m-m-adams
commented
2 weeks ago When the clock falls behind it attempts to catch up by sending two clocks at once, resulting in inconsistent midi clock and no gate output.
The clock falls behind when two ticks need to be output in the same render window. As the nightly build is targeting larger consistent window sizes this now occurs more frequently, however it is present in official and the RC as well (#2132 and unstable midi clock reports on discord).
The fix will likely be to proactively schedule ticks further in advance
Please describe the problem:
Alternate title: Trigger Clock Output Fails to Meet Realtime Deadline
When using the analog clock at the gate output, it will randomly stop transmitting for long periods.
On 8aa506f this happens after a few seconds. According to @m-m-adams it happens after a few minutes on Release Candidate 3 564cba31cd14d9fbfc16a28114ab6d4dba9441e6.
What is the expected behavior?
The trigger clock putters along more-or-less constantly, within a margin of error that most humans (and downstream sequencers) consider indistinguishable from lock-step.
Is there a relevant Pull request?
No response
What hardware did you reproduce it with?
OLED
What firmware did you reproduce it with?
Nightly
What is the firmware name:
8aa506f
If possible provide the steps to reproduce the issue and upload additional media:
Additional Info
Tested with multiple downstream devices, but it is so severe that it is visible with the naked eye.
Watching a blinking LED connected to the output is sufficient to detect the dropouts at higher PPQNs, it is that major! The LED (or any downstream sequencer/clock divider) will receive no signal at all for up to nearly a second(?!) before the Deluge resumes transmitting, leaving the downstream devices hopelessly out of sync.
The improvements between the latest nightlies and the current community branch is likely due to improvements in the yield/scheduler code that have gone in, but it remains. I'm not sure what commit introduced the issue.
NOTE: This is different to the other analog clock output bug that I reported.
UPDATE 2024-06-19
Some attempts at code fixes have improved the situation, so that the dropouts occur less often, sometimes only after dozens of minutes (depending on the BPM). But the issue persists and the underlying cause has not yet been identified. What is suspected is that the flag which tells the Deluge when another clock pulse is needed gets double set, resulting in triggers being delayed. @m-m-adams noticed that something similar happens in MIDI, where the number of ticks per minute is correct but get delayed and bunched up. This may be the same behaviour happening with gates, but the gate circuitry isn't fast enough to send all of those pulses so close together so they all get blended together, resulting in the clock not being transmitted in discrete pulses but instead as a single gate.