On Mon, 26 Oct 2020 at 16:16, Mike via Boost
Gesendet: Montag, 26. Oktober 2020 um 11:41 Uhr Von: "Mateusz Loskot via Boost"
On Mon, 26 Oct 2020 at 11:17, Mike via Boost wrote: E.g. I think the first c++17 bits have been implemented in gcc 6 and gcc 7 had mostly complete support, but it took till gcc 9 before the standard library gained support for pmr and to_chars is - to the bestof my knowledge - still not implemented for floating point types even in g++-10/11.
One of motivations is to drop support for GCC 5, It just seem simpler to assume C++17 as requirement with comment that some compilers w/ partial C++17 support may still work.
The problem I see is: If you are pedantic, then there isn't a complete c++17 toolchain on linux yet.
GIL will not require feature-complete C++17 compiler We should be good with the last of the GCC 6 kind [1] or we may require GCC 7. The specifics will be decided as we go, as contributors will require. The deal is to announce C++11 phasing out early/now, to let ourselves time, at least two releases, but in practices it will be longer before we bump required compiler versions high.
So unless you are saying, all current version of gcc/clang/libstdc++ "may still work" but are not really supported, you need to be more specific in what you require and support. Where support means (to me) "Will be regularly tested and will continue to work for the next couple of releases/years".
Yes, we will aim to cover claimed support with CI and Boost regression tests matrix.
It may also be possible to list required C++17 features per Boost (GIL) release, but that list may change in any next release, obviously.
I wouldn't go there, because it isn't easy to actually verify that the documented features and the used features stay in sync
I made a mistake. Above I meant to write "Boost (GIL) release notes", where we can list GIL's required C++17 features for particular Boost release, just for user's convenience and change/remove that list from release notes in future. [1] https://gcc.gnu.org/legacy-ml/gcc/2018-10/msg00113.html Best regards, -- Mateusz Loskot, http://mateusz.loskot.net