[R] Overlapping assignment

[Rob Steele]

Prof Brian Ripley wrote:
> On Mon, 24 Apr 2006, Rob Steele wrote:
> 
>> Is it valid to assign a slice of an array or vector to an overlapping
>> slice of the same object?  R seems to do the right thing but I can't
>> find anything where it promises to.
> 
> Yes.  Remember R is a vector language, so it is not doing this 
> index-by-index.  Also, what you are doing is
> 
> a <- `[<-`(a, 4:12, 1:9)
> 
> which should make this easier to understand: you get a new object, and you 
> don't expect
> 
> x <- x^2
> 
> to be a problem, do you?
> 
>>> a <- 1:12
>>> a[4:12] <- a[1:9]
>>> a
>>  [1] 1 2 3 1 2 3 4 5 6 7 8 9
>>
>>> b <- 1:12
>>> b[1:9] <- b[4:12]
>>> b
>>  [1]  4  5  6  7  8  9 10 11 12 10 11 12
>>

Thank you very much.  I understand that '[<-' is a function that 
produces a new "a" but I don't see that it necessarily follows that 
overlapping assignment is valid.  Even if R worked index-by-index (which 
at some level it must be doing) x <- x^2 would behave as expected.

It seems to me that R must either be smart about whether to copy left to 
right or right to left or else must copy the original object and then 
perform the destination indexing on the copy and the source indexing on 
the original.  Either way would keep it from overwriting a cell before 
it has a chance to copy it to its new location.  And if R does it by 
being smart about left and right, how does it handle index vectors that 
are out of order?  For example:

 > a <- 1:12
 > a[12:4] <-  a[9:1]
 > a
  [1] 1 2 3 1 2 3 4 5 6 7 8 9

It looks like it must be working with both the original and the copy 
during the assignment.