taken from SO, https://stackoverflow.com/questions/68131615/order-of-messages-to-console-within-functions-when-piped/68132380#68132380

testdata <- data.frame(a=c(3,5,7,1,2),b=c(2,5,6,3,9))

foo_1 <- function(DF){
  message("Hello 1")
  Obj <- DF %>% 
    mutate(Test_1=a*b)
  message("Bye 1")
  return(Obj)
} 

foo_2 <- function(DF){
  message("Hello 2")

  Obj <- DF %>% 
    mutate(Test_2=Test_1*2)
  message("Bye 2")
  return(Obj)
}

foo_3 <- function(DF){
  message("Hello 3")
  Obj <- DF %>% 
    mutate(Test_3=Test_2/a)

  message("Bye 3")
  return(Obj)
}

testdata %>% 
  foo_1() %>% 
  foo_2() %>% 
  foo_3()

The result of the above is that the message appear out of order. If we use the following then they appear in order.

testdata %>% 
  foo_1 %>% 
  { flush.console(); foo_2(.) } %>%
  { flush.console(); foo_3(.) }

Some way to address this would be desirable. %>% could issue a flush.console() or there could be a separate type of pipe that does that or there could be an option to control this.

[lionel-]

This is to be expected because of the lazy evaluation of function arguments in R. If you need sequential evaluation, add force(DF) at the start of your functions. You will get the intended order of messages for both f(a) and a |> f(). See the magrittr 2.0 release notes for more information https://www.tidyverse.org/blog/2020/11/magrittr-2-0-is-here/

[ggrothendieck]

The point is that there are features that make sense to be added to pipes to gain feature advantages over pure function composition.

[lionel-]

If you don't want to force inputs and need sequential piping you can use:

`%>%` <- magrittr::pipe_eager_lexical

[ggrothendieck]

So it already exists. I certainly missed that. Maybe magrittr could predefine a pipe operator for it?

[lionel-]

yes maybe, if there is interest in this.

Visitors please feel free to reply or react to this thread if this is something that you need.

[thuettel]

Since I was the OP of the linked Stackoverflow-Post I'd definitely appreciate implementing it.

[ggrothendieck]

Here is another application of the magrittr sequential operator. This would not work if we used %>% See https://stackoverflow.com/questions/68792277/pipeable-assign-function-for-r-does-not-assign-anything/68792380#68792380

`%s>%` <- pipe_eager_lexical
f3 <- function(x) x %s>% assign("thirteen", .) %>% { . + thirteen }

if (exists("thirteen")) rm(thirteen)
f3(13)
## [1] 26

[ggrothendieck]

Seems this comes up repeatedly on SO. Here is another relevant question there: https://stackoverflow.com/questions/68858736/access-result-later-in-pipe/68859879#68859879

[ggrothendieck]

Another example: https://stackoverflow.com/questions/70848367/r-in-non-standardard-evaluation-nse-how-do-i-determine-the-right-calling-env/70850216#70850216