For performance reasons, R delays copying as long as possible. Consider this code:
x <- rnorm(10)
y <- xThere is no reason for R to copy the elements of x into y at this point. It is sufficient - and much faster - for y to hold a reference to x. If this is the only thing that’s ever done with y
print(y[4])
print(y[7])there is no reason to ever make a copy of x, because a reference is just fine until there’s mutation of x or y. This is a problem if you’re working with y from D. If you did the above, and then you changed the elements of y from inside a D function, you’d actually be changing the elements of both x and y, because there’s nothing that triggers the copy.
The library handles all of this for you. If you’re doing the calls to R yourself, though, you’ll have to account for it. One way is to use braces when you do the copy. Here’s an example for a matrix:
m <- matrix(1:4, ncol=2)
mm <- m[]Or you can do an assignment after creating mm:
m <- matrix(1:4, ncol=2)
mm <- m
mm[1,1] <- mm[1,1]As soon as you do the assignment, mm is no longer a reference. You can see an example of this in testing/memory.d.