Functions with identical body but different enclosing environments

[bluaze]

To solve #38, which is different functions have same parameters (including empty parameter), body of the function to be memoised is introduced in digest in 70d959b
To solve #58 #61, which is srcref will differ across sources, including different computers, body of the function is converted to character in 611cfad

However, when two functions are identical except of their enclosing environments, problem arises

fn <- function(x) function(y) x+y

digest::digest(fn(1))
digest::digest(fn(2))

digest::digest(body(fn(1)))
digest::digest(body(fn(2)))

As examples above demonstrate, the two functions have different digests themselves, but identical digest of the bodies

As there is a function called removeSource in utils, which exactly removes all source related attributes of a function, it might be appropriate to use removeSource(fn) instead of as.character(body(fn))

digest::digest(utils::removeSource(fn(1)))
digest::digest(utils::removeSource(fn(2)))

[jimhester]

Good suggestion! Want to create a Pull Request with this change?

[bluaze]

if enclosing environment is introduced in hashing function to be memoisd, not just body, functions relies on their enclosing environments will caching rightly. but functions that doesn't rely on enclosing environment might behave unexpectedly, as show below

test_env <- new.env()

evalq({
  fn <- function(x) { x+1 }
  
  i <- 0
  cat(digest::digest(removeSource(fn)),"\n")
  
  i <- 1
  cat(digest::digest(removeSource(fn)),"\n")
  
},test_env)

Although fn doesn't rely on any values in test_env, the digest changes as an irrelevant variant i changes. however, if these code execute in global environment directly, the digest won't change, as serialize used by digest underlying wouldn't serialize the global environment.

as memoization might be more suitable for functions that are referentially transparent, it could be not appropriate to support functions rely on enclosing environment