Re: [release] 1.75.0 post-beta merges

From: Adam Wulkiewicz (adam.wulkiewicz_at_[hidden])
Date: 2020-12-02 21:55:42

śr., 2 gru 2020 o 22:37 Glen Fernandes <glen.fernandes_at_[hidden]>
napisał(a):

> On Wed, Dec 2, 2020 at 4:27 PM Adam Wulkiewicz
> <adam.wulkiewicz_at_[hidden]> wrote:
> >
> > wt., 1 gru 2020 o 01:34 Glen Fernandes <glen.fernandes_at_[hidden]>
> napisał(a):
> >>
> >> On Mon, Nov 30, 2020 at 6:56 PM Adam Wulkiewicz via Boost
> >> <boost_at_[hidden]> wrote:
> >> >
> >> > W dniu 29.11.2020 o 21:19, Glen Fernandes pisze:
> >> > > Got it. OK to merge (all of them, including 42bd7cfe).
> >> >
> >> > Thanks! I'd like to merge one more trivial workaround:
> >> >
> https://github.com/boostorg/geometry/commit/d2d3a7954a0006af308fb87e5485750d31a1d228
> >>
> >> That removes it for MSVC 2015 v14.0 (i.e. 1900 or older), but I was
> >> under the impression that even MSVC 2017 v15.0 (i.e. 1910) is not
> >> enough. I thought the constexpr issues were fixed with MSVC 2017 v15.8
> >> or higher   (i.e. 1915).
> >>
> >> For that reason BOOST_NO_CXX14_CONSTEXPR is also defined for all
> >> versions of MSVC below 2017 v15.8 (1915).
> >
> >
> > I think your understanding is technically correct. AFAIU it's because
> the general-purpose Boost.Config macro has to work in all cases and
> accommodate for various quirks of different compilers. However in this
> particular case this workaround seems to be enough. All of our tests
> compile, with this workaround even with VS2015.
>
> If you're testing 2017 v15.0  (not 15.8+) and it passes, then using >=
> 1910 is fine. You should make it BOOST_MSVC not _MSC_VER since I'm
> sure here you mean actual MSVC and not anything else (cygwin, mingw or
> whatever else) that defines it to pretend to be MSVC.
>

Yes, I probably should've used BOOST_MSVC instead. But doing it now, as a
part of this change I think would not be wise. It's because there are 43
other places in Boost.Geometry when this should be done. So I'd prefer to
do it later, consistently across the whole library.

Adam