[cfe-dev] Standards interpretation question wrt to Clang trunk

[Howard Hinnant]

On Nov 22, 2012, at 11:27 AM, Ryan Molden <ryanmolden at gmail.com> wrote:

> On Thu, Nov 22, 2012 at 2:14 AM, Richard Smith <richard at metafoo.co.uk> wrote:
> On Wed, Nov 21, 2012 at 9:52 PM, Ryan Molden <ryanmolden at gmail.com> wrote:
> > Is this the right alias for discussion of Clang implementation specifics? Or
> > would that be better over on cfe-commits?
> 
> This is the right place.
> 
> > Assuming this is an appropriate alias (or that I can simply forward this on
> > if not), I am fixing some missing compiler intrinsics that Clang doesn't
> > provide but the VS 2012 STL headers require for some of their type traits
> > implementations.
> >
> > In doing this I have found a couple of things in Clang's current type trait
> > support that seem incorrect by my reading of the standards, but I wanted to
> > run this by other people before I changed anything.
> >
> > The first issue is this code:
> >
> > struct NotTrivial {
> >   NotTrivial() = delete;
> > };
> >
> > int main(int,char**)
> > {
> >     static_assert(!__is_trivial(NotTrivial), "A class with no default
> > constructor doesn't meet the definition of a trivial class");
> >     return 0;
> > }
> >
> > This fires today in Clang trunk.
> >
> > Looking at the standard I see this:
> >
> > The definition of a trivial type is n3376 3.9 [basic.types]/9
> >
> > Arithmetic types (3.9.1), enumeration types, pointer types, pointer to
> > member types (3.9.2), std::nullptr_t, and cv-qualified
> > versions of these types (3.9.3) are collectively called scalar types. Scalar
> > types, POD classes (Clause 9), arrays of such types and cv-qualified
> > versions of these types (3.9.3) are collectively called POD types. Scalar
> > types, trivially copyable class types (Clause 9), arrays of such types, and
> > cv-qualified versions of these types (3.9.3) are collectively called
> > trivially copyable types. Scalar types, trivial class types (Clause 9),
> > arrays of such types and cv-qualified versions of these types (3.9.3) are
> > collectively called trivial types. Scalar types, standard-layout class types
> > (Clause 9), arrays of such types and cv-qualified
> > versions of these types (3.9.3) are collectively called standard layout
> > types.
> >
> > The referenced Clause 9 says (n3376 9 [class]/6):
> >
> > A trivial class is a class that has a trivial default constructor (12.1) and
> > is trivially copyable.
> > [ Note: In particular, a trivially copyable or trivial class does not have
> > virtual functions or virtual base
> >
> > Since NotTrivial has no default constructor it seems it can't meet this
> > definition.
> 
> Clang is behaving correctly. NotTrivial has a default constructor,
> which is deleted, and thus not user-provided. It meets the other
> constraints for triviality, so it is a trivial default constructor.
> 
> (Note that deleted functions still exist, and can still be trivial.)
> 
> > The second issue (this one I am less sure about) is in this code (this also
> > fires in Clang trunk):
> >
> > struct NotTriviallyCopyable {
> >   NotTriviallyCopyable(const NotTriviallyCopyable&) = delete;
> >   NotTriviallyCopyable& operator=(const NotTriviallyCopyable&) = delete;
> > };
> >
> > int main(int,char**)
> > {
> >     static_assert(!__is_trivially_copyable(NotTriviallyCopyable), "A class
> > with no copy constructor or copy assignment operator is not trivially
> > copyable");
> >     return 0;
> > }
> >
> > This one hinges on the definition of trivially-copyable, which seems like it
> > can be read two ways:
> >
> > n3376 9 [class]/6:
> >
> > A trivially copyable class is a class that:
> >   — has no non-trivial copy constructors (12.8),
> >   — has no non-trivial move constructors (12.8),
> >   — has no non-trivial copy assignment operators (13.5.3, 12.8),
> >   — has no non-trivial move assignment operators (13.5.3, 12.8), and
> >   — has a trivial destructor (12.4).
> >
> > It seems since this class has NO copy constructors or copy assignment
> > operators that it shouldn't be considered trivially copyable.
> 
> Clang is correct here too. This class has a trivial (deleted) copy
> constructor and a trivial (deleted) copy assignment. It has no move
> constructor and no move assignment. It also has a trivial destructor.
> Therefore it is trivially copyable.
> 
> Interesting. My conceptualization of deleted functions was that they any reference to them in code would be ill-formed. It isn't clear how one uses some of the type traits (like is_default_constructible, is_copy_constructible, etc..) in meaningful ways in the face of deleted functions. It seems like it would lead one down a path of choosing functions/code paths that end up being ill-formed because the thing you tested for and were told existed by the compiler was actually deleted.
> 
> Another example, I notice that UTT_HasTrivialDefaultConstructor maps to the intrinsic __has_trivial_constructor, which reports true for a class with a deleted default constructor. Wouldn't that naturally lead one to think code like T t; is valid, when it would not be?

Some traits are more useful than others.  I recommend std::is_trivially_default_constructible, std::is_trivially_copy_constructible and std::is_trivially_copy_assignable for answering the questions you are asking.

Howard