Avoiding preprocessor for mutual exclusive function call in C++20 - c++

Please consider the following little function. It provides compiler abstraction for breaking the debugger programmatically:
inline constexpr void BreakDebug()
{
#ifdef __GNUC__
__builtin_trap();
#elif defined _MSC_VER
__debugbreak();
#endif
}
I would like to rewrite the function and replace the preprocessor instructions using C++20 code. Since __builtin_trap and __debugbreak are compiler specific and mutual exclusive, I can't use a simple if constexpr since I would get a compilation error.
Assuming I would wrap the compiler macros __GNUC__ and _MSC_VER using a constexpr enumeration constant... how could that be done?

Apparently, those builtin functions can also be turned into dependent name. I checked this here. So the usual trick that consists in making potentially ill-formed code dependent works:
enum compiler_t{
gnu,
msvc
};
inline constexpr compiler_t compiler =
#ifdef __GNUC__
compiler_t:: gnu;
#else
compiler_t:: msvc;
#endif
template <class...Args>
inline constexpr void BreakDebug(Args...args)
{
if constexpr (compiler == gnu){
__builtin_trap(args...);
} else {
__debugbreak(args...);
}
}
int main(){
BreakDebug();
return 0;
}

You only need a declaration of the function that is not called (and no definition, as it won't be called):
inline constexpr void BreakDebug()
{
#ifdef __GNUC__
constexpr bool GNU = true;
constexpr void __debugbreak();
#elif defined _MSC_VER
constexpr bool GNU = false;
constexpr void __builtin_trap();
#endif
if constexpr (GNU){
__builtin_trap();
} else {
__debugbreak();
}
}
By declaring only that function that does not exist and that you will never call, you can avoid linker errors in case the signature is not the correct one.

Related

C++ constexpr boolean vs macro

I have been reading about using constexpr instead of using macros for better safety, as macros can lead to undefined behavior. However, if I have the code in the following example, is it okay to do this, and just use preprocessor if statements (#if), or should I change the macros to constexpr bool.
Here is what I wanted to do:
#if NDEBUG
#define ENABLE_VALIDATION_LAYERS 0
#else
#define ENABLE_VALIDATION_LAYERS 1
#endif
vs
#if NDEBUG
constexpr bool enable_validation_layers = false;
#else
constexpr bool enable_validation_layers = true;
#endif
Code usage right now looks like this:
//if validation layers are not enabled, the validation support does not need to be checked
#if ENABLE_VALIDATION_LAYERS
if (!CheckValidationLayerSupport()) //function that returns a bool to check if validation layers are supported
{
throw std::runtime_error("validation layers requested, but not available!");
}
#endif

Is there any way to change function prototypes at compile time in C++?

suppose you have to write code that should compile under C++,C++11,C++17,etc.
A function like this for example.
bool Ispalindrome(const std::string &str)
{
// Code
}
That compiles under all C++ implementations. But if you want to use the old and the new the C++17 string_view feature you have deal with something similar to
#ifdef LEGACYCPP
bool Ispalindrome(const std::string &str)
{
// Code
}
#elseif CPP17
bool Ispalindrome(std::string_view str)
{
// OMG Repeated Code
}
#endif
Using conditional compiling is right, but have to repeat code.
Is there any way to choose a function prototype at compile time ? Or other way to circumvent double coding ? (in situations where it can be applied)
Thanks
In a header file, you'll need to do something like
#if __cplusplus >= 201703L
#include <string_view>
bool Ispalindrome(std::string_view str);
#else
#include <string>
bool Ispalindrome(const std::string &str);
#endif
In your definition, you'll need to include the header and then do
#if __cplusplus >= 201703L
bool Ispalindrome(std::string_view str)
#else
bool Ispalindrome(const std::string &str)
#endif
{
// OMG no repeated code
}
__cplusplus is specified in the standard, and is predefined in every compilation unit. 201703L indicates a C++17 compiler, larger values more recent standards.
This assumes an implementation (compiler and library) that correctly claims compliance with the respective C++ standard.
If you can be sure that the code section of your function will be equivalent in each case, then you can use a macro definition for the parameter list:
#ifdef ISCPP17
#define PALIN_PARAMS std::string_view str
#else
#define PALIN_PARAMS const std::string& str
#endif
bool Ispalindrome(PALIN_PARAMS)
{
// Code
}
#undef PALIN_PARAMS
There are, of course, many variations on this theme: you could leave out the "str" part in the macro definition and have (PALIN_PARAMS str) in the signature. But, using "as is" will also allow for multiple parameters with different types.
But I'm not sure this sort of thing will pass the Inquisitions of the C++ Puritans.
Another (possibly more robust) way would be to use conditional compilation blocks with using (or typedef) statements to define argument types:
#ifdef ISCPP17
using cnststrg = std::string_view; // Or typedef std::string_view cnststrg;
#else
using cnststrg = const std::string&;
#endif
bool Ispalindrome(cnststrg str)
{
// Code
return true;
}
Provided it's exactly repeated code then, do following minor change:
#ifdef CPP17
bool Ispalindrome(std::string_view str)
#else // all versions of LEGACYCPP
bool Ispalindrome(const std::string &str)
#endif
{
// Same Code
}
If a minor part of function is unique to versions, then apply the above same trick there too.

Why noreturn/__builtin_unreachable prevents tail call optimization

I have come to fact that all major compilers will not do tail call optimization if a called function does not return (i.e. marked as _Noreturn/[[noreturn]] or there is a __builtin_unreachable() after the call). Is this an intended behavior and not a missed optimization, and if so why?
Example 1:
#ifndef __cplusplus
#define NORETURN _Noreturn
#else
#define NORETURN [[noreturn]]
#endif
void canret(void);
NORETURN void noret(void);
void foo(void) { canret(); }
void bar(void) { noret(); }
C: https://godbolt.org/z/pJfEe-
C++: https://godbolt.org/z/-4c78K
Example 2:
#ifdef _MSC_VER
#define UNREACHABLE __assume(0)
#else
#define UNREACHABLE __builtin_unreachable()
#endif
void f(void);
void foo(void) { f(); }
void bar(void) { f(); UNREACHABLE; }
https://godbolt.org/z/PFhWKR
It's intentional, though perhaps controversial since it can seriously harm stack usage properties; for this reason I've even resorted to tricking the compiler to think a function that can't return can. The reasoning is that many noreturn functions are abort-like (or even call abort), and that it's likely someone running a debugger wants to be able to see where the call happened from -- information which would be lost by a tail call.
Citations:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=10837
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56165
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67327
etc.

C++: cleanest practice to write an empty function body [duplicate]

I have a cross platform application and in a few of my functions not all the values passed to functions are utilised. Hence I get a warning from GCC telling me that there are unused variables.
What would be the best way of coding around the warning?
An #ifdef around the function?
#ifdef _MSC_VER
void ProcessOps::sendToExternalApp(QString sAppName, QString sImagePath, qreal qrLeft, qreal qrTop, qreal qrWidth, qreal qrHeight)
#else
void ProcessOps::sendToExternalApp(QString sAppName, QString sImagePath, qreal /*qrLeft*/, qreal /*qrTop*/, qreal /*qrWidth*/, qreal /*qrHeight*/)
#endif
{
This is so ugly but seems like the way the compiler would prefer.
Or do I assign zero to the variable at the end of the function? (which I hate because it's altering something in the program flow to silence a compiler warning).
Is there a correct way?
You can put it in "(void)var;" expression (does nothing) so that a compiler sees it is used. This is portable between compilers.
E.g.
void foo(int param1, int param2)
{
(void)param2;
bar(param1);
}
Or,
#define UNUSED(expr) do { (void)(expr); } while (0)
...
void foo(int param1, int param2)
{
UNUSED(param2);
bar(param1);
}
In GCC and Clang you can use the __attribute__((unused)) preprocessor directive to achieve your goal.
For example:
int foo (__attribute__((unused)) int bar) {
return 0;
}
C++17 now provides the [[maybe_unused]] attribute.
http://en.cppreference.com/w/cpp/language/attributes
Quite nice and standard.
C++17 Update
In C++17 we gain the attribute [[maybe_unused]] which is covered in [dcl.attr.unused]
The attribute-token maybe_unused indicates that a name or entity is possibly intentionally unused. It shall
appear at most once in each attribute-list and no attribute-argument-clause shall be present.
...
Example:
[[maybe_unused]] void f([[maybe_unused]] bool thing1,
[[maybe_unused]] bool thing2) {
[[maybe_unused]] bool b = thing1 && thing2;
assert(b);
}
Implementations should not warn that b is unused, whether or not NDEBUG is defined. —end example ]
For the following example:
int foo ( int bar) {
bool unused_bool ;
return 0;
}
Both clang and gcc generate a diagnostic using -Wall -Wextra for both bar and unused_bool (See it live).
While adding [[maybe_unused]] silences the diagnostics:
int foo ([[maybe_unused]] int bar) {
[[maybe_unused]] bool unused_bool ;
return 0;
}
see it live.
Before C++17
In C++11 an alternative form of the UNUSED macro could be formed using a lambda expression(via Ben Deane) with an capture of the unused variable:
#define UNUSED(x) [&x]{}()
The immediate invocation of the lambda expression should be optimized away, given the following example:
int foo (int bar) {
UNUSED(bar) ;
return 0;
}
we can see in godbolt that the call is optimized away:
foo(int):
xorl %eax, %eax
ret
Your current solution is best - comment out the parameter name if you don't use it. That applies to all compilers, so you don't have to use the pre-processor to do it specially for GCC.
An even cleaner way is to just comment out variable names:
int main(int /* argc */, char const** /* argv */) {
return 0;
}
gcc doesn't flag these warnings by default. This warning must have been turned on either explicitly by passing -Wunused-parameter to the compiler or implicitly by passing -Wall -Wextra (or possibly some other combination of flags).
Unused parameter warnings can simply be suppressed by passing -Wno-unused-parameter to the compiler, but note that this disabling flag must come after any possible enabling flags for this warning in the compiler command line, so that it can take effect.
A coworker just pointed me to this nice little macro here
For ease I'll include the macro below.
#ifdef UNUSED
#elif defined(__GNUC__)
# define UNUSED(x) UNUSED_ ## x __attribute__((unused))
#elif defined(__LCLINT__)
# define UNUSED(x) /*#unused#*/ x
#else
# define UNUSED(x) x
#endif
void dcc_mon_siginfo_handler(int UNUSED(whatsig))
macro-less and portable way to declare one or more parameters as unused:
template <typename... Args> inline void unused(Args&&...) {}
int main(int argc, char* argv[])
{
unused(argc, argv);
return 0;
}
Using preprocessor directives is considered evil most of the time. Ideally you want to avoid them like the Pest. Remember that making the compiler understand your code is easy, allowing other programmers to understand your code is much harder. A few dozen cases like this here and there makes it very hard to read for yourself later or for others right now.
One way might be to put your parameters together into some sort of argument class. You could then use only a subset of the variables (equivalent to your assigning 0 really) or having different specializations of that argument class for each platform. This might however not be worth it, you need to analyze whether it would fit.
If you can read impossible templates, you might find advanced tips in the "Exceptional C++" book. If the people who would read your code could get their skillset to encompass the crazy stuff taught in that book, then you would have beautiful code which can also be easily read. The compiler would also be well aware of what you are doing (instead of hiding everything by preprocessing)
I have seen this instead of the (void)param2 way of silencing the warning:
void foo(int param1, int param2)
{
std::ignore = param2;
bar(param1);
}
Looks like this was added in C++11
Lol! I dont think there is another question on SO that reveal all the heretics corrupted by Chaos better that this one!
With all due respect to C++17 there is a clear guideline in C++ Core Guidelines. AFAIR, back in 2009 this option was available as well as today. And if somebody says it is considered as a bug in Doxygen then there is a bug in Doxygen
First off the warning is generated by the variable definition in the source file not the header file. The header can stay pristine and should, since you might be using something like doxygen to generate the API-documentation.
I will assume that you have completely different implementation in source files. In these cases you can either comment out the offending parameter or just write the parameter.
Example:
func(int a, int b)
{
b;
foo(a);
}
This might seem cryptic, so defined a macro like UNUSED. The way MFC did it is:
#ifdef _DEBUG
#define UNUSED(x)
#else
#define UNUSED(x) x
#endif
Like this you see the warning still in debug builds, might be helpful.
Is it not safe to always comment out parameter names? If it's not you can do something like
#ifdef _MSC_VER
# define P_(n) n
#else
# define P_(n)
#endif
void ProcessOps::sendToExternalApp(
QString sAppName, QString sImagePath,
qreal P_(qrLeft), qreal P_(qrTop), qreal P_(qrWidth), qreal P_(qrHeight))
It's a bit less ugly.
Using an UNREFERENCED_PARAMETER(p) could work. I know it is defined in WinNT.h for Windows systems and can easily be defined for gcc as well (if it doesn't already have it).
UNREFERENCED PARAMETER(p) is defined as
#define UNREFERENCED_PARAMETER(P) (P)
in WinNT.h.
Use compiler's flag, e.g. flag for GCC:
-Wno-unused-variable
In C++11, this is the solution I'm using:
template<typename... Ts> inline void Unreferenced(Ts&&...) {}
int Foo(int bar)
{
Unreferenced(bar);
return 0;
}
int Foo2(int bar1, int bar2)
{
Unreferenced(bar1, bar2);
return 0;
}
Verified to be portable (at least on modern msvc, clang and gcc) and not producing extra code when optimizations are enabled.
With no optimization, the extra function call is performed and references to the parameters are copied to the stack, but there are no macros involved.
If the extra code is an issue, you can use this declaration instead:
(decltype(Unreferenced(bar1, bar2)))0;
but at that point, a macro provides better readability:
#define UNREFERENCED(...) { (decltype(Unreferenced(__VA_ARGS__)))0; }
This works well but requires C++11
template <typename ...Args>
void unused(Args&& ...args)
{
(void)(sizeof...(args));
}
You can use __unused to tell the compiler that variable might not be used.
- (void)myMethod:(__unused NSObject *)theObject
{
// there will be no warning about `theObject`, because you wrote `__unused`
__unused int theInt = 0;
// there will be no warning, but you are still able to use `theInt` in the future
}
I found most of the presented answers work for local unused variable only, and will cause compile error for unused static global variable.
Another macro needed to suppress the warning of unused static global variable.
template <typename T>
const T* UNUSED_VARIABLE(const T& dummy) {
return &dummy;
}
#define UNUSED_GLOBAL_VARIABLE(x) namespace {\
const auto dummy = UNUSED_VARIABLE(x);\
}
static int a = 0;
UNUSED_GLOBAL_VARIABLE(a);
int main ()
{
int b = 3;
UNUSED_VARIABLE(b);
return 0;
}
This works because no warning will be reported for non-static global variable in anonymous namespace.
C++ 11 is required though
g++ -Wall -O3 -std=c++11 test.cpp
void func(void *aux UNUSED)
{
return;
}
smth like that, in that case if u dont use aux it wont warn u
I don't see your problem with the warning. Document it in the method/function header that compiler xy will issue a (correct) warning here, but that theses variables are needed for platform z.
The warning is correct, no need to turn it off. It does not invalidate the program - but it should be documented, that there is a reason.

How do I best silence a warning about unused variables?

I have a cross platform application and in a few of my functions not all the values passed to functions are utilised. Hence I get a warning from GCC telling me that there are unused variables.
What would be the best way of coding around the warning?
An #ifdef around the function?
#ifdef _MSC_VER
void ProcessOps::sendToExternalApp(QString sAppName, QString sImagePath, qreal qrLeft, qreal qrTop, qreal qrWidth, qreal qrHeight)
#else
void ProcessOps::sendToExternalApp(QString sAppName, QString sImagePath, qreal /*qrLeft*/, qreal /*qrTop*/, qreal /*qrWidth*/, qreal /*qrHeight*/)
#endif
{
This is so ugly but seems like the way the compiler would prefer.
Or do I assign zero to the variable at the end of the function? (which I hate because it's altering something in the program flow to silence a compiler warning).
Is there a correct way?
You can put it in "(void)var;" expression (does nothing) so that a compiler sees it is used. This is portable between compilers.
E.g.
void foo(int param1, int param2)
{
(void)param2;
bar(param1);
}
Or,
#define UNUSED(expr) do { (void)(expr); } while (0)
...
void foo(int param1, int param2)
{
UNUSED(param2);
bar(param1);
}
In GCC and Clang you can use the __attribute__((unused)) preprocessor directive to achieve your goal.
For example:
int foo (__attribute__((unused)) int bar) {
return 0;
}
C++17 now provides the [[maybe_unused]] attribute.
http://en.cppreference.com/w/cpp/language/attributes
Quite nice and standard.
C++17 Update
In C++17 we gain the attribute [[maybe_unused]] which is covered in [dcl.attr.unused]
The attribute-token maybe_unused indicates that a name or entity is possibly intentionally unused. It shall
appear at most once in each attribute-list and no attribute-argument-clause shall be present.
...
Example:
[[maybe_unused]] void f([[maybe_unused]] bool thing1,
[[maybe_unused]] bool thing2) {
[[maybe_unused]] bool b = thing1 && thing2;
assert(b);
}
Implementations should not warn that b is unused, whether or not NDEBUG is defined. —end example ]
For the following example:
int foo ( int bar) {
bool unused_bool ;
return 0;
}
Both clang and gcc generate a diagnostic using -Wall -Wextra for both bar and unused_bool (See it live).
While adding [[maybe_unused]] silences the diagnostics:
int foo ([[maybe_unused]] int bar) {
[[maybe_unused]] bool unused_bool ;
return 0;
}
see it live.
Before C++17
In C++11 an alternative form of the UNUSED macro could be formed using a lambda expression(via Ben Deane) with an capture of the unused variable:
#define UNUSED(x) [&x]{}()
The immediate invocation of the lambda expression should be optimized away, given the following example:
int foo (int bar) {
UNUSED(bar) ;
return 0;
}
we can see in godbolt that the call is optimized away:
foo(int):
xorl %eax, %eax
ret
Your current solution is best - comment out the parameter name if you don't use it. That applies to all compilers, so you don't have to use the pre-processor to do it specially for GCC.
An even cleaner way is to just comment out variable names:
int main(int /* argc */, char const** /* argv */) {
return 0;
}
gcc doesn't flag these warnings by default. This warning must have been turned on either explicitly by passing -Wunused-parameter to the compiler or implicitly by passing -Wall -Wextra (or possibly some other combination of flags).
Unused parameter warnings can simply be suppressed by passing -Wno-unused-parameter to the compiler, but note that this disabling flag must come after any possible enabling flags for this warning in the compiler command line, so that it can take effect.
A coworker just pointed me to this nice little macro here
For ease I'll include the macro below.
#ifdef UNUSED
#elif defined(__GNUC__)
# define UNUSED(x) UNUSED_ ## x __attribute__((unused))
#elif defined(__LCLINT__)
# define UNUSED(x) /*#unused#*/ x
#else
# define UNUSED(x) x
#endif
void dcc_mon_siginfo_handler(int UNUSED(whatsig))
macro-less and portable way to declare one or more parameters as unused:
template <typename... Args> inline void unused(Args&&...) {}
int main(int argc, char* argv[])
{
unused(argc, argv);
return 0;
}
Using preprocessor directives is considered evil most of the time. Ideally you want to avoid them like the Pest. Remember that making the compiler understand your code is easy, allowing other programmers to understand your code is much harder. A few dozen cases like this here and there makes it very hard to read for yourself later or for others right now.
One way might be to put your parameters together into some sort of argument class. You could then use only a subset of the variables (equivalent to your assigning 0 really) or having different specializations of that argument class for each platform. This might however not be worth it, you need to analyze whether it would fit.
If you can read impossible templates, you might find advanced tips in the "Exceptional C++" book. If the people who would read your code could get their skillset to encompass the crazy stuff taught in that book, then you would have beautiful code which can also be easily read. The compiler would also be well aware of what you are doing (instead of hiding everything by preprocessing)
I have seen this instead of the (void)param2 way of silencing the warning:
void foo(int param1, int param2)
{
std::ignore = param2;
bar(param1);
}
Looks like this was added in C++11
Lol! I dont think there is another question on SO that reveal all the heretics corrupted by Chaos better that this one!
With all due respect to C++17 there is a clear guideline in C++ Core Guidelines. AFAIR, back in 2009 this option was available as well as today. And if somebody says it is considered as a bug in Doxygen then there is a bug in Doxygen
First off the warning is generated by the variable definition in the source file not the header file. The header can stay pristine and should, since you might be using something like doxygen to generate the API-documentation.
I will assume that you have completely different implementation in source files. In these cases you can either comment out the offending parameter or just write the parameter.
Example:
func(int a, int b)
{
b;
foo(a);
}
This might seem cryptic, so defined a macro like UNUSED. The way MFC did it is:
#ifdef _DEBUG
#define UNUSED(x)
#else
#define UNUSED(x) x
#endif
Like this you see the warning still in debug builds, might be helpful.
Is it not safe to always comment out parameter names? If it's not you can do something like
#ifdef _MSC_VER
# define P_(n) n
#else
# define P_(n)
#endif
void ProcessOps::sendToExternalApp(
QString sAppName, QString sImagePath,
qreal P_(qrLeft), qreal P_(qrTop), qreal P_(qrWidth), qreal P_(qrHeight))
It's a bit less ugly.
Using an UNREFERENCED_PARAMETER(p) could work. I know it is defined in WinNT.h for Windows systems and can easily be defined for gcc as well (if it doesn't already have it).
UNREFERENCED PARAMETER(p) is defined as
#define UNREFERENCED_PARAMETER(P) (P)
in WinNT.h.
Use compiler's flag, e.g. flag for GCC:
-Wno-unused-variable
In C++11, this is the solution I'm using:
template<typename... Ts> inline void Unreferenced(Ts&&...) {}
int Foo(int bar)
{
Unreferenced(bar);
return 0;
}
int Foo2(int bar1, int bar2)
{
Unreferenced(bar1, bar2);
return 0;
}
Verified to be portable (at least on modern msvc, clang and gcc) and not producing extra code when optimizations are enabled.
With no optimization, the extra function call is performed and references to the parameters are copied to the stack, but there are no macros involved.
If the extra code is an issue, you can use this declaration instead:
(decltype(Unreferenced(bar1, bar2)))0;
but at that point, a macro provides better readability:
#define UNREFERENCED(...) { (decltype(Unreferenced(__VA_ARGS__)))0; }
This works well but requires C++11
template <typename ...Args>
void unused(Args&& ...args)
{
(void)(sizeof...(args));
}
You can use __unused to tell the compiler that variable might not be used.
- (void)myMethod:(__unused NSObject *)theObject
{
// there will be no warning about `theObject`, because you wrote `__unused`
__unused int theInt = 0;
// there will be no warning, but you are still able to use `theInt` in the future
}
I found most of the presented answers work for local unused variable only, and will cause compile error for unused static global variable.
Another macro needed to suppress the warning of unused static global variable.
template <typename T>
const T* UNUSED_VARIABLE(const T& dummy) {
return &dummy;
}
#define UNUSED_GLOBAL_VARIABLE(x) namespace {\
const auto dummy = UNUSED_VARIABLE(x);\
}
static int a = 0;
UNUSED_GLOBAL_VARIABLE(a);
int main ()
{
int b = 3;
UNUSED_VARIABLE(b);
return 0;
}
This works because no warning will be reported for non-static global variable in anonymous namespace.
C++ 11 is required though
g++ -Wall -O3 -std=c++11 test.cpp
void func(void *aux UNUSED)
{
return;
}
smth like that, in that case if u dont use aux it wont warn u
I don't see your problem with the warning. Document it in the method/function header that compiler xy will issue a (correct) warning here, but that theses variables are needed for platform z.
The warning is correct, no need to turn it off. It does not invalidate the program - but it should be documented, that there is a reason.