CODEWITHSUNDEEP
rrcpp
F. Privé
F. Privé

Reputation: 11738

R's NA_REAL conversion to integer in C++ now gives 0 on Mac (instead of NA_INTEGER)

Reprex:

// [[Rcpp::export]]
int dbl2int(double x) { return x; }

/*** R
dbl2int(NA_real_)
*/

Rcpp::sourceCpp("test-dbl2int.cpp")

On Mac, it outputs 0, while it outputs NA_INTEGER on Windows and Linux (cf. https://github.com/privefl/testbigstatsr/actions/runs/10737751205/job/29779881355#step:6:165). Note: this issue appeared only recently on Mac (it has given NA_INTEGER for years..).

Any idea what's going on, and how to do this conversion so that it works for all platforms and is still efficient?

Edit: It seems that Inf is now converted to 2147483647 instead of NA_INTEGER; but -Inf is still converted to NA..

Upvotes: 2

Views: 71

Answers (1)

Ada Lovelace
Ada Lovelace

Reputation: 1265

This may well be risky code. You are relying an a very non-standard NA conversion: only R knows about NA for integer, so you are betting on the appropriate R conversion being found. I am not sure we can rely on that.

Two things come to mind. First, maybe see if you can distill this into a pure C API for R function and then report to r-devel. (Also, if and when that works and this does not, file an Rcpp bug showing that.)

Second, use a safer converter. Something like

Rcpp::cppFunction("int dbl2int(double x) { 
     if (ISNAN(x)) return NA_INTEGER; else return static_cast<int>(x); }")

You can also test via R_IsNA(), there is a bit of overlap between test predicates.

Upvotes: 5

Related Questions