should cpp11 support string_view?

[CreRecombinase]

R strings seem like a pretty perfect use case for std::string_view. For compilers that don't support c++17, there are backports available. Any interest in a pull request?

[jimhester]

I don't think it is entirely trivial to do this. First as you mentioned we would need to use compatibility library so we could use it with C++11.

Second and perhaps more importantly cpp11 uses translateCharUT8() to access character data from R vectors. This translates the characters to UTF-8 and (if needed) allocates memory that reclaimed at the end of the enclosing .Call(). If we use string_view() on this data and the string_view() outlasts the .Call() boundary the data will no longer be valid.

It might be that this particular case is rare enough and the benefits of string_view are substantial enough that it makes sense, I am not sure.

[jimhester]

Another pretty serious issue with using string_view in cpp11 is that nearly all R APIs that have const char* expect null terminated strings, which may not be the case with a string_view.

[CreRecombinase]

Yeah the point about possible allocation by translateCharUTF8 totally makes sense. I guess I was thinking of string_view as modeling the contents of a CHARSXP "as it is". Operationally, I think of a string_view as

1. non-owning
2. cheap to copy
3. immutable (unlike other "data" SEXP, you can't muck easily around with the contents of a CHARSXP in-place because of the CHARSXP cache. string_view, unlike its cousin span, will only give you a const pointer to its contents)
4. "knowing" its own length

[jimhester]

Closing this issue for now, while I do think using string_view is a nice idea in theory, I don't think it really will work in practice for cpp11 due to the concerns expressed above.