use __slots__ where possible

[albertedwardson]

Summary

this breaks weakrefs, but I believe the objects in question shouldn't be weakref`ed anyway

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.

[codspeed-hq]

Merging this PR will not alter performance

✅ 24 untouched benchmarks

---

_{Comparing albertedwardson:slots (91da498) with main (02bed6f)}

[hrv-dys]

The __slots__ additions make sense for the protocol classes — these are instantiated per-connection, so reducing per-instance memory overhead adds up under load.

A few things:

1. ServerState as a @dataclass(slots=True) — clean conversion. The field(default_factory=...) for the mutable defaults (set, list) is correct and avoids the classic mutable-default-argument footgun.

2. FlowControl as a dataclass — the __post_init__ to call self._is_writable_event.set() is a nice touch. However, I notice repr=False, eq=False on both dataclasses — is that to avoid accidentally comparing or printing these in hot paths? Might be worth a brief comment for future readers.

3. The limit_max_requests change from @functools.cached_property to __init__ assignment — this is necessary because __slots__ and cached_property don't play well together (cached_property needs __dict__). The rename to _limit_max_requests() as a regular method called in __init__ is a reasonable workaround. Just want to flag: this changes the laziness — it's now computed eagerly at construction time. In practice this shouldn't matter since Server.__init__ is called once, but it's a behavioral change worth noting.

4. H11Protocol and HttpToolsProtocol slots lists — I spot-checked these against the __init__ methods and they look complete. If a future PR adds an attribute and forgets to update __slots__, it'll silently fall through to a per-instance __dict__ (unless I'm wrong and asyncio.Protocol doesn't define __dict__?). Actually, since the base asyncio.Protocol doesn't use __slots__, subclasses still get __dict__. So the slots here act as a performance hint rather than a strict contract. That's fine, just worth being aware of.

Tests pass with only pre-existing failures (websockets deprecation warnings unrelated to this PR).

[albertedwardson]

@hrv-dys unfortunately llm didn't catch that my changes with @dataclass changes the __init__ signature

[albertedwardson]

init=False or don't use dataclasses at all

[albertedwardson]

can be closed in favour of #2697 that already introduces slots

or maybe just deffered until the mentioned one will be merged

[Kludex]

@albertedwardson I think if we can find meaningful proof of performance improvement then it's worth it, otherwise I would prefer to close it for the time being. Thanks for the push tho.