Casting from double to integer does not handle overflow correctly

[akbertram]

As noticed by @kalibera, overflow during conversion from double to integer is not handled correctly.

Most basically, the expression as.integer(2^40) should overflow to NA, not the integer max value.

In cases where double-precision values are used as indices, it gets more complicated. For example, in the code below:

a <- as.numeric(1:1000000)
c <- a[c(-1*sum(a), sum(a))]

Both -1 * sum(a) and sum(a) would overflow to NA when cast to 32-bit integers, but is still within the range of a 64-bit integer, and so in GNU R, this leads to

Error in a[c(-1 * sum(a), sum(a))] : 
  only 0's may be mixed with negative subscripts

whereas Renjin returns

[1] NA NA