perf: Eliminate recursive calls - Improved by approximately **17%**

[medz]

Recursive calls vs. No recursion Benchmark Comparison Report

Benchmark Comparison Table:

Test Category	Test Case	Recursive calls (μs)	No recursion (μs)	Change (μs)	Change (%)	Performance Change
Propagation	avoidablePropagation	104272	80050	-24222	-23.2%	Significant Improvement (Large Decrease)
	broadPropagation	183044	184638	1594	0.9%	Little Change (Essentially Flat)
	deepPropagation	77191	54170	-23021	-29.8%	Significant Improvement (Large Decrease)
	diamond	126867	107900	-18967	-15.0%	Improvement (Noticeable Decrease)
	mux	197853	194765	-3088	-1.6%	Little Change (Slight Decrease)
	triangle	53633	52221	-1412	-2.6%	Slight Improvement (Slight Decrease)
Observers	repeatedObservers	14215	12850	-1365	-9.6%	Slight Improvement (Slight Decrease)
Large-Scale	molBench	325145	330550	5405	1.7%	Slight Regression (Slight Increase)
Components	comp_0to1	979	873	-106	-10.8%	Improvement (Noticeable Decrease)
	comp_1to1	2883	1654	-1229	-42.6%	Significant Improvement (Large Decrease)
	comp_2to1	1794	1412	-382	-21.3%	Improvement (Noticeable Decrease)
	comp_4to1	1916	1587	-329	-17.2%	Improvement (Noticeable Decrease)
	comp_1000to1	13	6	-7	-53.8%	Significant Improvement (Large Decrease)
	comp_1to2	1505	1253	-252	-16.7%	Improvement (Noticeable Decrease)
	comp_1to4	1503	3079	1576	104.9%	Significant Regression (Large Increase)
	comp_1to8	1558	2507	949	60.9%	Significant Regression (Large Increase)
	comp_1to1000	429	460	31	7.2%	Slight Regression (Slight Increase)
Updates	update_1to1	3547	2671	-876	-24.7%	Improvement (Noticeable Decrease)
	update_2to1	1425	1511	86	6.0%	Slight Regression (Slight Increase)
	update_4to1	671	686	15	2.2%	Slight Regression (Slight Increase)
	update_1000to1	11	10	-1	-9.1%	Slight Improvement (Slight Decrease)
	update_1to2	1535	1528	-7	-0.5%	Little Change (Essentially Flat)
	update_1to4	695	690	-5	-0.7%	Little Change (Essentially Flat)
	update_1to1000	23	23	0	0.0%	No Change
Cellular	cellx1000	4163	4383	220	5.3%	Slight Regression (Slight Increase)
	cellx2500	10207	11839	1632	16.0%	Noticeable Regression (Noticeable Increase)
	cellx5000	27705	26234	-1471	-5.3%	Slight Improvement (Slight Decrease)
Grid	10x5 - 2 sources - read 20.0% (simple)	204985	203574	-1411	-0.7%	Little Change (Essentially Flat)
	10x10 - 6 sources - dynamic - read 20.0% (dynamic)	120424	116255	-4169	-3.5%	Slight Improvement (Slight Decrease)
	1000x12 - 4 sources - dynamic (large)	229409	222962	-6447	-2.8%	Slight Improvement (Slight Decrease)
	1000x5 - 25 sources (wide dense)	303129	303695	566	0.2%	Little Change (Essentially Flat)
	5x500 - 3 sources (deep)	124582	119665	-4917	-4.0%	Slight Improvement (Slight Decrease)
	100x15 - 6 sources - dynamic (very dynamic)	168315	166657	-1658	-1.0%	Little Change (Essentially Flat)

• Simple averages show that the average improvement for the improved test cases is about 17.2%.
• A conservative estimate is that the overall average performance improvement for "No recursion" is between 10% and 17%.

[johnsoncodehk]

It may also be worth to try keep checkDirty simple and only eliminate recursive calls for propagate. Because only propagate can obtains an additional fast path by eliminating recursive calls, see stackblitz/alien-signals#7 (comment).

[medz]

It may also be worth to try keep checkDirty simple and only eliminate recursive calls for propagate. Because only propagate can obtains an additional fast path by eliminating recursive calls, see stackblitz/alien-signals#7 (comment).

OK，I will re-experiment.