gcc 5.1.1 requires linkage with -lpthread?

[Jonathan Wakely]

If you're using std::thread then you always needed to link to libpthread or
you'd get an exception at run time.

Now you get a failure earlier at link time, but there's no new dependency.

On Jun 6, 2015 6:24 PM, "Sam Varshavchik" <mrsam@courier-mta.com> wrote:

> Rebuilding a bunch of code from 4.9.2 to 5.1.1 resulted in a bunch of
> complaints about unresolved references to pthread_create(), and admonishing
> me for not linking with -lpthread :-)
>
> Which was odd, since the code in question uses std::thread exclusively,
> and should only need to be linked with the default libstdc++, helpfully
> supplied by the compiler.
>

No. Libstdc++ never linked to libpthread, it was always your job.


> So, I chased this down to <thread> now making visible an explicit
> reference to pthread_create, in std::thread's constructor:
>
>    template<typename _Callable, typename… _Args>
>      explicit
>      thread(_Callable&& __f, _Args&&… __args)
>      {
>        _M_start_thread(_M_make_routine(std::__bind_simple(
>                std::forward<_Callable>(__f),
>                std::forward<_Args>(__args)…)),
>     reinterpret_cast<void(*)()>(&pthread_create));
>      }
>
> So that's what I'm compiling. This may be because of the ld.so
> configuration in Fedora. Even if libstdc++ is linked with -lpthread, this
> ends up b compiled as part of my code, at -O2, apparently, and I'm not
> linking with -lpthread.
>
> Not really a complicated thing to fix up, of course. Just seems a bit odd,
> to end up requiring a linking dependency, in this manner.
>
>