Consider fast_float and std::from_chars

[alugowski]

Great library!

I did notice that the integer and floating-point routines parsing are home-grown. They do seem to perform well and handle common values well, but it's easy to see places where they'd break down.

I faced the same problem and settled on fast_float for floats and std::from_chars for integers. The former has basically the same performance as read_fp, but can be considered correct (including error checking!). It's already been, or soon will be, included in all the major compilers. The latter has basically universal compiler support. I don't see a minimum C++ standard declaration anywhere, so I assume C++17 is ok.

[kasimirgabert]

This is a great suggestion. For the systems that this library was built for (and currently used on), C++11 through older compilers is all that is available. However, from_chars and fast_float would be perfect for those that can use it. The current plan is to develop a parameter that can allow different character and float parsing to be used, with the homegrown one an option available for those that do not have access to other implementations.

[alugowski]

fast_float itself is C++11, it just uses an interface similar to the C++17 routines. Only std::from_chars requires C++17.

I'm noticing not too much downside with strtol and friends for integers, though I bet your homegrown is faster. I'm too paranoid of missing some edgecases to use a homegrown solution. For example the homegrown does not appear to detect overflow. That's probably a worthwhile tradeoff for many users though.