Patchbay connections disappear when changing quantum

[nyxkn]

When changing quantum (pipewire-jack), jack connections in the patchbay disappear (both in raysession and patchance), and we need to run a manual refresh of the canvas to see them again.

vokoscreenNG-2025-11-18_11-43-21.mp4

Other patchbays like cable and qjackctl seem to handle the quantum change without needing manual refresh.

[Houston4444]

Ok, quite strange ! I see that hydrogen ports have different names after buffer size change, probably Hydrogen renames them. Is it the same for connections which two ports are not renamed ?

Also, if you change this buffer size (quantum) from another program, is it the same ?

[AlexJakeGreen]

I also observe the similar issue with jack_mixer connections. Physical devices' connections do not disappear. I do not observe ports renamed. In my observations, the connections disappear only for the first change of buffer size:

Screencast.From.2026-01-09.02-15-08.mp4

If I change buffer size from another application, the behavior is the same: jack_mixer related connection dissappear

Screencast.From.2026-01-09.02-16-59.mp4

Why I switch the buffer: Switching the buffer size helps me to solve another unrelated to this issue problem, it is a very rare unknown and sporadical issue with pipewire when it starts to produce a lot of noise and cracking, and switching buffer is a fast workaround to get things back working.

[Houston4444]

Ok Thanks. I can reproduce it on a VM.

Starting raysession with raysession --dbg patch_engine, I can see that JACK send disconnections events but no connection event at buffersize change. I guess the only workaround solution would be to re-scan connections after buffersize change.

DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:midi in" 4294967435
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:midi out" 4294967436
DEBUG:patch_engine.patch_engine - ports connected False ('jack_mixer.mixer:Monitor L', 'Audio interne Stéréo analogique:playback_FL')
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:Monitor L" 4294967437
DEBUG:patch_engine.patch_engine - ports connected False ('jack_mixer.mixer:Monitor R', 'Audio interne Stéréo analogique:playback_FR')
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:Monitor R" 4294967438
DEBUG:patch_engine.patch_engine - ports connected False ('Audio interne Stéréo analogique:capture_FL', 'jack_mixer.mixer:Tradi L')
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:Tradi L" 4294967439
DEBUG:patch_engine.patch_engine - ports connected False ('Audio interne Stéréo analogique:capture_FR', 'jack_mixer.mixer:Tradi R')
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:Tradi R" 4294967440
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:Tradi Out L" 4294967441
DEBUG:patch_engine.patch_engine - port registration False "jack_mixer.mixer:Tradi Out R" 4294967442
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:midi in" 4294967435
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:midi out" 4294967436
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:Monitor L" 4294967437
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:Monitor R" 4294967438
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:Tradi L" 4294967439
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:Tradi R" 4294967440
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:Tradi Out L" 4294967441
DEBUG:patch_engine.patch_engine - port registration True "jack_mixer.mixer:Tradi Out R" 4294967442

[Houston4444]

Ok. The bug should be fixed now. Feedback welcome.

[AlexJakeGreen]

In my opinion, a quantum change behaves like a global reset of the graph, therefore a full rescan is required. This may not be merely a workaround, but the proper fix

[Houston4444]

In my opinion, a quantum change behaves like a global reset of the graph, therefore a full rescan is required. This may not be merely a workaround, but the proper fix

I tried to fix it this way, but there were problems (ports added by PipeWire just after the re-scan. I finally think it is better this way. BTW, I think I have to report an issue to PipeWire, this behavior is strange, and the Hydrogen port renaming is even stranger (I can reproduce it).