edlee123 / decorator Goto Github PK

A “decorator” is a function that wraps another function. This allows creating new functions effortlessly, and is a powerful tool to add functionality to functions as soon as they are defined.

For a runnable example, see the example vignette

For instance, let’s say we want a given function to be always called twice. This can of course be achieved easily by just executing the function body twice (for instance in a loop). However, a better way would be to just tell the function “hey, execute twice”; so that, when the function is called, it is automatically executed twice:

echo = function (msg)
    message(msg)

echo('hi')
# Desired output:
# hi
# hi

Since functions are objects in R and can be passed to other functions, we can “tell” echo to execute twice by modifying it:

echo = twice(echo)

Now we just need to define twice, which takes a function and returns a new function that calls the original function twice, forwarding all its arguments. However, the above is somewhat cumbersome because we first need to define echo and then modify it.

Alternatively, wrapping the original definition inside a function call works but becomes unreadable — especially once the function becomes longer:

echo = twice(
    function (msg)
        message(msg)
)

Now consider that we may wrap the function inside multiple other functions. Even worse. The “decorators” module allows us to write the following instead:

echo = twice %@% function (msg)
    message(msg)

A more realistic example might be a logged function that causes a function to be logged. Decorators can be chained, so we might write:

echo = logged('log.txt') %@% twice %@% function (msg)
    message(msg)

For comparison, in Python this would be written as

@logged('log.txt')
@twice
def echo(msg):
    print(msg)

The implementation of the decorator twice is trivial:

#' Call decorated function twice, in succession
twice = decorator %@% function (f) {
    function (...) {
        f(...)
        f(...)
    }
}

twice is a normal function that takes a function and returns a different function (which, when called, calls f twice). The only thing that’s special about twice is that it’s itself decorated by the decorator function. decorator makes a decorator out of a regular function.¹

logged is only slightly more complex:

#' Log each call to the decorated function
logged = function (file) decorator %@% function (f) {
    function (...) {
        cat(deparse(match.call()), '\n', file = file, append = TRUE)
        f(...)
    }
}

The “decorators” module takes care of pretty-printing decorated functions to hide the uninteresting decorator wrappers. Instead, the original function body is shown, along with the declared decorators (as one might expect intuitively):

print(echo)
# logged("log.txt") %@%
# twice %@%
# function (msg)
#     message(msg)
# <environment: 0x7fd7d4f6a3c0>

¹ The decorator decorator is necessary to allow easy chaining of decorators.

The cache decorator causes a function to cache its results: calling the function again with the same arguments returns the cached result without re-executing the function. This is useful when re-calculating the results is significantly more expensive than maintaining and querying a cache.

This can also be used to elegantly (if somewhat inefficiently) implement dynamic programming algorithms. Consider this implementation of the Fibonacci numbers:

fib = cache %@% function (n) if (n < 2) 1 else fib(n - 1) + fib(n - 2)

This function has roughly linear runtime and is thus asymptotically more efficient than a naive recursive implementation (i.e. the same implementation without caching), which has exponential runtime.

Computations on functions can be very useful and fit easily into the concept of R programming, since R treats functions as proper first-class objects and provides powerful introspection and metaprogramming capabilities. However, until now, systematically applying modifications to functions as soon as they are declared was syntactically not trivial. The decorator syntax encourages more widespread use of function modifications.

The concept of R function decorators is based on Python function decorators, which also inspired the syntax. The syntax is designed to be as unobtrusive as possible, and to blend into the normal function declaration syntax.