On 25. 06. 22 13:28, Takashi Sakamoto wrote:
Hi Jaroslav,
Recent years I'm bothered about unexpected abort of pulseaudio and pipewire processes when testing user-defined control element set. They aborts at element removal event.
pulseaudio: mixer.c:149: hctl_elem_event_handler: Assertion `bag_empty(bag)' failed. wireplumber: mixer.c:149: hctl_elem_event_handler: Assertion `bag_empty(bag)' failed.Would I ask your opinion about the design of alsa-lib mixer API?
As long as I investigate, these programs seem to have careless coding as alsa-lib mixer API application. Both of them attaches an instance of snd_mixer_elem_t to an instance of snd_hctl_elem_t by calling snd_mixer_elem_attach() when detecting element addition, but never detach it even if detecting element removal.
In the case, the link list (=bag) of mixer API internal never becomes empty. It has link entries as much as the number of registered mixer classes which attaches snd_mixer_elem_t. Then it hits the assertion.
I think the design of alsa-lib mixer API demands mixer class implementation to detach at element removal which attached at element addition. But I have less conviction about it since enough unfamiliar.
Yes, if the REMOVE event is delivered to the mixer class, the reference to the associated hctl element should be removed. The assert does the check for this consistency.
I'm glad if receiving your opinion about it.
I wrote test program for the issued behaviour:
For test, please use sample script in alsa-gobject project:
The patches for pulseaudio/pipewire are prepared:
Your fixes seem correct. Please, create the PA/PW merge request with this code. Please add me (@perexg) to your merge message.
Thank you for your work on this.
Jaroslav