Rcpp::String silently drops embedded NUL bytes

[kevinushey]

For example:

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
SEXP embeddedNul() {
  std::string hasNullByte("abc\0abc", 7);
  String converted(hasNullByte);
  return wrap(converted);
}

/*** R
embeddedNul()
*/

Sourcing this gives me:

> embeddedNul()
[1] "abc"

I would hope for either:

1. An error, since we shouldn't silently lose data following the NUL byte; or
2. A conversion to e.g. a raw character vector, preserving the embedded NUL.

It's worth stating that attempting to do this on the R side would give an error, e.g.

data <- as.raw(c(97, 98, 99, 0, 97, 98, 99))
rawToChar(data)

giving

> data <- as.raw(c(97, 98, 99, 0, 97, 98, 99))
> data
[1] 61 62 63 00 61 62 63
> rawToChar(data)
Error in rawToChar(data) : embedded nul in string: 'abc\0abc'

[kevinushey]

I believe this occurs because we use Rf_mkCharCE() throughout the String classes implementation:

$ ag -Q "mkCharCE"
inst/include/Rcpp/String.h
55:        String(): data(Rf_mkCharCE("", CE_UTF8)), buffer(), valid(true), buffer_ready(true), enc(CE_UTF8) {
369:            return valid ? data : Rf_mkCharCE(buffer.c_str(), enc);
398:                data = Rcpp_ReplaceObject(data, Rf_mkCharCE(Rf_translateCharUTF8(data), encoding));
400:                data = Rf_mkCharCE(buffer.c_str(), encoding);
472:                data = Rf_mkCharCE(buffer.c_str(), enc);

But since buffer itself is just a std::string, we should pass both the c_str() and the length() on and use Rf_mkCharLenCE(). I'll try making a PR.