NumberParser::tryParseFloat is calling this overload:
bool strToDouble(const std::string& str, double& result, char decSep, char thSep, const char* inf, const char* nan)

Since it accepts a const std::string& str,, calling with s.c_str() will create a temporary std::string to call strToDouble

Maybe a possible patch for strToDouble would be change the strToDouble to receive a string by value without const ref and remove the std::string tmp from the code, thisway can avoid some possible multiple duplications when calling with a char* or similar.

bool strToDouble(std::string str, double& result, char decSep, char thSep, const char* inf, const char* nan)
{
	if (str.empty()) return false;

	using namespace double_conversion;

	trimInPlace(str);
	removeInPlace(str, thSep);
	replaceInPlace(str, decSep, '.');
	removeInPlace(str, 'f');
	result = strToDouble(str.c_str(), inf, nan);
	return !FPEnvironment::isInfinite(result) &&
		!FPEnvironment::isNaN(result);
}