perf: let EventedSet use clear() method of underlying set

[DanGonite57]

Motivated by napari/napari#6746, napari/napari#6895

The default clear method implemented by collections.abc.MutableSet is very slow, which I guess is due to the fact that it is not backed by the C implementation of the builtin set.

I have attempted to implement a custom interface for clear, modelled on the existing one for discard. However, I don't entirely understand the intentions for the various hooks, BAILs, etc, so please do let me know if anything is inconsistent with the behaviours and styles of other methods.

bench.py

from psygnal.containers import EventedSet

num = 100_000

my_set = EventedSet(range(num))
my_set.events.items_changed.connect(print)
my_set.clear()

Before:

> hyperfine --warmup 1 .\bench.py
Benchmark 1: .\bench.py
  Time (mean ± σ):      4.257 s ±  0.050 s    [User: 3.640 s, System: 0.047 s]
  Range (min … max):    4.192 s …  4.345 s    10 runs

After:

> hyperfine --warmup 1 .\bench.py
Benchmark 1: .\bench.py
  Time (mean ± σ):     236.3 ms ±   3.4 ms    [User: 59.9 ms, System: 31.0 ms]
  Range (min … max):   231.6 ms … 243.9 ms    12 runs

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (b0c2952) to head (c296570).

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #307   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           22        22           
  Lines         2071      2078    +7     
=========================================
+ Hits          2071      2078    +7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[codspeed-hq]

CodSpeed Performance Report

Merging #307 will improve performances by 11.78%

_{Comparing DanGonite57:perf-clear (c296570) with main (b0c2952)}

Summary

⚡ 1 improvements
✅ 65 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	DanGonite57:perf-clear	Change
⚡	test_emit_time[lambda-2]	127.8 µs	114.3 µs	+11.78%

[tlambert03]

wow. thanks this is fantastic. LGTM

Sorry about the weird BAIL stuff... had to remind myself why it's there (clearly not obvious), I think it was mostly there for things like add where subclasses might want the ability to cancel the event, or to avoid an event entirely as you've done in the case if len(self) == 0. but your PR is perfect, thanks!

[tlambert03]

btw, the codspeed benchmarks shown above don't capture this, but the asv benchmarks do and are awesome 🤩

Change	Before [b0c2952]	After [3b7af9d]	Ratio	Benchmark (Parameter)
-	1.15±0.01μs	689±7ns	0.6	benchmarks.EventedSetWithCallbackSuite.time_clear(10)
-	1.16±0.03μs	682±2ns	0.59	benchmarks.EventedSetSuite.time_clear(10)
-	4.83±0s	763±2ns	0	benchmarks.EventedSetSuite.time_clear(100000)
-	4.84±0.02s	794±7ns	0	benchmarks.EventedSetWithCallbackSuite.time_clear(100000)

"infinite speedup" 😂