Re: [Boost-users] error_code using c++11 and boost

Subject: Re: [Boost-users] error_code using c++11 and boost
From: Gavin Lambert (boost_at_[hidden])
Date: 2018-12-06 22:20:38

On 7/12/2018 03:34, Cristian Morales Vega wrote:
> One "new" thing listed in the 1.69 release notes is "Boost.System is
> now header-only". But I can see a reference to
> BOOST_ERROR_CODE_HEADER_ONLY in Boost.System from 2012
> (https://github.com/boostorg/system/commit/65e14ff1670ea4cf41fb31d989b31ce9d93a03af).
> And Chrono documentation has been documenting the "undocumented
> feature" (BOOST_ERROR_CODE_HEADER_ONLY) for a long time. Up to 1.54 it
> said "but it is buggy (see #7347 duplicate symbol while
> BOOST_ERROR_CODE_HEADER_ONLY is defined)", but that warning was
> removed in 1.55.

BOOST_ERROR_CODE_HEADER_ONLY only worked correctly with *some* build 
configurations and platforms -- notably, error categories that should 
compare equal did *not* do so with some combinations of DLLs and shared 
libraries.

It was not defined by default so that people have to consciously decide 
to shoot themselves in the foot (or that they have tested that for their 
particular application and platform it is not a problem).

In 1.69, the way that error_category comparison works has been altered 
(see the new constructor mentioned in the release notes), which permits 
distinct category objects to still compare equal, which means that it is 
now safe to build the standard error_category objects header-only in all 
cases.

What perhaps should be emphasized a bit stronger in the release notes is 
that it would be a very good idea for any custom error_category objects 
(especially those in libraries) to make use of these ids as well.  (It's 
documented how to do that in the reference docs for the constructor.)