Strange bracket assignment call ('[<-') with matrix argument

Question

Recently I've stumbled upon this bit of code:

y <- NULL

y[cbind(1:2, 1:2)] <- list( list(1,2), list(2,3))

From the second answer here.

But it doesn't seem to differ from y <- list(...), as the comparisons below show:

> identical(y, list( list(1,2), list(2,3)))
[1] TRUE
> identical(y, y[cbind(1:2, 1:2)])
[1] FALSE

What is going on in the bracket assignment here? Why it doesn't throw an error? And why is it different from the non-assigment version in the last line of code?

Answer 1

Matrix indexing only applies when y has dim. Combine this with standard R recycling and the fact that all matrices are actually vectors, and this behavior makes sense.

When you initialize y to NULL, you ensure it has no dim. Therefore, when you index y by a matrix, say ind, you get the same results as having called y[as.vector(ind)]

identical(y[ind], y[as.vector(ind)])
# [1] TRUE

If there are repeat values in ind and you are also assigning, then for each index, only the last value assigned ot it will remain. For example Lets assume we are executing

y <- NULL; y[cbind(1:2, 2:1)] <- list( list(1,2), list(3,4) )
#   y has no dimension, so `y[cbind(1:2, 2:1)]` 
#   is the equivalent of   `y[c(1:2, 2:1)]`

When you assign y[c(1, 2, 2, 1)] <- list("A", "B") , in effect what happens is analogous to:

    y[[1]] <- "A"
    y[[2]] <- "B"
    y[[2]] <- "B"  # <~~ 'Overwriting' previous value 
    y[[1]] <- "A"  # <~~ 'Overwriting' previous value 

Here is a further look at the indexing that occurs: (Notice how the first two letters are being repeated)

ind <- cbind(1:2, 1:2)
L <- as.list(LETTERS)
L[ind]
# [[1]]
# [1] "A"
# 
# [[2]]
# [1] "B"
# 
# [[3]]
# [1] "A"
# 
# [[4]]
# [1] "B"

Here is the same thing, now with assignment. Notice how only the 3rd and 4th values being assigned have been kept.

L[ind] <- c("FirstWord", "SecondWord", "ThirdWord", "FourthWord")
L[ind]
# [[1]]
# [1] "ThirdWord"
# 
# [[2]]
# [1] "FourthWord"
# 
# [[3]]
# [1] "ThirdWord"
# 
# [[4]]
# [1] "FourthWord"

Try a different index for further clarity:

ind <- cbind(c(3, 2), c(1, 3))  ## will be treated as c(3, 2, 1, 3) 
L <- as.list(LETTERS)
L[ind] <- c("FirstWord", "SecondWord", "ThirdWord", "FourthWord")
L[1:5]
#  [[1]]
#  [1] "ThirdWord"
#  
#  [[2]]
#  [1] "SecondWord"
#  
#  [[3]]
#  [1] "FourthWord"
#  
#  [[4]]
#  [1] "D"
#  
#  [[5]]
#  [1] "E"

L[ind]
#  [[1]]
#  [1] "FourthWord"
#  
#  [[2]]
#  [1] "SecondWord"
#  
#  [[3]]
#  [1] "ThirdWord"
#  
#  [[4]]
#  [1] "FourthWord"

Edit regarding @agstudy's questions:

Looking at the src for [ we have the following comments:

• The special [ subscripting where dim(x) == ncol(subscript matrix)
• is handled inside VectorSubset. The subscript matrix is turned
• into a subscript vector of the appropriate size and then
• VectorSubset continues.

Looking at the function static SEXP VectorSubset(SEXP x, SEXP s, SEXP call) the relevant check is the following:

/* lines omitted */ 
attrib = getAttrib(x, R_DimSymbol);
/* lines omitted */
if (isMatrix(s) && isArray(x) && ncols(s) == length(attrib)) {
    /* lines omitted */
...