[cfe-users] clang++-3.3 reject valid ?

[Дмитрий Дьяченко]

g++ ICE reported as http://gcc.gnu.org/PR56961

Thanks again, David

Dmitry

2013/4/15 David Blaikie <dblaikie at gmail.com>:
> On Thu, Apr 11, 2013 at 8:51 AM, Дмитрий Дьяченко <dimhen at gmail.com> wrote:
>> Hi!
>>
>> The following code
>> $ cat -n x.ii
>>      1    extern "C" typedef struct
>>      2    {
>>      3        volatile struct
>>      4        {
>>      5        } fields;
>>      6    }
>>      7    CSPHandleState;
>>      8    typedef struct
>>      9    {
>>     10        CSPHandleState state;
>>     11    }
>>     12    CSPHandle;
>>     13    CSPHandle a;
>>     14    void
>>     15    fn1 ()
>>     16    {
>>     17        CSPHandleState b;
>>     18        b = a.state;
>>     19    }
>>
>>
>> rejected by clang++
>> $ /home/dimhen/build/csa/Release+Asserts/bin/clang++ -fsyntax-only -x c++ x.ii
>> x.ii:1:20: error: member function 'operator=' not viable: 'this'
>> argument has type
>>       'volatile struct <anonymous struct at x.ii:3:14>', but function
>> is not marked
>>       volatile
>> extern "C" typedef struct
>>                    ^~~~~~
>> x.ii:3:14: note: 'operator=' declared here
>>     volatile struct
>>              ^
>> x.ii:18:7: note: implicit default copy assignment operator for
>> 'CSPHandleState' first
>>       required here
>>     b = a.state;
>>       ^
>> 1 error generated.
>>
>> $ /home/dimhen/build/csa/Release+Asserts/bin/clang++ -v
>> clang version 3.3 (trunk 179273)
>> Target: x86_64-unknown-linux-gnu
>> Thread model: posix
>>
>>
>> but gcc accept this
>> $ g++ -Wall -Wextra -fsyntax-only x.ii
>> [dimhen at dim clang]$ echo $?
>> 0
>>
>> gcc versions:
>> gcc version 4.7.2 20121109 (Red Hat 4.7.2-8) (GCC)
>> gcc version 4.9.0 20130411 (experimental) [trunk revision 197693] (GCC)
>>
>> Original file (not the reduced testcase) was successfully compiled with
>> many gcc versions (many versions from 3.2.3 to 4.9), xlc (AIX 5/6/7),
>> suncc (Solaris 9/10 x64/sparc), MS Visual Studio (many versions from
>> 2005 to 2012)
>>
>>
>> So, my question is: what is right behavior?
>
> Seems like a case of accepts-invalid in other compilers. If you do the
> copy explicitly rather than via the implicit copy ctor, it fails:
>
> struct foo { };
> volatile foo f;
> f = foo(); // passing 'volatile foo' as 'this' argument of 'foo&
> foo::operator=(const foo&)' discards qualifiers [-fpermissive]
>
> (additional errors occur if you try to copy from a volatile as well as
> to a volatile)
>
> In fact, while reducing your test case, I even managed to ICE GCC:
>
> truct foo { };
>   typedef struct
>   {
>       volatile foo fields;
>   }
>   CSPHandleState;
>   CSPHandleState a;
>   void
>   fn1 ()
>   {
>       CSPHandleState b;
>       b.fields = foo();
>   }