Bug in arithmetic operation with NA as argument

[TDiazT]

• R program that triggers the difference

1 + NA

(Any sum using NA)

• runR.native Output

[1] NaN

• R Output

[1] NA 

• Additional information

I am wondering if it has something to do with using Double.add.
I noticed in bisect that this execution goes all the way from do_arith to R_binary. The checks at the source code (first, snd) return false (which seems ok). Then it jumps straight to calling R_binary.

In R_binary it skips the checks for array (makes sense) and in this check it enters. It ends up calling real_binary (source) and ends ups performing this operation (coqR code)

[TDiazT]

Actually it happens with every arithmetic operation.

1 + NA
1 * NA
NA/NaN
1 - NA

NaN/NA produces the correct result though.

[Mbodin]

If I remember correctly, there was an issue with NA_real, which displays NaN or something like that.

We are messing with OCaml's nan to differenciate NA from NaN and I guess that OCaml doesn't like that (because of some optimisations?). I don’t think that this is too big an issue for now.