Hi. SUBJ.
* no more optional-ness and possible troubles from aligned or sized allocation/deallocation operators. Since c++17 has them both.
* c++17 is modern enough for us to be able to use some nice helper codes from abseil or similar libraries (one I am thinking of right now is absl::FunctionRef) as well as generally keeping pace with "modern" c++
In terms of "enterprise" Linux distros it cuts Ubuntu LTS releases 14.04 (gcc 4.8), 16.04 (gcc 5) and potentially 18.04 (gcc 7).
https://wiki.ubuntu.com/Releases states they still all have some support from Ubuntu. Users on those distros (if any; those *are* old releases), will have to find a way to get an upgraded compiler.
The case of RHEL and original Centos is covered via RedHat dev toolset stuff (my kudos to Red Hat folks doing it; super handy). I am semi-regularly testing on as far as rhel 6 and at least semi-modern compilers are available there.
I don't have data on SLES, but hopefully SuSE folks do similar or even better job than rhat folks.
Most recent Debian stable is easily good enough too. Up-to-date OSX should be covered easily. In terms of Visual Studio it'll be an upgrade of the minimal version to VS2019.
Various BSDes, well, I do know for certain that their most recent versions are definitely good enough. QNX should be covered (they ship a version of gcc 8 with their sdk).
Solaris/AIX and possibly other less commonly used cases, I have no idea (I recently tested most recent opensolaris and it fails already; I may or may not find time to get this working). But recent opensolaris at least ships up-to-date gcc. I don't have resources to test Oracle's custom compiler.
In terms of less common C++ compilers (I already mentioned Oracle's above), we're seemingly going to lose some.
https://en.wikipedia.org/wiki/List_of_compilers#C++_compilers But do note that the wikipedia text explicitly states that list might be stale. Plus I do have the impression the industry appears to be consolidating towards clang.
Any concerns/comments/questions ?