[cfe-dev] (libc++) Implicit type converstion, value initialization and initializer lists

[Andrew Parker]

I picked up an assert in a libc++ test with MSVC and am unsure whether
actually MSVC is actually performing as per the standard in this case.

The test is:

std/containers/sequences/forwardlist/forwardlist.modifiers/insert_after_init.pass.cpp

The code in question is:

*        typedef int T;*
*        typedef std::forward_list<T> C;*
*        typedef C::iterator I;*
*        C c;*
*        I i = c.insert_after(c.cbefore_begin(), {});*
*        assert(i == c.before_begin());*

The intention is that calling insert_after with an empty initializer list
will insert nothing, hence the assert criterion.  However, empty
initializers for *int* should be subject to value initialization, i.e. {}
should be interpreted as int i{};.  So should this call not be equivalent
to:

*I i = c.insert_after(c.cbefore_begin(), 0);?*

MSVC compiles the code in this way, i.e.

*iterator **insert_after(const_iterator __p, value_type&& __v){...}*

is called instead of

*iterator insert_after(const_iterator __p, initializer_list<value_type>
__il){...}*

hence the list grows by one in size.

My guess would be that MSVC is doing the wrong thing, as this is usually
the case.  However, I can't be certain in this case so appeal to the
experts on this.  If MSVC is correct then this actually points to a clang
bug not a libc++ one.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150714/4cf22369/attachment.html>