Compile Time Assertion is Unreliable?

Question

I am going through the first chapters of the book "Modern C++ Design". In particular, compile time assertions. I have some problem with the following code:

template<bool> struct CompileTimeChecker { CompileTimeChecker(...) {} }; template<> struct CompileTimeChecker<false> {}; #define STATIC_CHECK(expr, msg)\ {\ struct ERROR_##msg {ERROR_##msg() {}};\ CompileTimeChecker<((expr) != 0)>(ERROR_##msg());\ } int main() { STATIC_CHECK(0, MessageNull); STATIC_CHECK(1, MessageOne); }

This doesn't raise a compile time error for g++ 7.4.0 and clang++ 6.0.0. However, the following code does raise an error (as expected):

template<bool> struct CompileTimeChecker { CompileTimeChecker(...) {} }; template<> struct CompileTimeChecker<false> {}; #define STATIC_CHECK(expr, msg)\ {\ struct ERROR_##msg {ERROR_##msg(int i) {i;}};\ CompileTimeChecker<((expr) != 0)>(ERROR_##msg(0));\ } int main() { STATIC_CHECK(0, MessageNull); STATIC_CHECK(1, MessageOne); }

The only difference in the second code is the usage of a constructor with parameters.

The expected error message in both cases is:

g++: no matching function for call to ‘CompileTimeChecker<false>::CompileTimeChecker(main()::ERROR_MessageNull)
clang++: no matching conversion for functional-style cast from 'ERROR_MessageNull' to 'CompileTimeChecker<(0 != 0)>'

Also, it should be CompileTimeChecker<(expr)>, not CompileTimeChecker<(expr != 0)>, because this is a macro. — L. F.
– L. F., Commented Jul 23, 2019 at 10:38
@Yksisarvinen: Mostly because "Modern C++ Design" is no longer modern. It predates C++11 and static_assert. — MSalters
– MSalters, Commented Jul 23, 2019 at 14:12

L. F. · Accepted Answer · 2019-07-23 12:18:47Z

This is called the most vexing parse. The following statement:

CompileTimeChecker<expr>(Type());

is equivalent to

CompileTimeChecker<expr> Type();

which declares a function named Type. You can work around the issue by using the = form of initialization:

CompileTimeChecker<expr> var = Type();

This way, it cannot be interpreted as a declaration. You can also use the {} initialization since C++11. On the other hand,

CompileTimeChecker<expr>(Type(0));

is an expression statement that creates an object as desired because Type(0) cannot possibly declare a function.

Since C++11, just use static_assert.

Yes, and also it's specifically allowed to have both a struct and a function with the same name in the same scope, for header compatibility with C.
The parentheses didn't fix it for me because function((a)) is equal to function(a). But you are right about the function/constructor ambiguity. Writing CompileTimeChecker<expr>(Type{}); fixed it for me (i.e. using uniform initialization). Can you update your answer?
@L.F. CompileTimeChecker<expr>((Type())); is not working for me.
@image I am wrong. I confused this case with another case of most vexing parse. I have updated the answer. Hope it doesn't vex us again!

Collectives™ on Stack Overflow

Compile Time Assertion is Unreliable?

1 Answer 1

4 Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Related