[cfe-dev] thread(F&&) fails when F is not MoveAssignable

[Howard Hinnant]

On Dec 16, 2012, at 7:16 PM, Gordon Henriksen <gordonhenriksen at me.com> wrote:

> On 2012-12-16, at 15:49, Gordon Henriksen wrote:
> 
>> On 2012-12-16, at 15:36, Howard Hinnant wrote:
>> 
>>> let's zero in on a precise copy/pasteable test.
>> 
>> Agreed. I'm attempting to reduce a test case.
> 
> Found it; the correct reduction has one additional line:
> 
>    #define decltype __typeof // CULPRIT
> 
> #ifdefing out that 'help' for clang permits the test to work without modification.
> 
> Reduction is below.
> 
> — Gordon
> 
> 
> cat <<'EOT' >test.cpp
> #define decltype __typeof // CULPRIT
> #include <thread>
> #include <iostream>
> #include <cassert>
> 
> namespace {
>   class SimpleJoinThread {
>       bool &flag_;
>   public:
>       SimpleJoinThread(bool &flag) : flag_(flag) { }
>       void operator()() { flag_ = true; }
>   };
> } // End anonymous namespace.
> 
> int main()
> {
>   bool flag = false;
>   std::thread(SimpleJoinThread(flag)).join();
>   assert(flag);
> }
> EOT
> 
> xcrun clang++ -std=c++0x -stdlib=libc++ test.cpp
> ./test
> 

Interesting.  Nice detective work.

I'm actually not duplicating your symptoms, but that may be because I currently have later versions of libc++ installed.  Your detective work seems spot-on to me.  test bug.

Fwiw, libc++ does a similar dance in C++03 mode.  <__config> has:

#if !(__has_feature(cxx_decltype))
#define _LIBCPP_HAS_NO_DECLTYPE
#endif

// ...

#ifdef _LIBCPP_HAS_NO_DECLTYPE
#define decltype(x) __typeof__(x)
#endif

Howard