CODEWITHSUNDEEP
c++c++20cinterop
Michael Stum
Michael Stum

Reputation: 180944

Can I safely use a string_view if I need to eventually need to get a char pointer?

I'm working on some C++ code that also eventually needs to call OS-level C code, for example scandir.

I'd like to use C++ for most of the codebase, which (to me) means that I'm mostly working with std::string instead of char pointers.

I have a method that accepts a string_view, so that I can pass in both std::string and char*, depending on whether the C++ or the "C interop" code needs to call it:

std::optional<std::vector<FileInfo>> scan_directory(const std::string_view directory)
{
    if (directory.empty()) {
        return {};
    }

    struct dirent **namelist;
    int num = scandir(directory.data(), &namelist, nullptr, nullptr);
    [...]
}

Note the call to data() here, since scandir takes a const char *. Now, I saw this note:

Unlike std::basic_string::data() and string literals, data() may return a pointer to a buffer that is not null-terminated. Therefore it is typically a mistake to pass data() to a routine that takes just a const CharT* and expects a null-terminated string.

That got me thinking: Is there a better/safer way? I know that in practice, the callers will be null-terminated strings, but I don't want to create a hard-to-diagnose bug later on when I'm already aware there's a potential issue here. Though I guess that there's already no guarantee that a char* is null-terminated, so I'm not making the situation any worse.

Still, curious if there is a better option.

I'm on g++ (GCC) 10.2.1 20201016 (Red Hat 10.2.1-6) in C++20 mode via CMake's set(CMAKE_CXX_STANDARD 20).

Upvotes: 2

Views: 1014

Answers (1)

Sam Varshavchik
Sam Varshavchik

Reputation: 118340

When all you have is a std::string_view, you have no guarantees whatsoever that accessing beyond its size() is not undefined behavior. It's not an absolute guarantee that it's undefined behavior, but neither you have any guarantees that it's not. If the string_view was constructed using a pointer to a \0-terminated character string, but it's not included in the constructed string_view's size, then you're arguably safe (I note that recent versions of libstdc++ have an option to do some boundary checking on vectors and string accesses, and if at some point in the future boundary checking get introduced for string_views they'll trip you up, and you'll be out of luck). But it is certainly possible to construct a string_view with a pointer to a string that's not \0 terminated, with an exact character count. In that case looking beyond its size() will result in nasal demons. And you have no way to determine whether this is the case.

The simplest solution for you here is to declare the parameter as const char *, and create an overload that takes a const std::string & alternative parameter, and then calls this function using c_str().

Upvotes: 1

Related Questions