Kevin wrote a clever defer()
function in a pull request two months ago, and it almost killed me when I first saw it.1 It looked so short, but I’m really bad at reasoning code that uses a combination of substitute()
, evalq()
, and do.call()
with the envir
argument, especially when substitute()
is used three times, and there are three envir
arguments.
I was once confused by on.exit()
, so the defer()
function was completely magic to me. Today I found that I needed something similar – I wanted to insert an on.exit()
call to a parent function. I know how on.exit()
works in the current function and have used it many times before, but I have never tried to inject on.exit()
to a parent function, so I looked at Kevin’s code again, and simplified it a little bit (hopefully):
f = function() {
g()
print('printing in f()') # <- 2
}
g = function() {
x = 'an exit call added to f() by g()'
on.exit(print('exiting g()'), add = TRUE) # <- 1
do.call(
on.exit, list(substitute(print(x)), add = TRUE),
envir = parent.frame() # the key is this envir
) # <- 3
}
f()
# [1] "exiting g()"
# [1] "printing in f()"
# [1] "an exit call added to f() by g()"
You can see that first g()
exits, then the print()
expression in f()
is executed, and finally f()
exits, with the exit call added by g()
executed.
I was looking for this because I have several functions that should do the same thing on exit, and I don’t want to repeat the code on.exit()
in all these functions. It seems do.call(on.exit, envir = parent.frame())
solved my problem.
-
This PR is one of the greatest contributions in the history of R packages. I’ll talk about it next year. Some of you may be sleepless because of it. ↩︎