as.integer(x) rounds floating point numbers towards zero,
so it behaves approximately as as.integer(trunc(x)).
If x > INT_MAX, then as.integer(x) is NA. This is nothing
bad, but it is slightly more restrictive than necessary. An
alternative approach could be that
as.integer(x) is NA, if trunc(x) > INT_MAX and
as.integer(x) == INT_MAX for all x in [INT_MAX,INT_MAX+1).
Let me suggest the following patch to IntegerFromReal for consideration:
--- R-devel_2007-10-11-orig/src/main/coerce.c 2007-07-25 17:54:17.000000000
+0200
+++ R-devel_2007-10-11-asinteger/src/main/coerce.c 2007-10-12 07:10:06.000000000
+0200
@@ -173,7 +173,8 @@
{
if (ISNAN(x))
return NA_INTEGER;
- else if (x > INT_MAX || x <= INT_MIN ) {
+ x = trunc(x);
+ if (x > INT_MAX || x <= INT_MIN ) {
*warn |= WARN_NA;
return NA_INTEGER;
}
This patch changes the behavior as suggested above. Its effect may be seen
using the script
options(digits=12)
x <- seq(2^31 - 3, 2^31, length=7)
cbind(x,as.integer(trunc(x)),as.integer(x))
Original behavior:
[1,] 2147483645.0 2147483645 2147483645
[2,] 2147483645.5 2147483645 2147483645
[3,] 2147483646.0 2147483646 2147483646
[4,] 2147483646.5 2147483646 2147483646
[5,] 2147483647.0 2147483647 2147483647
[6,] 2147483647.5 2147483647 NA
[7,] 2147483648.0 NA NA
Using the patch:
[1,] 2147483645.0 2147483645 2147483645
[2,] 2147483645.5 2147483645 2147483645
[3,] 2147483646.0 2147483646 2147483646
[4,] 2147483646.5 2147483646 2147483646
[5,] 2147483647.0 2147483647 2147483647
[6,] 2147483647.5 2147483647 2147483647
[7,] 2147483648.0 NA NA
Petr Savicky.