Programmatically switch package in `::` call in R

user_5276212 0
Viewed 275

Given a call to a function bar::foo(), I would like to be able to programmatically switch the package bar so that the same syntax calls hello::foo().

An example:

  • Let's say I have three packages, parentPkg, childPkg1 and childPkg2.
  • In parentPkg I have a call to function childPkg1::foo()
  • foo() is also a function in childPkg2

  • I would like to be able, in parentPkg to use the :: operator to call foo() but to programatically switch the package name. Something like:

    dummy_pkg_name = ifelse(scenario=="child1", "childPkg1", "childPkg2")
dummy_pkg_name::foo()

Is it possible? How do I achieve it?

Some context

parentPkg is a function that interacts with a web application, takes some request and data and returns results from different statistical models depending on the scenarios.
Each scenario is quite complex and not everything can be generalised in parentPkg. For this reason, childPkg1 and childPkg2 (actually there are also 3 and 4) are sort of sub-packages that deals with the data cleaning and various alternatives for each scenario but return the same class of value.
The idea is that parentPkg would switch the package to the pertinent child depending on the scenario and call all of the necessary functions without having to write the same sequence for each child but just with a slightly different :: call.

r package
3 Answers

Since :: can be seen as a function, it looks like

`::`(dummy_pkg_name, foo)()

is what you want. Alternatively,

getFromNamespace("foo", ns = dummy_pkg_name)()

For instance,

`::`(stats, t.test)
# function (x, ...) 
# UseMethod("t.test")
# <bytecode: 0x102fd4b00>
# <environment: namespace:stats>

getFromNamespace("t.test", ns = "stats")
# function (x, ...) 
# UseMethod("t.test")
# <bytecode: 0x102fd4b00>
# <environment: namespace:stats>
user_13205350

You could also create a call() that could then be evaluated.

call("::", quote(bar), quote(foo()))
# bar::foo()

Put into use:

c <- call("::", quote(stats), quote(t.test))
eval(c)
# function (x, ...) 
# UseMethod("t.test")
# <bytecode: 0x4340988>
# <environment: namespace:stats>

Wrapped up in a function using setdiff as our default function:

f <- function(pkg, fn = setdiff) {
    pkg <- substitute(pkg)
    fn <- substitute(fn)
    eval(call("::", pkg, fn))
}

f(base)
# function (x, y) 
# {
#     x <- as.vector(x)
#     y <- as.vector(y)
#     unique(if (length(x) || length(y)) 
#         x[match(x, y, 0L) == 0L]
#     else x)
# }
# <bytecode: 0x30f1ea8>
# <environment: namespace:base>

f(dplyr)
# function (x, y, ...) 
# UseMethod("setdiff")
# <environment: namespace:dplyr>
user_30639100

To adhere to KISS, simply re-assign to new named functions in global environment. Be sure to leave out () since you are not requesting to run the function. 

parent_foo <- parentPkg::foo
child1_foo <- childPkg1::foo
child2_foo <- childPkg2::foo
child3_foo <- childPkg3::foo

Then, conditionally apply them as needed:

if (scenario=="child1") {
  obj <- child1_foo(...)
} 
else if (scenario=="child2") {
  obj <- child2_foo(...)
} 
...
user_14224510
Related