Introduction

If you are annoyed by std::setprecision(std::numeric_limits<long double>::digits10 + 1) this tool may be for you. It is in a draft state and should be used with care.

The purpose of numdata is to print floating point numbers to streams in a better way than it is possible using standard tools. It is based on std::locale::facet mechanism (https://en.cppreference.com/w/cpp/locale/locale/facet) and Google double-conversion library (https://github.com/google/double-conversion).

Features

• Prints floating point numbers in minimal representation that allows to parse them without ambiguity.
• Prints trailing . for integer values reducing value type ambiguity in the output.
• Does not produce excessive 0 values.
• Non-invasive.
• May actually work faster than standard output (https://github.com/miloyip/dtoa-benchmark), not verified.

Usage

#include <numdata/numdata.h>

...
double a = 0.1;
std::cout << numdata::On;
std::cout << a;
std::cout << numdata::Off;
...

Output example

See test/demo.cpp

numdata:
 0.
 inf
 nan
 2.2250738585072014e-308
 1.7976931348623157e+308
 1.
 1e+11
 1.111111e-1
 1e-15
 3.000020000001e-1

default:
 0
 inf
 nan
 2.22507e-308
 1.79769e+308
 1
 1e+11
 0.111111
 1e-15
 0.300002

std::scientific:
 0.000000e+00
 inf
 nan
 2.225074e-308
 1.797693e+308
 1.000000e+00
 1.000000e+11
 1.111111e-01
 1.000000e-15
 3.000020e-01

std::setprecision:
 0
 inf
 nan
 2.225073858507201383e-308
 1.797693134862315708e+308
 1
 100000000000
 0.1111111000000000043
 1.000000000000000078e-15
 0.300002000000100022

std::showpoint + std::setprecision:
 0.000000000000000000
 inf
 nan
 2.225073858507201383e-308
 1.797693134862315708e+308
 1.000000000000000000
 100000000000.0000000
 0.1111111000000000043
 1.000000000000000078e-15
 0.3000020000001000220

Notes

• Facet inherits from std::num_put which lacks specialization for float type, i.e., such values are converted to double.
• double-conversion is not the fastest library, see, e.g., https://github.com/jk-jeon/dragonbox, but others generally lack features required by numdata.