summaryrefslogtreecommitdiffstats
path: root/src/boost/libs/math/example/nonfinite_facet_simple.cpp
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-27 18:24:20 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-27 18:24:20 +0000
commit483eb2f56657e8e7f419ab1a4fab8dce9ade8609 (patch)
treee5d88d25d870d5dedacb6bbdbe2a966086a0a5cf /src/boost/libs/math/example/nonfinite_facet_simple.cpp
parentInitial commit. (diff)
downloadceph-upstream.tar.xz
ceph-upstream.zip
Adding upstream version 14.2.21.upstream/14.2.21upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/boost/libs/math/example/nonfinite_facet_simple.cpp')
-rw-r--r--src/boost/libs/math/example/nonfinite_facet_simple.cpp269
1 files changed, 269 insertions, 0 deletions
diff --git a/src/boost/libs/math/example/nonfinite_facet_simple.cpp b/src/boost/libs/math/example/nonfinite_facet_simple.cpp
new file mode 100644
index 00000000..3be01c2d
--- /dev/null
+++ b/src/boost/libs/math/example/nonfinite_facet_simple.cpp
@@ -0,0 +1,269 @@
+/** nonfinite_num_facet.cpp
+*
+* Copyright (c) 2011 Paul A. Bristow
+*
+* Distributed under the Boost Software License, Version 1.0.
+* (See accompanying file LICENSE_1_0.txt
+* or copy at http://www.boost.org/LICENSE_1_0.txt)
+*
+* This very simple program illustrates how to use the
+* `boost/math/nonfinite_num_facets.hpp' to obtain C99
+* representation of infinity and NaN.
+* (from the original
+* Floating Point Utilities contribution by Johan Rade.
+* Floating Point Utility library has been accepted into Boost,
+* but the utilities are incorporated into Boost.Math library.
+*
+\file
+
+\brief A very simple example of using non_finite_num facet for
+C99 standard output of infinity and NaN.
+
+\detail Provided infinity and nan are supported,
+this example shows how to create a C99 non-finite locale,
+and imbue input and output streams with the non_finite_num put and get facets.
+This allow output and input of infinity and NaN in a Standard portable way,
+This permits 'loop-back' of output back into input (and portably across different system too).
+This is particularly useful when used with Boost.Serialization so that non-finite NaNs and infinity
+values in text and xml archives can be handled correctly and portably.
+
+*/
+
+#ifdef _MSC_VER
+# pragma warning (disable : 4127) // conditional expression is constant.
+#endif
+
+#include <iostream>
+using std::cout;
+using std::endl;
+using std::cerr;
+
+#include <iomanip>
+using std::setw;
+using std::left;
+using std::right;
+using std::internal;
+
+#include <string>
+using std::string;
+
+#include <sstream>
+using std::istringstream;
+
+#include <limits>
+using std::numeric_limits;
+
+#include <locale>
+using std::locale;
+
+#include <boost/math/special_functions/nonfinite_num_facets.hpp>
+// from Johan Rade Floating Point Utilities.
+
+int main ()
+{
+ std::cout << "Nonfinite_num_facet very simple example." << std::endl;
+
+ if((std::numeric_limits<double>::has_infinity == 0) || (std::numeric_limits<double>::infinity() == 0))
+ {
+ std::cout << "Infinity not supported on this platform." << std::endl;
+ return 0;
+ }
+
+ if((std::numeric_limits<double>::has_quiet_NaN == 0) || (std::numeric_limits<double>::quiet_NaN() == 0))
+ {
+ std::cout << "NaN not supported on this platform." << std::endl;
+ return 0;
+ }
+
+ std::locale default_locale (std::locale::classic ()); // Note the currrent (default C) locale.
+
+ // Create plus and minus infinity.
+ double plus_infinity = +std::numeric_limits<double>::infinity();
+ double minus_infinity = -std::numeric_limits<double>::infinity();
+
+ // and create a NaN (NotANumber)
+ double NaN = +std::numeric_limits<double>::quiet_NaN ();
+
+ double negated_NaN = (boost::math::changesign)(std::numeric_limits<double>::quiet_NaN ());
+
+
+ // Output the nonfinite values using the current (default C) locale.
+ // The default representations differ from system to system,
+ // for example, using Microsoft compilers, 1.#INF, -1.#INF, and 1.#QNAN,
+ // Linux "inf", "-inf", "nan"
+ cout << "Using C locale" << endl;
+ cout << "+std::numeric_limits<double>::infinity() = " << plus_infinity << endl;
+ cout << "-std::numeric_limits<double>::infinity() = " << minus_infinity << endl;
+ cout << "+std::numeric_limits<double>::quiet_NaN () = " << NaN << endl;
+
+ // Display negated NaN.
+ cout << "negated NaN " << negated_NaN << endl; // "-1.IND" or "-nan".
+
+ // Create a new output locale, and add the nonfinite_num_put facet
+ std::locale C99_out_locale (default_locale, new boost::math::nonfinite_num_put<char>);
+ // and imbue the cout stream with the new locale.
+ cout.imbue (C99_out_locale);
+
+ // Or for the same effect more concisely:
+ cout.imbue (locale(locale(), new boost::math::nonfinite_num_put<char>));
+
+ // Output using the new locale:
+ cout << "Using C99_out_locale " << endl;
+ cout << "+std::numeric_limits<double>::infinity() = " << plus_infinity << endl;
+ cout << "-std::numeric_limits<double>::infinity() = " << minus_infinity << endl;
+ cout << "+std::numeric_limits<double>::quiet_NaN () = " << NaN << endl;
+ // Expect "inf", "-inf", "nan".
+
+ // Display negated NaN.
+ cout << "negated NaN " << negated_NaN << endl; // Expect "-nan".
+
+ // Create a string with the expected C99 representation of plus infinity.
+ std::string inf = "inf";
+ { // Try to read an infinity value using the default C locale.
+ // Create an input stream which will provide "inf"
+ std::istringstream iss (inf);
+
+ // Create a double ready to take the input,
+ double infinity;
+ // and read "inf" from the stringstream:
+ iss >> infinity;
+
+ // This will not work on all platforms! (Intel-Linux-13.0.1 fails EXIT STATUS: 139)
+ if (! iss)
+ { // Reading infinity went wrong!
+ std::cerr << "C locale input format error!" << std::endl;
+ }
+ } // Using default C locale.
+
+ { // Now retry using C99 facets.
+ // Create a new input locale and add the nonfinite_num_get facet.
+ std::locale C99_in_locale (default_locale, new boost::math::nonfinite_num_get<char>);
+
+ // Create an input stream which will provide "inf".
+ std::istringstream iss (inf);
+ // Imbue the stream with the C99 input locale.
+ iss.imbue (C99_in_locale);
+
+ // Create a double ready to take the input,
+ double infinity;
+ // and read from the stringstream:
+ iss >> infinity;
+
+ if (! iss)
+ { // Reading infinity went wrong!
+ std::cout << "C99 input format error!" << std::endl;
+ }
+ // Expect to get an infinity, which will display still using the C99 locale as "inf"
+ cout << "infinity in C99 representation is " << infinity << endl;
+
+ // To check, we can switch back to the default C locale.
+ cout.imbue (default_locale);
+ cout << "infinity in default C representation is " << infinity << endl;
+ } // using C99 locale.
+
+ {
+ // A 'loop-back example, output to a stringstream, and reading it back in.
+ // Create C99 input and output locales.
+ std::locale C99_out_locale (default_locale, new boost::math::nonfinite_num_put<char>);
+ std::locale C99_in_locale (default_locale, new boost::math::nonfinite_num_get<char>);
+
+ std::ostringstream oss;
+ oss.imbue(C99_out_locale);
+ oss << plus_infinity;
+
+ std::istringstream iss(oss.str()); // So stream contains "inf".
+ iss.imbue (C99_in_locale);
+
+ std::string s;
+
+ iss >> s;
+
+ cout.imbue(C99_out_locale);
+ if (oss.str() != s)
+ {
+ cout << plus_infinity << " != " << s << " loopback failed!" << endl;
+ }
+ else
+ {
+ cout << plus_infinity << " == " << s << " as expected." << endl;
+ }
+ }
+
+
+ // Example varying the width and position of the nonfinite representations.
+ // With the nonfinite_num_put and _get facets, the width of the output is constant.
+
+ #ifdef BOOST_NO_CXX11_NUMERIC_LIMITS
+ cout << "BOOST_NO_CXX11_NUMERIC_LIMITS is defined, so no max_digits10 available." << endl;
+ std::streamsize max_digits10 = 2 + std::numeric_limits<double>::digits * 30103UL / 100000UL;
+#else
+ // Can use new C++0X max_digits10 (the maximum potentially significant digits).
+ std::streamsize max_digits10 = std::numeric_limits<double>::max_digits10;
+#endif
+ cout << "std::numeric_limits<double>::max_digits10 is " << max_digits10 << endl;
+ cout.precision(max_digits10);
+
+ double pi = 3.141592653589793238462643383279502884197169399375105820974944;
+ // Expect 17 (probably) decimal digits (regardless of locale).
+ // cout has the default locale.
+ cout << "pi = " << pi << endl; // pi = 3.1415926535897931
+ cout.imbue (C99_out_locale); // Use cout with the C99 locale
+ // (expect the same output for a double).
+ cout << "pi = " << pi << endl; // pi = 3.1415926535897931
+
+ cout << "infinity in C99 representation is " << plus_infinity << endl;
+
+ //int width = 2; // Check effect if width too small is OK.
+ // (There was a disturbed layout on older MSVC?).
+ int width = 20;
+
+ // Similarly if we can switch back to the default C locale.
+ cout.imbue (default_locale);
+ cout << "infinity in default C representation is " << plus_infinity << endl;
+ cout << "infinity in default C representation (setw(" << width << ") is |" << setw(width) << plus_infinity <<'|' << endl;
+ cout << "infinity in default C representation (setw(" << width << ") is |" << left << setw(width) << plus_infinity <<'|' << endl;
+ cout << "infinity in default C representation (setw(" << width << ") is |" << internal << setw(width) << plus_infinity <<'|' << endl;
+
+ cout.imbue (C99_out_locale);
+ cout << "infinity in C99 representation (setw(" << width << ") is |" << right << setw(width) << plus_infinity <<'|'<< endl;
+ cout << "infinity in C99 representation (setw(" << width << ") is |" << left << setw(width) << plus_infinity <<'|'<< endl;
+ cout << "infinity in C99 representation (setw(" << width << ") is |" << internal << setw(width) << plus_infinity <<'|'<< endl;
+
+ return 0;
+} // int main()
+
+// end of test_nonfinite_num_facets.cpp
+
+/*
+
+Output:
+
+simple_nonfinite_facet.vcxproj -> J:\Cpp\MathToolkit\test\Math_test\Release\nonfinite_facet_simple.exe
+ Nonfinite_num_facet very simple example.
+ Using C locale
+ +std::numeric_limits<double>::infinity() = 1.#INF
+ -std::numeric_limits<double>::infinity() = -1.#INF
+ +std::numeric_limits<double>::quiet_NaN () = 1.#QNAN
+ Using C99_out_locale
+ +std::numeric_limits<double>::infinity() = inf
+ -std::numeric_limits<double>::infinity() = -inf
+ +std::numeric_limits<double>::quiet_NaN () = nan
+ infinity in C99 representation is inf
+ infinity in default C representation is 1.#INF
+ 3
+ 3
+ inf == inf as expected.
+ std::numeric_limits<double>::max_digits10 is 17
+ pi = 3.1415926535897931
+ C locale input format error!
+ pi = 3.1415926535897931
+ infinity in C99 representation is inf
+ infinity in default C representation is 1.#INF
+ infinity in default C representation (setw(20) is 1.#INF|
+ infinity in default C representation (setw(20) is 1.#INF |
+ infinity in default C representation (setw(20) is 1.#INF|
+ infinity in C99 representation (setw(20) is inf|
+ infinity in C99 representation (setw(20) is inf |
+ infinity in C99 representation (setw(20) is inf|
+
+*/