gh-74690: Micro-optimise typing._get_protocol_attrs

[AlexWaygood]

This PR is a bunch of micro-optimisations for typing._get_protocol_attrs that together have a significant impact on the speed of isinstance checks against runtime-checkable protocols.

Benchmark:

importtimefromtypingimportProtocol, runtime_checkable@runtime_checkableclassHasX(Protocol): x: intclassFoo: @propertydefx(self) ->int: return42classBar: x=42classBaz: def__init__(self): self.x=42classEgg: ... classNominal(HasX): def__init__(self): self.x=42classRegistered: ... HasX.register(Registered) num_instances=500_000foos= [Foo() for_inrange(num_instances)] bars= [Bar() for_inrange(num_instances)] bazzes= [Baz() for_inrange(num_instances)] basket= [Egg() for_inrange(num_instances)] nominals= [Nominal() for_inrange(num_instances)] registereds= [Registered() for_inrange(num_instances)] defbench(objs, title): start_time=time.perf_counter() forobjinobjs: isinstance(obj, HasX) elapsed=time.perf_counter() -start_timeprint(f"{title}: {elapsed:.2f}") bench(foos, "Time taken for objects with a property") bench(bars, "Time taken for objects with a classvar") bench(bazzes, "Time taken for objects with an instance var") bench(basket, "Time taken for objects with no var") bench(nominals, "Time taken for nominal subclass instances") bench(registereds, "Time taken for registered subclass instances")

Results on dfc4c95:

Time taken for objects with a property: 2.17 Time taken for objects with a classvar: 2.10 Time taken for objects with an instance var: 2.13 Time taken for objects with no var: 2.47 Time taken for nominal subclass instances: 2.20 Time taken for registered subclass instances: 4.72 

Results with this PR:

Time taken for objects with a property: 1.65 Time taken for objects with a classvar: 1.60 Time taken for objects with an instance var: 1.61 Time taken for objects with no var: 1.89 Time taken for nominal subclass instances: 1.61 Time taken for registered subclass instances: 3.79 

I'm once again skipping NEWS, since #103034 will counteract the boost to performance here, so the overall impact on performance for runtime-checkable protocols in 3.12 is still TBD.

• Issue: Performance of typing._ProtocolMeta._get_protocol_attrs and isinstance #74690

[AlexWaygood]

May once again be of interest to @leycec and @posita :)

[carljm]

Looks fine to me.

Given @hauntsaninja 's feedback in #103141 , are you planning to land a single news entry for this and that one and #103034 that summarizes the changes and the overall perf impact?

[hauntsaninja]

Hooray, thanks for working on this!

[AlexWaygood]

Given @hauntsaninja 's feedback in #103141 , are you planning to land a single news entry for this and that one and #103034 that summarizes the changes and the overall perf impact?

Yeah, I'll try to rework the NEWS entry in #103034 to summarise the changes made in the performance-related PRs as well as that PR 👍

[hauntsaninja]

Looks great!