EOL-ing Python 3.7-3.9

[nir0s]

At some point, we agreed that we won't be supporting Python versions that are EOL. Python 3.7-3.9 are EOL, so we should consider whether we want to remove official support for them or not.

[nir0s]

Is there a code-quality reason to drop 3.7–3.9?

Short answer: there is almost nothing in 3.10 specifically that would improve this codebase. The real, tangible cleanups live at the 3.7 → 3.8 boundary, and they're unlocked just by dropping 3.7. Going all the way to a 3.10 floor buys very little we can't already have today.

What dropping each version actually buys us

Dropping 3.7 (the only high-value one for the code): We get to delete two genuine compatibility shims in distro.py right now:

• The cached_property backport class (~18 lines) — functools.cached_property is 3.8+.
• The TypedDict import fallback (TypedDict = dict) — typing.TypedDict is 3.8+.

These are the only two version-gated shims in the entire module, and both target 3.7. That's the single biggest win, and it costs us nothing past 3.7.

Dropping through to 3.9: Enables PEP 585 at runtime — Dict[str, str] → dict[str, str], List/Tuple → list/tuple, shrinking the typing import block. Purely cosmetic.

Dropping through to 3.10: Enables PEP 604 (Optional[bool] → bool | None, removing Optional) and the match statement. But:

• match has no natural fit here — the id()/version() lookups are "first non-empty in priority order" chains and dict-table normalizations, not structural pattern matches. Forcing match in would be worse, not better.
• The bool | None syntax win is real but small.

The catch that undercuts the 3.10 case

We can get all the syntax modernization (X | None, list[str], etc.) today, on 3.7, by adding from __future__ import annotations. Every annotation in the module is used only in signatures and the two TypedDict class bodies — none are subscripted at runtime — so making them lazy strings lets us write the modern syntax with no version bump.

What from __future__ import annotations cannot fix are the two runtime shims (cached_property, the TypedDict import), because those execute. So those genuinely require a 3.8 floor — which lands us right back at "dropping 3.7 is the only change that pays for itself in code."

Recommendation

Drop them, but for the right reason and not too aggressively:

1. Set the floor to 3.9. It removes 3.7 (deletes both shims) plus the special-cased ubuntu-22.04 CI runners that exist only to host 3.7/3.8. Justify it as EOL hygiene + CI cost, not feature payoff — that's the honest framing.
2. Don't push to 3.10 for the code's sake — there's no 3.10 feature this library wants. If we ever do, it's a packaging decision, not a code-quality one.
3. Weigh our user base before going fast. distro is a foundational, very-widely-vendored dependency, and its whole job is identifying old/enterprise/embedded systems — exactly the environments most likely to be stuck on old Python. A conservative floor has real value here that a typical app wouldn't have.