In WinVector/wrapr: Wrap R Tools for Debugging and Parametric Programming

The wrapr dot arrow pipe was designed to both be an effective R function application pipe and also an experimental test-bed for pipe effects.

Let's take a look at implementing a new effect from a user perspective. The idea we want to implement is delayed evaluation through a collecting object we call a "locum" or stand-in. The locum is intended to collect operations without executing them, for later use. This is similar to a lambda or function abstraction. The code is now in the wrapr package, but could be implemented by any user as it uses only public or semi-public wrapr interfaces.

The locum was for a while a part of wrapr, but now it is user code in the file locum.R.

Let's start loading the wrapr package.

library(wrapr)
source("locum.R")

We can use the locum to collect the operations, and then print them.

y <- 4
p <- locum() %.>% 
  sin(.) %.>% 
  cos(.) %.>% 
  atan2(., y)

print(p)

We can now replace the locum with the value we want to apply the pipeline to.

5 %.>% p

This yields the same answer as the following function application.

atan2(cos(sin(5)), 4)

We can also add later intended arguments to the pipeline formatting.

print(p, 'start' = 4)

We can do some fun things, such as combining locum pipelines.

p1 <- locum() %.>% sin(.)
p2 <- locum() %.>% cos(.)

p12 <- p1 %.>% p2
p12

4 %.>% p12

cos(sin(4))

The idea is: wrapr dot arrow pipe is designed for expansion through the apply_right, apply_left S3 interfaces, and the apply_right_S4 S4 interface. And users can access the implementation through the pipe_impl function. This example and the formal article should give users a good place to start. rquery, rqdatatable, and cdata already use the extension interfaces to implement interesting features.