Reading n chars from stream to string

Question

I need to read n chars from a binary file into a string. Currently, what I do is:

static string Read(istream &stream, uint32_t count)
{
 auto bytes = unique_ptr<char[]>(new char[count]);
 stream.read(bytes.get(), count);
 return string(bytes.get(), count);
}

I found the way I deal with the array of chars quite messy. If I used new and delete[] directly, it would make the code messy in another way (I would need to add a local variable for the result). And I'm trying to avoid delete as a general rule.

Is there a clear way to write this code? The fact that it uses twice as much memory as it needs is probably not a big deal, but fixing that would be nice too.

• Konrad Rudolph: I think this is actually quite clean – albeit low-level – code. Nothing really wrong with it, and it’s probably more efficient than ruds’ implementation, although that’s arguably even cleaner.
• Loki Astari: @KonradRudolph: There is no need to be calling new here. Especially since the string object will deal with all that for you.
• Konrad Rudolph: @Loki Ah true, I’d completely forgotten about C++11’s contiguity requirement.

Answer 1

Why not:

static string Read(istream &stream, uint32_t count)
{
 std::string result(count, ' ');
 stream.read(&result[0], count);
 return result;
}

Though not strictly C++03 compatible that is easily validated. One of the reasons the committee found it easy to add the new constraint in C++11 was that no implementation did not use contiguous memory (Random Access Iterators are a hint).

But a C++03 strict implementation would be:

static string Read(istream &stream, uint32_t count)
{
 std::vector<char> result(count); // Because vector is guranteed to be contiguous in C++03
 stream.read(&result[0], count);
 return std::string(&result[0], &result[count]);
}

• svick: Yeah, that's certainly better.
• ruds: Note that this requires C++11 in a non-obvious way (C++03 didn't require that &result[0] worked this way). Otherwise, LGTM.
• Josh Kelley: Depending on your use case, you may want to do result.resize(stream.gcount());, in case not all of count bytes could be read.

Answer 2

First of all, it seems that you've got a using namespace std; somewhere in your code. Don't do that. (No, really).

Here's a function that should meet your needs.

static std::string Read(std::istream &stream, std::string::size_type count)
{
 std::string out;
 out.reserve(count);
 std::copy_n(std::istreambuf_iterator(stream), count, std::back_inserter(out));
 return out;
}

• Jamal: uint32_t should have an std:: as well.
• svick: I don't have using namespace std; there. Instead, I have using std::uint32_t; using std::unique_ptr; using std::string; using std::istream; there. It's kind of annoying, but I thought it would be better than writing std:: everywhere.
• ruds: @Jamal You're right. Actually, I've changed it to std::string::size_type, as that's the size class we're interested in.
• ruds: @svick I think it's odd to read code with names from std but with no std:: in front. Personal tastes vary, of course.
• svick: Is this going to insert the characters one by one? I worry that doing that would be too slow (the result might be one megabyte or so). (Of course, the proper way to find that out would be to measure it.)