:: Re: [Libbitcoin] [PATCH] Fix C++11 …
Αρχική Σελίδα
Delete this message
Reply to this message
Συντάκτης: Amir Taaki
Ημερομηνία:  
Προς: libbitcoin
Αντικείμενο: Re: [Libbitcoin] [PATCH] Fix C++11 language violations
For others interested:

http://stackoverflow.com/questions/6893700/how-to-construct-stdarray-object-with-initializer-list

On 14/02/14 08:43, William Swanson wrote:
> On Thu, Feb 13, 2014 at 11:02 PM, Amir Taaki <genjix@???> wrote:
>> Hey, the thing is that clang is broken and isn't implementing C++11 fully.
>>
>> https://en.wikipedia.org/wiki/C%2B%2B11#Uniform_initialization
>>
>> It's an error with the compiler, not the code and the double syntax
>> actually means something different (create a list with a single element
>> initialized to those items). But due to some C++ things it compiles and
>> casts it back which isn't what you expect is happening.
>>
>> I'm hoping clang fixes their compiler, but for now it can work on mac
>> using gcc (we've had a few people do it).
>
> I have already researched this, and no, Clang is actually correct! As
> it turns out, the std::array type is an aggregate type, which,
> according to the standard, provides neither constructor or destructor.
> Therefore, initialization works just like old-style C struct
> initialization. You need a pair of braces for the class, and another
> pair of braces for the array inside. It is similar to initializing
> this thing:
>
> class example {
> public:
> int data[2];
> int x;
> };
>
> example x = {{0, 0}, 0} /* This is obviously correct */
> example x = {0, 0, 0}; /* But C also allowed this! */
> example x{{0, 0}, 0} /* C++11 now allows this */
> example x{0, 0, 0}; /* C++11 does not allow this, but GCC wrongly accepts it! */
>
> The reason your code worked before is because of a bug in GCC, which
> would wrongly accept that last example line. Clang correctly rejects
> it.
>
> This is not the only error the patch addresses, but it is the most
> common. For example, you used "std::string" in a file that didn't have
> an "#include <string>". This only worked by accident with that
> particular implementation of the standard library. In another case,
> you passed an int into a uniform initializer which expected a size_t.
> This is a narrowing conversion, which the C++11 standard does not
> allow here (see the Wikipedia article you just linked).
>
> So, this isn't some fly-by-night patch; it's something I've actually
> verified according to the standard. The Android Clang has other C++11
> problems besides these, but I have solved those by patching the
> standard library rather that libbitcoin.
>
> -William
>