Re: [boost] [chrono][timer][release] Non-inline functions in inlined headers

Subject: Re: [boost] [chrono][timer][release] Non-inline functions in inlined headers
From: Vicente J. Botet Escriba (vicente.botet_at_[hidden])
Date: 2017-12-14 10:51:36

Le 14/12/2017 à 10:52, Vicente J. Botet Escriba via Boost a écrit :
> Le 13/12/2017 à 11:24, Daniel James via Boost a écrit :
>> Hi,
>>
>> We've had a link problem reported. for this code on Visual Studio,
>> using auto-linking:
>>
>> #include "boost\chrono.hpp"
>> #include "boost\timer\timer.hpp"
>>
>> int main()
>> {
>> Â Â Â Â  auto now = boost::chrono::steady_clock::now();
>> Â Â Â Â  boost::timer::cpu_timer cput;
>> Â Â Â Â  return 0;
>> }
>>
>> I haven't got the binaries installed yet, so I haven't reproduced it,
>> but I think it's caused because timer now includes the chrono headers
>> using BOOST_CHRONO_HEADER_ONLY, and in the chrono inlined headers not
>> all functions are BOOST_CHRONO_INLINE. It looks like this problem was
>> always in chrono, but BOOST_CHRONO_HEADER_ONLY was widely used enough
>> for it to be detected.
>>
>> I'd quite like to fix this for 1.66.0, which requires either reverting
>> the change to timer, or fixing the problem in chrono. The error is in
>> chrono, but I'm more inclined to revert the change in timer for this
>> release, as it's a less disruptive way of quickly fixing the problem.
>>
>> https://github.com/boostorg/timer/pull/5
>>
> Hi,
>
>
> Sorry, I forgot completely about  BOOST_CHRONO_INLINE.
>
> I see two uses of inline
>
> mac/process_cpu_clocks.hpp:Â Â Â Â Â  inline long tick_factor() // 
> multiplier to convert ticks
> posix/process_cpu_clocks.hpp:Â  inline nanoseconds::rep 
> tick_factor()Â Â Â Â Â Â Â  // multiplier to convert ticks
This is not the source of any trouble, isn't it?

Sorry for the noise.

Vicente