ENH: Support calling a const NumberToString, add unit tests + style improvements

[N-Dekker]

Supported calling a const NumberToString function object, by declaring all NumberToString::operator() overloads const.

Added various GoogleTest unit tests, which do call const NumberToString objects.

Did various style improvements, including a reduction of the buf size, and removal of builder.Reset() calls from the implementation of NumberToString.

[thewtex]

Great to see the improved coverage 💚

[N-Dekker]

FYI, some interesting differences between the output of std::ostringstream and itk::NumberToString<double> that I encountered while writing the unit tests (using NL = std::numeric_limits<double>):

input	stream.str()	itk::NumberToString
-0.0	-0	0
1e-6	1e-06	0.000001
1e20	1e+20	100000000000000000000
NL::lowest()	-1.79769e+308	-1.7976931348623157e+308
NL::epsilon()	2.22045e-16	2.220446049250313e-16
NL::denorm_min()	4.94066e-324	5e-324
NL::min()	2.22507e-308	2.2250738585072014e-308
NL::quiet_NaN()	nan	NaN
NL::infinity()	inf	Infinity

As produced by the following code:

for (const double d :
     { -0.0, 1e-6, 1e20, NL::lowest(), NL::epsilon(), NL::denorm_min(), NL::min(), NL::quiet_NaN(), NL::infinity() })
{
  std::ostringstream stream;
  stream << d;
  std::cout << " | " << stream.str() << " | " << itk::NumberToString<double>{}(d) << '\n';
}

[dzenanz]

Now I see your comment. Perhaps in the case of

input	stream.str()	itk::NumberToString
NL::denorm_min()	4.94066e-324	5e-324

itk::NumberToString is less precise because of a "short" buffer?

[dzenanz]

If you wish to add some more unit tests, you can do so in a new PR specifically aimed at increasing test coverage. But that does not seem necessary.

[N-Dekker]

If you wish to add some more unit tests, you can do so in a new PR specifically aimed at increasing test coverage. But that does not seem necessary.

@dzenanz Thanks, I think it isn't necessary either to add more unit tests. 😃 I did add them especially to ensure that the STYLE commit c00531f wouldn't change the existing behavior. And that should be fine, as the tests are passed now both before and after the STYLE commit.

BTW, the tests might eventually break if the floating point format of itk::NumberToString is ever going to be changed. Instead of calling EcmaScriptConverter(), itk::NumberToString might be changed to use a different DoubleToStringConverter object, having different formatting flags. https://github.com/google/double-conversion/blob/master/double-conversion/double-to-string.h#L113-L121 But that's not my proposal, for now...!

[N-Dekker]

FYI, I just submitted a related issue: "Maximum string length DoubleToStringConverter::EcmaScriptConverter() ToShortest? 26 chars?", google/double-conversion#146

[N-Dekker]

Last weekend I submitted a related pull request to Google's double-conversion library, which aims to allow further simplifying the implementation of itk::NumberToString (as well as similar user code). Please have a look:

google/double-conversion#159 "Add DoubleToStringConverter::ToShortestString overload set"