Unaligned memory access in order_by #355
Comments
|
That is interesting. It is complaining about a static_cast to the derived class in a variadic CRTP. We might need a language lawyer to see if that is valid or not? |
|
I haven't seen CRTP before and I have hard time following the code. |
|
This is a drastically simplified version of what is happening: #include <iostream>
#include <string_view>
template<typename Derived>
struct Base1
{
constexpr auto& get1() const
{
return static_cast<const Derived&>(*this);
}
};
template<typename Derived>
struct Base2
{
constexpr auto& get2() const
{
return static_cast<const Derived&>(*this);
}
};
template<typename Derived>
struct Base3
{
constexpr auto& get3() const
{
return static_cast<const Derived&>(*this);
}
};
struct X : public Base1<X>, public Base2<X>, public Base3<X>
{
constexpr X(std::string_view d) : data{d} {}
std::string_view data;
};
int main()
{
constexpr auto x = X{"cheesecake"};
static_assert(x.get1().data == x.get2().data);
static_assert(&x.get1() == &x.get2());
static_assert(&x.get1() == &x.get3());
std::cout << "Base1: " << alignof(Base1<X>) << std::endl; // 1
std::cout << "Base2: " << alignof(Base2<X>) << std::endl; // 1
std::cout << "Base3: " << alignof(Base3<X>) << std::endl; // 1
std::cout << "X : " << alignof(X) << std::endl; // 8
}
This is using C++17, but that's just to make the code slightly more concise. Judging by the static_asserts, this seems to be OK despite the different alignments (I'll need to experiment with UB sanitizer, though). |
|
I tried to reproduce your code sample: #include <iostream>
#include <sqlpp11/sqlpp11.h>
#include <MockDb.h>
#include <Sample.h>
int main()
{
const auto t = test::TabBar{};
auto db = MockDb{};
for (const auto& row : db(select(all_of(t)).from(t).where(t.alpha > 15).order_by(t.alpha.asc()).limit(100u)))
{
std::cout << row.alpha;
}
}Compiled with clang-12 I got no alignment warnings when running it. I also tried with Can you create a minimal viable example of the problem? |
|
Here's a sample which produces the issue and then crashes: The issue occurs when I compile with gcc 9.3.0 but not with clang 11.0. |
|
BTW, one could add one more assertion that also passes: |
|
Thanks for the example! I can reproduce it with gcc-11, too. I'll try to create a minimal example that exposes the behavior and then start looking for a language lawyer. |
|
I found a minimal example and posted the question on stackoverflow. |
|
The answer on stackoverflow seems very plausible to me. I filed a bug for gcc. |
|
FYI: The gcc sanitizer bug is now fixed in trunk: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98206 I tested with the example above: Works fine now. |
We're running our unit tests with undefined behaviour sanitizer and it has caught the following misaligned access:
While executing
The text was updated successfully, but these errors were encountered: