Unclear error when r_string is used as an argument

[stephematician]

I know why this doesn't work; however it took a while to track down (and I didn't find help elsewhere) - so at least if I put this here it might help others.

Basic idea is what if you want to pass a 'scalar' character vector (i.e. invoke a function in R like foo("test")), it looks like the pattern I have to use is:

cpp11::cpp_source(code='
#include <string>
#include "cpp11.hpp"
[[cpp11::register]]
void foo(cpp11::strings bar) {
    if (bar.size() != 1) cpp11::stop("Expected one string");
    std::string str = bar[0];
}')
foo("hello")

No problems.

On the other hand, trying this (and hoping that cpp11 does the hard work) fails:

cpp11::cpp_source(code='
#include <string>
#include "cpp11.hpp"
[[cpp11::register]]
void foo(cpp11::r_string bar) { std::string str = bar;}')
foo("hello")

Error: 'translateCharUTF8' must be called on a CHARSXP, but got 'character'

As a suggestion, perhaps have a valid_type check on the data similar to r_vector; e.g.

// update the assignment in the constructor to check type
  r_string(SEXP data) : data_(valid_type(data)) {}
// definition of valid_type somewhere
inline SEXP r_string::valid_type(SEXP data) {
  if (data == nullptr) {
    throw std::exception( /* construct error msg */);
  }
  if (TYPEOF(data) != CHARSXP) {
    throw std::exception(/* construct error msg */);
  }
  return data;
}

[Eunsolfs]

hi,I met a problem in laerning R ,can you give me some help?
My question link is : How to pass parameters in .R file to cpp11 function in .cpp file

thank you very much if you can lend me some time!

[stephematician]

hi,I met a problem in laerning R ,can you give me some help? My question link is : How to pass parameters in .R file to cpp11 function in .cpp file

thank you very much if you can lend me some time!

I don't know if you did, but I would try https://www.stackoverflow.com to post problems like this.

[Eunsolfs]

thank you for your help. Finally i got the solution –—— n <- sum(c(1.1,2.2,3.5)) use c(...) to pass vector of numbers

…

---- Replied Message ---- | From | ***@***.***> | | Date | 04/15/2023 07:27 | | To | ***@***.***> | | Cc | ***@***.***>***@***.***> | | Subject | Re: [r-lib/cpp11] Unclear error when `r_string` is used as an argument (Issue #311) | hi,I met a problem in laerning R ,can you give me some help? My question link is : How to pass parameters in .R file to cpp11 function in .cpp file thank you very much if you can lend me some time! I don't know if you did, but I would try https://www.stackoverflow.com to post problems like this. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[DavisVaughan]

I think we want the r_string(SEXP data) constructor to be very very fast, as it is used on each element of a character vector in some cases, so I don't think we should add that type check.

FWIW we do allow conversion directly into a std::string, which is what you wanted I think

cpp11::cpp_source(code='
#include <string>
#include "cpp11.hpp"

[[cpp11::register]]
cpp11::writable::strings foo(std::string bar) {
  cpp11::writable::strings out = { bar }; 
  return out;
}'
)

foo("hello")
#> [1] "hello"

---

Note to self that this occurs because we have enable_if_constructible_from_sexp here:

cpp11/inst/include/cpp11/as.hpp

Lines 33 to 37 in b0576f2

	template <typename T, typename R = void>
	using enable_if_constructible_from_sexp =
	enable_if_t<!is_smart_ptr<T>::value && // workaround for gcc 4.8
	std::is_class<T>::value && std::is_constructible<T, SEXP>::value,
	R>;

Used in

cpp11/inst/include/cpp11/as.hpp

Lines 81 to 84 in b0576f2

	template <typename T>
	enable_if_constructible_from_sexp<T, T> as_cpp(SEXP from) {
	return T(from);
	}

And r_string is of course constructible from a SEXP, just a CHARSXP, not a STRSXP

cpp11/inst/include/cpp11/r_string.hpp

Line 17 in b0576f2

r_string(SEXP data) : data_(data) {}

It's possible we could add a more specific as_cpp() condition for something like enable_is_r_string, but I'm not sure that would conflict with enable_if_constructible_from_sexp since both would technically be valid.

[stephematician]

Thanks. I think I had an unusual use-case; most of the time casting to std::string works a-ok, but it does involve a copy.

FWIW - I'd be surprised if there was a substantial performance hit; between inlining, branch prediction, and the simple pointer access via TYPEOF, most CPUs will cruise through it ... but I am willing to be corrected if anyone did some profiling.

Thanks again for looking into this.