An overflow in a constant expression is not detected

[Quuxplusone]

Bugzilla Link	PR14043
Status	NEW
Importance	P enhancement
Reported by	Tomasz Mikolajczyk (tmmikolajczyk@gmail.com)
Reported on	2012-10-08 16:55:19 -0700
Last modified on	2012-10-09 09:18:39 -0700
Version	3.1
Hardware	PC Linux
CC	efriedma@quicinc.com, llvm-bugs@lists.llvm.org, pawel@32bitmicro.com, richard-llvm@metafoo.co.uk
Fixed by commit(s)
Attachments	fib.cpp (885 bytes, text/x-c++src)
Blocks
Blocked by
See also

Created attachment 9321
Application calculating Fibonacci seq at compile time

Consider this trivial Fibonacci sequence calculation at compile time:

template<int v>
struct fib
{
    static_assert(v > 1, "fib<0> and fib<1> are special");
    enum { value = fib<v-1>::value + fib<v-2>::value };
};

template<>
struct fib<0> { enum { value = 0 }; };

template<>
struct fib<1> { enum { value = 1 }; };

std::cout << fib<0>::value << std::endl;
std::cout << fib<1>::value << std::endl;
std::cout << fib<2>::value << std::endl;
std::cout << fib<3>::value << std::endl;
std::cout << fib<4>::value << std::endl;
std::cout << fib<5>::value << std::endl;

std::cout << fib<45>::value << std::endl; // 1134903170
std::cout << fib<46>::value << std::endl; // 1836311903
std::cout << fib<47>::value << std::endl; // integer overflow

gcc (4.7.1) complains when tries to instantiate the fib<47>:
~$ g++ -std=c++11 fib.cpp
fib.cpp: In instantiation of `struct fib<47>':
fib.cpp:30:25:   required from here
fib.cpp:7:10: warning: integer overflow in expression [-Woverflow]
fib.cpp:7:10: error: overflow in constant expression [-fpermissive]
fib.cpp:7:10: error: enumerator value for `value' is not an integer constant

However clang keeps going silently:
~$ clang++ -std=c++11 fib.cpp
~$ ./a.out
0
1
1
2
3
5
8
13
21
34
1134903170
1836311903
-1323752223
~$

Would be good to have the overflow detected as well. The fib.cpp file from the
attachment contains the whole app.

[Quuxplusone]

Attached fib.cpp (885 bytes, text/x-c++src): Application calculating Fibonacci seq at compile time

[Quuxplusone]

Hmm... we do print a warning with -pedantic, but it might be a good idea to print something by default in this case.

[Quuxplusone]

Ironically, we accept this by default for GCC compatibility. Looks like they upgraded this from pedwarn to permerror for C++ in 4.5 or 4.6. Bumping this from Extension to ExtWarn seems sensible to me.

GCC still treats this as a pedwarn in C, but it seems reasonable to me for us to give a warning by default in all language modes (though for C, we still wouldn't catch overflow, because the C ICE checker doesn't evaluate).