What I learned implementing argparse in C++, pt. 1

Krzysiek Karbowiak
by Krzysiek Karbowiak

Categories

  • docs
  • Programming

Tags

  • cpp-argparse
  • musings

Look around

When I decided to implement Python’s argparse in C++, I thought this was a really great idea. I was so excited to be working on this, I almost pat myself on the back for coming up with this :-)

What I did not think about was that other people surely thought of the same, and there must be other projects with the same goal. When it finally occured to me to do a search on Github for “cpp argparse” string, I saw that there were plenty of results. Some projects dated back to 2014!

Luckily for me, my findings did not invalidate my efforts, at least not entirely. Most of the projects were not header-only, were incomplete, or did not offer the facilities of modern C++.

But honestly, even if I found something suitable to use as is, or something I could contribute to, I would probably start my project nonetheless just to see how much I could do.

C++17, where art thou?

Where is C++17? Everywhere. Meaning that every major compiler1 supports it fully. Does that mean that you can write modern C++17 code, depending on the utilities, and compile it without problems on Linux and Windows? Unfortunately, not.

It turns out that compiler’s support for language standard is one thing, and the completeness of the implementation of the standard library the compiler ships with is another.

I was already bitten by this when I wanted to do 2016’s edition of Advent of Code in C++17. I wanted to utilise the newly added string to number conversions available in <charconv> header in the form of std::from_chars. To my dissapointment, GCC rejected my code stating that there was no such header file.

Back then I thought that it’s OK. C++17 was not yet officially approved and the compilers shipped a preview.

As it turns out, currently only Visual Studio 2019 ships with complete implementation of std::from_chars and std::to_chars (go see Stephan T. Lavavej’s great talk on the subject). Both GCC and Clang lag behind offering no support for floating point types.

It seems writing portable code no longer means relying on standard library and avoiding compiler extensions. You need to use the subset of standard library that is fully shipped by the compilers you will use.

  1. My list of major C++ compilers includes Clang, GCC, and MSVC. ↩