Two important features in C++11 received small upgrades in C++14 that had far-reaching effects.<\/p>\n
constexpr<\/code> function bodies were allowed only to be a single return<\/code> statement. In C++14, nearly all statements are now allowed in constexpr<\/code> functions, allowing for much more idiomatic C++. You can learn more about what\u2019s allowed in core constant expressions on CppReference<\/a> as well as learn about constexpr<\/code> on the same site<\/a>.<\/li>\n- An aggregate type in C++11 could not be initialized as an aggregate if it contained any members that had default member initializers. In C++14 the committee removed this restriction on non-static data members having default member initializers. You can find out more about aggregate initialization on CppReference<\/a> as well as about non-static data member initialization<\/a> (known as NSDMI.)<\/li>\n<\/ol>\n
You can see from the CppReference site<\/a> that while both of these changes were small changes to the standard (search for the “C++14” annotation on those pages) they had broad impact on the compiler\u2019s implementation and on what code developers can write.<\/p>\n\n- Previously,
constexpr<\/code> was basically all about expressions, but C++14 extended it to allow control flow statements. The difference in what you can write in your code is huge: now instead of just being able to write expressions you can create idiomatic compile-time value compilations and mutation.<\/li>\n- The changes to allow aggregate initialization for aggregates that contain non-static data members makes a useful case of aggregate initialization possible.<\/li>\n<\/ul>\n
We\u2019re happy to announce that the compiler shipping with Visual Studio 2017 supports both extended constexpr and NSDMI for aggregates fully.<\/p>\n
Extended constexpr<\/h4>\n
Our extended constexpr work was test-driven. We tested against a number of libraries that are heavy users of extended constexpr<\/code>, including:<\/p>\n\n- Sprout<\/li>\n
- Boost (with MSVC workarounds removed)<\/li>\n
- Boost Hana<\/li>\n
- Guidelines Support Library (GSL)<\/li>\n
- libc++ tests<\/li>\n
- Microsoft’s Range-v3 fork<\/li>\n
- C++ Standard Libraries<\/li>\n<\/ul>\n
We’ve also evaluated against test suites from McCluskey, Plumhall, and Perennial, as well as a test suite built from code snippets you sent us from an online survey, online forums, and early bug reports from the VC Blog.<\/p>\n
Taking advantage of C++14 constexpr<\/code> is almost as easy as putting the constexpr<\/code> keyword on existing functions, because extended constexpr<\/code> allows a developer to use idiomatic C++ for their constexpr<\/code> functions.<\/p>\n\n- Local variable declarations and conditionals can now be used in
constexpr<\/code> functions. [see example 1 below]<\/li>\n- Instead of recursion, all looping constructs (except goto) are allowed in constexpr functions [see example 2 below]<\/li>\n<\/ul>\n
Using iteration instead of recursion should perform better at compile time, and, as constexpr<\/code> functions can be used at runtime, will perform better at runtime as well. We’ve beefed up our constexpr<\/code> control flow engine to eagerly identify cases where a constexpr<\/code> function is guaranteed to evaluate illegal expressions, so a class of errors can be caught at constexpr<\/code>-authoring time rather than after a library has been shipped to developers.<\/p>\nconstexpr examples<\/h4>\n\n\/\/ Example 1a\nenum class Messages { \n\tHi, \n\tGoodbye \n};\n\nconstexpr Messages switch_with_declaration_in_control(int x)\n{\n switch (int value = x)\n {\n case 0: return Messages::Hi;\n default: return Messages::Goodbye;\n }\n}\n\n\n\/\/ Example 1b\nconstexpr bool even(int x) {\n\tif (x % 2 == 0)\n\t\treturn true;\n\treturn false;\n}\n\n\/\/ Example 2a\nconstexpr int simple_count_up_do_loop(int x) {\n\tint i = 0;\n\tdo\t{\n\t\t++i;\n\t} while (i < x);\n\treturn i;\n}\n\n\/\/ Example 2b\nconstexpr int multiple_initializers_multiple_incr_count_up(int x) {\n\tint higher = 0;\n\tfor (auto i = 0, j = 1; i <= x; ++i, j++)\n\t\thigher = j;\n\treturn higher;\n}\n\n\/\/ Negative test cases\nconstexpr bool always(int x) {\n\treturn x;\n}\n\nconstexpr int guaranteed_to_hit_true_path() {\n\tif (always(1))\n\t\tthrow "OH NO"; \/\/ illegal expression, guaranteed to be hit\n\treturn 0;\n}\n\nconstexpr int guaranteed_to_hit_false_path() {\n\tif (42 * 2 - 84)\n\t\treturn 1;\n\telse\n\t\tthrow "OH NO"; \/\/ illegal expression, guaranteed to be hit\n\treturn 0;\n}\n\nconstexpr int guaranteed_to_evaluate_while_loop() {\n\twhile (always(33)) {\n\t\tnew int(0); \/\/ illegal expression, guaranteed to be hit\n\t}\n\treturn 0;\n}\n\nconstexpr int guaranteed_to_evaluate_for_loop() {\n\tfor (; always(22); )\n\t\tnew int(); \/\/ illegal expression, guaranteed to be hit\n\treturn 0;\n}\n<\/pre>\nNSDMI for aggregates<\/h4>\n
The NSDMI for aggregates changes were more limited in scope but automatically improve a lot of existing code. In C++ 11, the presence of a default member initializer (NSDMI) would preclude the class from being an aggregate; therefore, it would not be eligible for aggregate initialization.<\/p>\n
An example:<\/p>\n
\nstruct S {\n int i = 1;\n int j = 2;\n};\n \nS s1; \/\/ OK; calls the default constructor, which initializes 'i' and 'j' to 1 and 2.\nS s2{}; \/\/ OK; not aggregate initialization because S is not an aggregate; calls the default constructor.\ns3{42}; \/\/ Error; S is not an aggregate and there is no appropriate constructor.\n<\/pre>\nIn C++14, S is now considered to be an aggregate class type, so aggregate initialization can be used:<\/p>\n
\nS s4{}; \/\/ OK with C++14; aggregate initialization; no initializer is provided for 'i' or 'j', so their respective default member initializers will be used to initialize them to 1 and 2.\nS s5{42}; \/\/ OK with C++14; aggregate initialization; 'i' is explicitly initialized to 42 and no initializer is provided for 'j', so its default member initializer will be used to initialize it to 2.\n<\/pre>\nIn closing<\/h4>\n
As always, we welcome your feedback. Feel free to send any comments through e-mail at visualcpp@microsoft.com<\/a>, through Twitter @visualc<\/a>, or Facebook at Microsoft Visual Cpp<\/a>.<\/p>\n