[Rd] treatment of zero and negative elements in matrix indices

[Tony Plate]

Matrix indexing seems to give rather "variable" results when zeros or 
negative values are included among the indices (in terms of both error 
messages and in terms of the number of returned values when there is no 
error message).

Is this the intended behavior?

I couldn't see any comments about zeros or negative values in matrix 
indices in either ?"[" or Section 3.4.2 "Indexing matrices and arrays" 
of the "R Language Definition" (Version 2.0.1), or in Section 5.3 "Index 
arrays" of "An Introduction to R" (Version 2.1.0).

(It looks like the special treatment of zeros and negative indices is 
being applied to the vector indices after they are derived by standard 
indexing arithmetic from the matrix indices -- so zeros and negative 
values in the matrix indices have no special meaning.  There does seem 
to be some range checking, but it's not consistent -- sometimes index 
values that are out of range get through.)

 > x <- matrix(1:6,ncol=2)
 > dim(x)
[1] 3 2
 > x
      [,1] [,2]
[1,]    1    4
[2,]    2    5
[3,]    3    6
 > x[rbind(c(1,1), c(2,2))]
[1] 1 5
 > x[rbind(c(1,1), c(2,2), c(0,1))]
[1] 1 5
 > x[rbind(c(1,1), c(2,2), c(0,0))]
Error: only 0's may mix with negative subscripts
 > x[rbind(c(1,1), c(2,2), c(0,2))]
[1] 1 5 3
 > x[rbind(c(1,1), c(2,2), c(0,3))]
Error: subscript out of bounds
 > x[rbind(c(1,1), c(2,2), c(1,0))]
Error: only 0's may mix with negative subscripts
 > x[rbind(c(1,1), c(2,2), c(2,0))]
Error: only 0's may mix with negative subscripts
 > x[rbind(c(1,1), c(2,2), c(3,0))]
[1] 1 5
 > x[rbind(c(1,1), c(2,2), c(1,2))]
[1] 1 5 4
 > x[rbind(c(1,1), c(2,2), c(-1,2))]
[1] 1 5 2
 > x[rbind(c(1,1), c(2,2), c(-2,2))]
[1] 1 5 1
 > x[rbind(c(1,1), c(2,2), c(-3,2))]
[1] 1 5
 > x[rbind(c(1,1), c(2,2), c(-4,2))]
Error: only 0's may mix with negative subscripts
 > x[rbind(c(1,1), c(2,2), c(-1,-1))]
Error: subscript out of bounds
 >
 > # range checks are at least sometimes not applied
 > x <- matrix(1:6, ncol=3)
 > dim(x)
[1] 2 3
 > x[rbind(c(1,1), c(2,2), c(-3,3))]
[1] 1 4 1
 > x[rbind(c(1,1), c(2,2), c(-4,3))]
[1] 1 4
 >


 > version
          _
platform i386-pc-mingw32
arch     i386
os       mingw32
system   i386, mingw32
status
major    2
minor    0.1
year     2004
month    11
day      15
language R
 >

(I see the same behavior in R 2.1.0 beta of 2005/04/04)

-- Tony Plate

PS: FWIW, S-PLUS 6.2 does something more like one would get when using 
each row of the index matrix as arguments to a call to "[", except that 
negative values are not allowed at all (any zero value in a row causes 
that row to be omitted from the result).