State of support for windows<10; how to specify lack of support for a given package

[h-vetinari]

AFAIU we don't actually capture the windows version of the user anywhere in the package metadata; i.e. __win has no value, and conda info on windows yields (for example):

       virtual packages : __archspec=1=skylake
                          __conda=24.11.0=0
                          __win=0=0

It's not completely lost, as it appears in the user-agent string for example, but I'm not aware this this is used anywhere, or indeed that it could be used for package metadata at all.

               platform : win-64
             user-agent : conda/24.11.0 requests/2.32.3 CPython/3.11.11 Windows/10 Windows/10.0.22631 solver/libmamba conda-libmamba-solver/24.11.1 libmambapy/2.0.4
                                                                                ^^         ^^^^^^^^^^

The trigger for opening this issue is stumbling over the boost 1.87 release notes, which state:

Filesystem:

As was announced in 1.84.0, Windows versions prior to 10 are no longer supported.

Boost is not alone in starting to drop windows <10, and I suspect that our coverage of this is pretty bad, because the windows agents in CI are all obviously much newer already.

Our ucrt package describes itself as "Redistributable files for Windows SDK. This is only needed Windows <10", so presumably we should still be broadly compatible (also because windows API is very stable).

However, this raises two immediate questions:

• how can we avoid that boost 1.87 gets installed on win<10 systems?
• do we have an idea how wide-spread win<10 still is among our users, and do we want to consider dropping support at some point?

[xhochy]

how can we avoid that boost 1.87 gets installed on win<10 systems?

The only obvious solution coming to my mind would be to make the __win version the Windows version.

[jaimergp]

IIRC, mamba did export some __win version in their virtual packages. See mamba-org/mamba#2815.

We'll probably need a small CEP to sync with mamba and rattler (which doesn't seem to have any Windows logic, so also versionless?).

[h-vetinari]

We'll probably need a small CEP

To give an update here, @jaimergp valiantly dove down this "small" rabbit hole: conda/ceps#103

For the __win situation specifically, it gets populated correctly since:

• conda v25.1.0 (conda/conda@7d075cf; released 2025-01-17)
• mamba v2.0.0 (mamba-org/mamba@3275347; released 2024-09-25; c.f. also this issue)
• pixi v0.41.0 (prefix-dev/pixi@35291bd which vendors conda/rattler@b17de29; released 2025-02-05; c.f. issue)