[vedchaudhari]

util.debounce() and util.throttle() add an abort listener to options.signal when they're created and never remove it, so a long-lived signal keeps every function created with it in memory:

const ac = new AbortController();
for (let i = 0; i < 5; i++) util.debounce(() => {}, 100, { signal: ac.signal });
events.getEventListeners(ac.signal, 'abort').length;
// before: 5
// after:  0

This only keeps the listener while calls are pending, and removes it with the Symbol.dispose method that addAbortListener() returns. Calls after an abort still reject, since every call already checks signal.aborted.

All functions share one long-lived AbortSignal:

Debounced functions created	Heap after GC on main	Heap after GC with this PR
10,000	29 MB	4 MB
30,000	78 MB	4 MB
50,000	127 MB	4 MB

Normal calls take the same time as before. Creating a debounced function with a signal is about 20% faster, and abort() with 5,000 idle functions takes 0.2 ms instead of 9.4 ms. The only extra cost is under 1 µs when a call starts waiting, to add and remove the listener.

Performance numbers

Median of 9 runs, old and new code on the same build, Windows x64:

Scenario	main	This PR
Calls within one debounce window	110 ns	115 ns
Throttle calls that run immediately	368 ns	356 ns
Creating a debounced function with a signal	5.1 µs	4.1 µs
abort() with 5,000 idle functions	9.4 ms	0.2 ms
Worst case: debounce call + flush() every time, with a signal	457 ns	1,147 ns
Worst case: throttle queues one call every time, with a signal	416 ns	1,049 ns

Added tests; they fail on main and pass with this change.

Refs: #65899

Assisted-by: Claude Code. I reproduced the leak on main, reviewed every change, and ran the debounce/throttle tests and lint myself.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.49%. Comparing base (f88a1d2) to head (9d4466e).
⚠️ Report is 20 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff            @@
##             main   #66839    +/-   ##
========================================
  Coverage   90.49%   90.49%            
========================================
  Files         797      797            
  Lines      277780   278249   +469     
  Branches    53360    53465   +105     
========================================
+ Hits       251388   251813   +425     
- Misses      16791    16806    +15     
- Partials     9601     9630    +29     

Files with missing lines	Coverage Δ
lib/internal/util/debounce.js	97.61% <100.00%> (+0.15%)	⬆️
lib/internal/util/throttle.js	94.27% <100.00%> (+0.21%)	⬆️

... and 55 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[bakkot]

The addAbortListener helper returns an object with a Symbol.dispose method that removes the listener; might be simpler to just use that?

[vedchaudhari]

The addAbortListener helper returns an object with a Symbol.dispose method that removes the listener; might be simpler to just use that?

Thanks, good point. Switched to addAbortListener() and its Symbol.dispose. I had used addEventListener and removeEventListener directly for a small speedup but it is not worth the extra code. Also updated the numbers in the description.

[vedchaudhari]

@cjihrig would you have a moment to take a look? It removes the abort listener that debounce/throttle leave behind on the signal, so it no longer leaks. Thanks!