Breaking Changes in Visual C++
When you upgrade to a new version of the Visual C++ compiler, you might encounter compilation and/or runtime errors in code that previously compiled and ran correctly. Changes in the new version that cause such problems are known as breaking changes, and typically they're required by modifications in the C++ language standard, function signatures, or the layout of objects in memory.
In Visual C++, although POD (plain old data) object layout and COM interfaces are guaranteed not to break from version to version, other kinds of object layout—for example, in scenarios that involve inheritance or template instantiation—are subject to change.
To avoid run-time errors that are difficult to detect and diagnose, we recommend that you never statically link to binaries that were compiled by using different versions of the compiler. Also, when you upgrade an EXE or DLL project, make sure to upgrade the libraries that it links to. If you're using CRT (C Runtime) or STL (Standard Template Library) types, don't pass them between binaries (including DLLs) that were compiled by using different versions of the compiler. For more information, see Potential Errors Passing CRT Objects Across DLL Boundaries.
We further recommend that you never write code that depends on a particular layout for an object that is not a COM interface or a POD object. If you do write such code, then you must ensure that it works after you upgrade. For more information, see Portability At ABI Boundaries (Modern C++).
The rest of this article describes specific breaking changes in Visual C++ in Visual Studio 2013.
Visual C++ Compiler
The final keyword now generates an unresolved symbol error where it would have compiled previously:
struct S1 { virtual void f() = 0; }; struct S2 final : public S1 { virtual void f(); }; int main(S2 *p) { p->f(); }
In earlier versions, an error wasn't issued because the call was a virtual call; nevertheless, the program would crash at runtime. Now, a linker error is issued because the class is known to be final. In this example, to fix the error, you would link against the obj that contains the definition of S2::f.
When you use friend functions in namespaces, you must re-declare the friend function before you refer to it or you will get an error because the compiler now conforms to the ISO C++ Standard. For example, this no longer compiles:
namespace NS { class C { void func(int); friend void func(C* const) {} }; void C::func(int) { NS::func(this); // error } }
To correct this code, declare the friend function:
namespace NS { class C { void func(int); friend void func(C* const) {} }; void func(C* const); // conforming fix void C::func(int) { NS::func(this); } }
The C++ Standard does not allow explicit specialization in a class. Although Visual C++ allows it in some cases, in cases such as the following example, an error is now generated because the compiler doesn't consider the second function to be a specialization of the first one.
template <int N> class S { public: template void f(T& val); template <> void f(char val); }; template class S<1>;
To correct this code, modify the second function:
template <> void f(char& val);
Visual C++ no longer tries to disambiguate the two functions in the following example, and now emits an error:
template<typename T> void Func(T* t = nullptr); template<typename T> void Func(...); int main() { Func<int>(); // error }
To correct this code, clarify the call:
template<typename T> void Func(T* t = nullptr); template<typename T> void Func(...); int main() { Func<int>(nullptr); // ok }
Before the compiler was made compliant with ISO C++11, the following code would have compiled and caused x to resolve to type int:
auto x = {0}; int y = x;
This code now resolves x to a type of std::initializer_list<int> and causes an error on the next line that tries to assign x to type int. (There is no conversion by default.) To correct this code, use int to replace auto:
int x = {0}; int y = x;
Aggregate initialization is no longer allowed when the type of the right-hand value does not match the type of the left-hand value that's being initialized, and an error is issued because the ISO C++11 Standard requires uniform initialization to work without narrowing conversions. Previously, if a narrowing conversion was available, a C4242 warning would have been issued instead of an error.
int i = 0; char c = {i}; // error
To correct this code, add an explicit narrowing conversion:
int i = 0; char c = {static_cast<char>(i)};
The following initialization is no longer allowed:
void *p = {{0}};
To correct this code, use either of these forms:
void *p = 0; // or void *p = {0};
Name lookup has changed. The following code is resolved differently in Visual C++ in Visual Studio 2012 and Visual C++ in Visual Studio 2013:
enum class E1 {a}; enum class E2 {b}; int main() { typedef E2 E1; E1::b; }
In Visual C++ in Visual Studio 2012, the E1 in expression E1::b resolved to ::E1 in the global scope. In Visual C++ in Visual Studio 2013, E1 in expression E1::b resolves to the typedef E2 definition in main() and has type ::E2.
Object layout has changed. On x64, the object layout of a class may change from previous releases. If it has a virtual function but it doesn’t have a base class that has a virtual function, the object model of the compiler inserts a pointer to a virtual function table after the data member layout. This means the layout may not be optimal in all cases. In previous releases, an optimization for x64 would try to improve the layout for you, but because it failed to work correctly in complex code situations, it was removed in Visual C++ in Visual Studio 2013. For example, consider this code:
__declspec(align(16)) struct S1 { }; struct S2 { virtual ~S2(); void *p; S1 s; };
In Visual C++ in Visual Studio 2013, the result of sizeof(S2) on x64 is 48, but in previous releases, it evaluates to 32. To make this evaluate to 32 in Visual C++ in Visual Studio 2013 for x64, add a dummy base class that has a virtual function:
__declspec(align(16)) struct S1 { }; struct dummy { virtual ~dummy() {} }; struct S2 : public dummy { virtual ~S2(); void *p; S1 s; };
To find places in your code that an earlier release would have tried to optimize, use a compiler from that release together with the /W3 compiler option and turn on Warning 4370. For example:
#pragma warning(default:4370) __declspec(align(16)) struct S1 { }; struct S2 { virtual ~S2(); void *p; S1 s; };
On Visual C++ compilers before Visual C++ in Visual Studio 2013, this code outputs this message:
warning C4370: 'S2' : layout of class has changed from a previous version of the compiler due to better packing
The x86 compiler has the same sub-optimal layout issue in all versions of Visual C++. For example, if this code is compiled for x86:
struct S { virtual ~S(); int i; double d; };
The result of sizeof(S) is 24. However, this can be reduced to 16 if you use the workaround just mentioned for x64:
struct dummy { virtual ~dummy() {} }; struct S : public dummy { virtual ~S(); int i; double d; };
Visual C++ Libraries
Standard Template Library
To enable new optimizations and debugging checks, the Visual Studio implementation of the C++ Standard Library intentionally breaks binary compatibility from one version to the next. Therefore, when the C++ Standard Library is used, object files and static libraries that are compiled by using different versions can't be mixed in one binary (EXE or DLL), and C++ Standard Library objects can't be passed between binaries that are compiled by using different versions. Such mixing emits linker errors about _MSC_VER mismatches. (_MSC_VER is the macro that contains the compiler's major version—for example, 1800 for Visual C++ in Visual Studio 2013.) This check cannot detect DLL mixing, and cannot detect mixing that involves Visual C++ 2008 or earlier.
Visual C++ in Visual Studio 2013 detects mismatches in _ITERATOR_DEBUG_LEVEL, which was implemented in Visual C++ 2010, and RuntimeLibrary mismatches. These occur when compiler options /MT (static release), /MTd (static debug), /MD (dynamic release), and /MDd (dynamic debug) are mixed. For more information, see Breaking Changes in Visual C++ 2012.
If your code acknowledges the previous release's simulated alias templates, you have to change it. For example, instead of allocator_traits<A>::rebind_alloc<U>::other, now you have to say allocator_traits<A>::rebind_alloc<U>. Although ratio_add<R1, R2>::type is no longer necessary and we now recommend that you say ratio_add<R1, R2>, the former will still compile because ratio<N, D> is required to have a "type" typedef for a reduced ratio, which will be the same type if it's already reduced.
You must use #include <algorithm> when you call std::min() or std::max().
If your existing code uses the previous release’s simulated scoped enums—traditional unscoped enums wrapped in namespaces—you have to change it. For example, if you referred to the type std::future_status::future_status, now you have to say std::future_status. However, most code is unaffected—for example, std::future_status::ready still compiles.
explicit operator bool() is stricter than operator unspecified-bool-type(). explicit operator bool() permits explicit conversions to bool—for example, given shared_ptr<X> sp, both static_cast<bool>(sp) and bool b(sp) are valid—and Boolean-testable "contextual conversions" to bool—for example, if (sp), !sp, sp && whatever. However, explicit operator bool() forbids implicit conversions to bool, so you can't say bool b = sp; and given a bool return type, you can't say return sp.
Now that real variadic templates are implemented, _VARIADIC_MAX and related macros have no effect. If you're still defining _VARIADIC_MAX, it is just ignored. If you acknowledged our macro machinery intended to support simulated variadic templates in any other way, you have to change your code.
In addition to ordinary keywords, STL headers now forbid the macro-izing of the context-sensitive keywords "override" and "final".
reference_wrapper/ref()/cref() now forbid binding to temporary objects.
<random> now strictly enforces its compile-time preconditions.
Various STL type traits have the precondition "T shall be a complete type". Although the compiler now enforces this more strictly, it may not enforce it in all situations. (Because STL precondition violations trigger undefined behavior, the Standard doesn't guarantee enforcement.)
The STL does not support /clr:oldSyntax.
The C++11 specification forcommon_type<> had unexpected and undesired consequences; in particular, it makes common_type<int, int>::type return int&&. Therefore, Visual C++ implements the Proposed Resolution for Library Working Group issue 2141, which makes common_type<int, int>::type return int.
As a side-effect of this change, the identity case no longer works (common_type<T> does not always result in type T). This complies with the Proposed Resolution, but it breaks any code that relied on the previous behavior.
If you require an identity type trait, don't use the non-standard std::identity that's defined in <type_traits> because it won't work for <void>. Instead, implement your own identity type trait to suit your needs. Here's an example:
template <typename T> struct Identity { typedef T type; };
MFC and ATL
MFC MBCS Library is no longer included in Visual Studio because Unicode is so popular and use of MBCS is significantly reduced. This change also keeps MFC more closely aligned with the Windows SDK itself, because many of the new controls and messages are Unicode-only. However, if you must continue to use the MFC MBCS library, you can download it from the MSDN Download Center. The Visual C++ Redistributable Package still includes this library.
Accessibility for the MFC ribbon is changed. Instead of a one-level architecture, there is now a hierarchical architecture. You can still use the old behavior by calling CRibbonBar::EnableSingleLevelAccessibilityMode().
CDatabase::GetConnect method is removed. To improve security, the connection string is now stored encrypted and is decrypted only as needed; it cannot be returned as plain text. The string can be obtained by using the CDatabase::Dump method.
Signature of CWnd::OnPowerBroadcast is changed. The signature of this message handler is changed to take an LPARAM as the second parameter.
Signatures are changed to accommodate message handlers. The parameter lists of the following functions have been changed to use newly added ON_WM_* message handlers:
CWnd::OnDisplayChange changed to (UINT, int, int) instead of (WPARAM, LPARAM) so that the new ON_WM_DISPLAYCHANGE macro can be used in the message map.
CFrameWnd::OnDDEInitiate changed to (CWnd*, UINT, UNIT) instead of (WPARAM, LPARAM) so that the new ON_WM_DDE_INITIATE macro can be used in the message map.
CFrameWnd::OnDDEExecute changed to (CWnd*, HANDLE) instead of (WPARAM, LPARAM) so that the new ON_WM_DDE_EXECUTE macro can be used in the message map.
CFrameWnd::OnDDETerminate changed to (CWnd*) as the parameter instead of (WPARAM, LPARAM) so that the new ON_WM_DDE_TERMINATE macro can be used in the message map.
CMFCMaskedEdit::OnCut changed to no parameters instead of (WPARAM, LPARAM) so that the new ON_WM_CUT macro can be used in the message map.
CMFCMaskedEdit::OnClear changed to no parameters instead of (WPARAM, LPARAM) so that the new ON_WM_CLEAR macro can be used in the message map.
CMFCMaskedEdit::OnPaste changed to no parameters instead of (WPARAM, LPARAM) so that the new ON_WM_PASTE macro can be used in the message map.
#ifdefs in the MFC header files are removed. Numerous #ifdefs in the MFC header files related to unsupported versions of Windows (WINVER < 0x0501) are removed.
ATL DLL (atl120.dll) is removed. ATL is now provided as headers and a static library (atls.lib).
Atlsd.lib, atlsn.lib, and atlsnd.lib are removed. Atls.lib no longer has character-set dependencies or code that's specific for debug/release. Because it works the same for Unicode/ANSI and debug/release, only one version of the library is required.
ATL/MFC Trace tool is removed together with the ATL DLL, and the tracing mechanism is simplified. The CTraceCategory constructor now takes one parameter (the category name), and the TRACE macros call the CRT debug reporting functions.