Conversation
jfsantos
pushed a commit
to tone-3000/tone3000-plugin
that referenced
this pull request
Aug 13, 2026
Three crashes reproduced live (gdb + ASan) on every device/channel/buffer change on a Focusrite Scarlett Solo, all upstream JUCE 9.0.1 bugs with open, unmerged fixes already filed (same author, all dated the day before this investigation started): - ALSAThread::close() waits a fixed 400ms for the audio thread to exit before force-closing the ALSA handles (closeNow()) out from under it. A large buffer at a low sample rate needs longer than 400ms for one i/o period, so a thread that's merely mid-read gets torn down as if it were stuck, use-after-freeing whatever it was reading from. Confirmed live: closeNow() firing with numCallbacks == callbacksToStop (no callback landed during the wait - not actually stuck). juce-framework/JUCE#1707 - ALSAThread::open() builds the active-channel bitset and the callback's channel-pointer array in one loop over the requested mask, then widens the bitset alone up to the hardware's channel-count floor afterward - so on a device fixed at 2 channels, a mono request leaves the bitset claiming 2 active channels with only 1 real pointer behind it. juce-framework/JUCE#1706 - AudioDeviceManager's CallbackMaxSizeEnforcer sizes its stored channel vectors from that (now too-wide) bitset, fills in only the channels the callback actually has, and forwards the full vector size regardless - so the unfilled tail is a null pointer that gets dereferenced downstream (AudioProcessorPlayer, LevelMeter::updateLevel). This is the direct segfault: confirmed live that JUCE's own jassert in this exact spot fires the moment a channel-mask change lands. juce-framework/JUCE#1708 All three applied as idempotent configure-time source patches, matching the existing pattern in this file. Drop each once its PR lands upstream. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Merged
4 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Opening an ALSA capture device whose minimum channel count is higher than the number of input channels selected crashes on the audio thread, in the input level meter. To reproduce, open a capture device whose ALSA minimum is 2 channels (a dsnoop device in my case) with only the first input channel enabled.
getActiveInputChannels() ends up reporting more channels than the audio callback is actually handed pointers for. open() fills inputChannelDataForCallback and currentInputChans together in one loop, and then widens currentInputChans to minChansIn afterwards, so the bitset gains a bit with no matching entry in the channel array. AudioDeviceManager sizes CallbackMaxSizeEnforcer's stored channel vectors from getActiveInputChannels(), fills in only the channels the device passes, and forwards the rest as null pointers, which is what then crashes in the meter.
This moves the widening up so that it applies to inputChannels before the loop runs, which is how the output path a few lines below already does it, and the two stay in agreement. A request for no input channels at all is left alone rather than being widened to the minimum.