src: use ostringstream instead of string in FromFilePath()

[pluris]

When performing string concatenation operations, it is more advantageous in terms of performance to use std::ostringstream rather than std::string. This approach is also used elsewhere, such as debug_utils and node_errors.

Refs: #46410

[nodejs-github-bot]

Review requested:

• @nodejs/url

[anonrig]

What is the performance impact of the new proposed change? Did you had any chance to benchmark it?

[pluris]

I ran benchmark/url/* and there doesn't seem to be much difference.
Also, this is not a binding API, so I don't know if I should create a benchmark. Could you please give me some advice?
To compare std::string and std::ostringstream, I created the some code as follows.

#include <iostream>
#include <chrono>
#include <sstream>

std::string FromFilePath1(const std::string_view file_path) {
  std::string escaped_file_path;
  for (size_t i = 0; i < file_path.length(); ++i) {
    escaped_file_path += file_path[i];
    if (file_path[i] == '%') escaped_file_path += "25";
  }
  return escaped_file_path;
}

std::string FromFilePath2(const std::string_view file_path) {
  std::ostringstream escaped_file_path;
  for (size_t i = 0; i < file_path.length(); ++i) {
    escaped_file_path << file_path[i];
    if (file_path[i] == '%') escaped_file_path << "25";
  }
  return escaped_file_path.str();
}

int main() {
    const std::string_view sample_url= "https://nodejs.org/en/blog/";

    std::chrono::system_clock::time_point start = std::chrono::system_clock::now();
    for (int i = 0;i < 1000000;i++)
        FromFilePath1(sample_url);
    std::chrono::duration<double> sec = std::chrono::system_clock::now() - start;
    std::cout << "result 1 : " << sec.count() <<" seconds"<< std::endl;
    // Prints : result 1 : 1.63845 seconds
    start = std::chrono::system_clock::now();
    
    for (int i = 0;i < 1000000;i++)
        FromFilePath2(sample_url);
    
    sec = std::chrono::system_clock::now() - start;
    std::cout << "result 2 : " << sec.count() <<" seconds"<< std::endl;
    // Prints : result 2 : 1.17444 seconds

    return 0;
}

[lemire]

I ran my own benchmarks. Please see my blog post:

For processing strings, streams in C++ can be slow

Here is the gist:

At least in this case, I find that the stream version is four times slower than naive string processing, and it is 30 times slower than the optimized ‘find’ approach. Your results will vary depending on your system, but I generally consider the use of streams in C++ as a hint that there might be poor performance.

I understand that there has been performance engineering guided by the belief that streams in C++ are efficient. My recommendation is to re-assess this belief and conduct benchmarks.

In fact, it is entirely possible that removing streams where they have been added could be a strategy to improve performance.

For benchmarking, I recommend using relatively large inputs. Here is why: Benchmarking is hard: processors learn to predict branches.

[pluris]

@lemire Thanks for the good information. This PR will be closed. 😄