https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68525
Martin Sebor <msebor at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Resolution|--- |INVALID
--- Comment #2 from Martin Sebor <msebor at gcc dot gnu.org> ---
Actually, I see that 5.1 accepts the code in C++ 98 mode but rejects it in C++
11 mode. Clang behaves the same way. Probably due to a change to the
standard.
In hindsight I agree that the test case is invalid. I even made that point
recently in a review of a patch submitted to make it valid (forgetting I had
previously raised this bug):
https://gcc.gnu.org/ml/gcc-patches/2016-01/msg00055.html
Unfortunately, there is a bug here that makes GCC accept the following
(equivalent and also invalid) code:
cat x.cpp && ~/bin/gcc-5.1.0/bin/g++ x.cpp -c -std=c++11
template <class T>
struct A {
int n;
T a[];
A (): n (), a () { }
};
struct B { B (int); };
A<B> ab;
A few days ago I submitted bug 69127 for a similar problem with zero-length
arrays. I suppose that bug could be used to fix this problem as well, even
though it's about flexible array members.
Let me close this bug as INVALID.
