On Mon, May 16, 2022 at 6:58 PM Michael Paquier <michael@paquier.xyz> wrote:
> Cutting support for now-unsupported versions of Windows is in the air
> for a couple of months, and while looking at the code a first cleanup
> that looked rather obvious to me is the removal of support for VS
> 2013, as of something to do for v16~.
Not a Windows person so I couldn't comment on the details without more
research, but this general concept seems good to me. That's a nice
reduction in (practically) untestable/dead code (no CI, no build
farm).
For comparison, I picked 3 random C/C++ (OK, C++) projects I could
think of to see how they deal with VS versions, and all require 2017+:
* MariaDB supports the last two major versions, so currently VS 2019
and VS 2022, 2022 is preferred[1]
* Chrome requires VS 2017+ but currently 2019 is preferred[2]
* OpenJDK requires VS 2017+[3]
Looking at the published lifecycle info, 2017 is the oldest still in
'mainstream' support[4], so it wouldn't be too crazy to drop VS 2015
too, just like those other projects. That said, it sounds like there
is no practical benefit to being more aggressive than you are
suggesting currently (as in, we wouldn't get to delete any more crufty
untestable dead code by dropping 2015, right?), so maybe that'd be
enough for now.
[1] https://mariadb.com/kb/en/Building_MariaDB_on_Windows/
[2] https://chromium.googlesource.com/chromium/src/+/main/docs/windows_build_instructions.md#Visual-Studio
[3] https://openjdk.java.net/groups/build/doc/building.html
[4] https://docs.microsoft.com/en-us/visualstudio/productinfo/vs-servicing
