Normalize platform_system to sys_platform

[charliermarsh]

Summary

A revival of an old idea (#9344) that I have slightly more confidence in now. I abandoned this idea because (1) it couldn't capture that, e.g., platform_system == 'Windows' and sys_platform == 'foo' (or some other unknown value) are disjoint, and (2) I thought that Android returned "android" for one of sys_platform or platform_system, which would've made this logic incorrect.

However, it looks like Android... doesn't do that? And the values here are almost always in a small, known set. So in the end, the tradeoffs here actually seem pretty good.

Vis-a-vis our current solution, this can (e.g.) simplify out expressions like sys_platform == 'win32' or platform_system == 'Windows'.

[zanieb]

nice!

[zanieb]

I'm into this. Some churn for outputs though... I wonder if we should hold off on this until another minor version? (though these are less meaningful than some of the other resolution changes we've been making...)

[charliermarsh]

I'm somewhat in favor of just shipping it, since it won't invalidate existing lockfiles.

[samypr100]

I thought that Android returned "android"

Note, per PEP-738, it will return "android" starting in 3.13. Similar as "ios" in 3.13 for iPhones per PEP-730

[charliermarsh]

Thank you! Luckily it looks like it will also return "Android" for platform.system(), phew.

[konstin]

I would feel more confident if we could codify this upstream (packaging.python.org probably), so uv doesn't end up making assumptions that python redistributors don't make.

[charliermarsh]

Can you open a convo about it?

[BurntSushi]

Also, do we have any context on why there is platform_system and sys_platform?

[BurntSushi]

And maybe it's obvious, but I think a comment saying why we normalize to sys_platform (instead of the other way around) would be good too.

[zanieb]

https://discuss.python.org/t/clarify-usage-of-platform-system/70900 from Konsti's prompt in the PyPA Discord

[BurntSushi]

Nice!

[BurntSushi]

Also, do we have any context on why there is platform_system and sys_platform?

[BurntSushi]

And maybe it's obvious, but I think a comment saying why we normalize to sys_platform (instead of the other way around) would be good too.

[BurntSushi]

(Meant to approve.)

[charliermarsh]

From the PyPA Discord, it sounds like this isn't really safe to assume in all cases:

I gave the example of iPad, but looking at the docs, but some Unix OSes will have the major version number appended in sys.platform... in fact, that should be the case for FreeBSD, NetBSD, OpenBSD, and SunOS in your example

It might be fine for macOS, Windows, and Linux though...? The downside is that the fewer values we rewrite, the more we risk messing up when we see a marker like platform_system == "Darwin" and platform_system == "FreeBSD", since we'd then rewrite that to sys_platform == "darwin" and platform_system == "FreeBSD", and no longer realize they're disjoint.

[charliermarsh]

Maybe we run with this for macOS, Linux, Windows, and any other platforms that don't encode a version number.

Then, we can use our existing "incompatible markers" infrastructure to mark those other platforms as incompatible with these sys_platform values? Like, we can mark platform_system == "FreeBSD" as incompatible with all of these sys_platform values.

[charliermarsh]

Ok, I did a second pass on this with more care around known semantics. I think it's good to go. Our strategy has two parts: (1) we normalize platform_system to sys_platform when we can; and (2) when we can't, we encode as many known incompatibilities as possible. (1) is preferable to (2), but (1) isn't always possible.