STR: 1. Load http://cappel-nord.de/webaudio/acid-defender/ 2. Select "Jam Mode". 3. Click a bunch of squares and select all the Kick, Snare, and Hat sound buttons at the bottom. 4. Wait about 20 seconds. RESULT: The audio will be distorted, slow down, and eventually stop. This is a regression, but I'm having difficulty bisecting it. The problem is easy reproducible in Nightly 41, somewhat reproducible in 38–40, but not reproducible in ESR 31. Could this be a regression from OscillatorNode bug 1106649 in Firefox 37? This might be the same problem I reported in BiquadFilterNode bug 939491.

Pushlog: https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=b60d1a95bfa3&tochange=b8d51d20fd97 via local build last good: b60d1a95bfa3 first bad: ffab1d2bd4ce Triggered by: ffab1d2bd4ce Paul Adenot — Bug 1106649 - Use band-limited wave tables to implement basic waveforms. r=karlt

This should now be fixed in a Nightly build that contains the patches from bug 1173016.

I'll verify the fix when it lands in Nightly.

Paul, I can still reproduce this bug in Nightly 42 build 2015-07-10 on my MacBook Pro, though the distortion takes longer (~1 minute) to reproduce than before (~20 seconds).

Looking at what the app does (using the web audio editor in the devtools), this will be covered by karlt's work on collecting nodes more aggressively/not computing audio for nodes that don't have input. I seem to recall he's starting soon on that, or maybe he's already started, I can't remember the bug number.

And the drop in cpu usage after GC is further evidence.

This bug was fixed sometime between Firefox 42 and 45.