[pulseaudio-discuss] [PATCH 2/4] alsa-ucm: Don't probe UCM

[Tanu Kaskinen]

On Thu, 2013-09-19 at 08:56 +0530, Arun Raghavan wrote:
> On Wed, 2013-09-18 at 21:26 +0200, David Henningsson wrote:
> > On 09/18/2013 07:41 PM, Felipe Tonello wrote:
> > > Hi David,
> > > 
> > > On Wed, Sep 18, 2013 at 7:01 AM, David Henningsson
> > > <david.henningsson at canonical.com> wrote:
> > >> In case of UCM, it usually targets a very specific hardware. The
> > >> person writing the UCM file for that hardware knows what (s)he is
> > >> doing, hopefully.
> > >>
> > >> Does it harm to probe all verbs at startup? Yes, because it can
> > >> cause longer startup time, clicks, or other side effects.
> > >>
> > >> (This is a difference to the normal probing mechanism, where we try
> > >> to make one profile set file fit many different sound cards, hence
> > >> we need probing there. But not here.)
> > >>
> > >> Signed-off-by: David Henningsson <david.henningsson at canonical.com>
> > >> ---
> > >>  src/modules/alsa/alsa-ucm.c |    4 +++-
> > >>  1 file changed, 3 insertions(+), 1 deletion(-)
> > >>
> > >> diff --git a/src/modules/alsa/alsa-ucm.c b/src/modules/alsa/alsa-ucm.c
> > >> index 81d0aeb..ab23259 100644
> > >> --- a/src/modules/alsa/alsa-ucm.c
> > >> +++ b/src/modules/alsa/alsa-ucm.c
> > >> @@ -1367,6 +1367,7 @@ static int ucm_create_profile(
> > >>      return 0;
> > >>  }
> > >>
> > >> +#if 0
> > >>  static snd_pcm_t* mapping_open_pcm(pa_alsa_ucm_config *ucm, pa_alsa_mapping *m, int mode) {
> > >>      snd_pcm_t* pcm;
> > >>      pa_sample_spec try_ss = ucm->core->default_sample_spec;
> > >> @@ -1514,6 +1515,7 @@ static void ucm_probe_profile_set(pa_alsa_ucm_config *ucm, pa_alsa_profile_set *
> > >>
> > >>      pa_alsa_profile_set_drop_unsupported(ps);
> > >>  }
> > >> +#endif
> > >>
> > >>  pa_alsa_profile_set* pa_alsa_ucm_add_profile_set(pa_alsa_ucm_config *ucm, pa_channel_map *default_channel_map) {
> > >>      pa_alsa_ucm_verb *verb;
> > >> @@ -1539,7 +1541,7 @@ pa_alsa_profile_set* pa_alsa_ucm_add_profile_set(pa_alsa_ucm_config *ucm, pa_cha
> > >>          ucm_create_profile(ucm, ps, verb, verb_name, verb_desc);
> > >>      }
> > >>
> > >> -    ucm_probe_profile_set(ucm, ps);
> > >> +/*    ucm_probe_profile_set(ucm, ps); */
> > >>      ps->probed = true;
> > >>
> > >>      return ps;
> > > 
> > > Why not just remove the code?
> > 
> > This is just the first draft. We might end up with some way of
> > configuring whether to probe or not (suggestions welcome).
> 
> Summarising IRC discussion:
> 
> * The lags, pops, clicks etc. are likely UCM bugs, and having probing
> off by default is likely to cause these to not be fixed.

If the bugs don't get fixed unless probing is enabled, I guess that
means that the bugs don't annoy people during normal operation. Hence,
I'm not sure they need fixing. If the only reason to probe is to find
bugs that manifest themselves only during probing, I don't understand
why anyone would care about such bugs, and why we should slow down the
startup for everyone just to find those bugs that nobody cares about.

> * I'm okay with an on-by-default "probe" modarg to disable probing in
> module-alsa-card

Can we have it off-by-default for UCM cards?

> * We were thinking about a special PA-specific Value to allow passing
> such parameters to PA. In this case, it would need to be at the card
> config level.

I don't like PA-specific stuff in UCM. If you really need to be able to
declare a UCM configuration as "bug-free", (i.e. pass probe=no to
module-alsa-card), there can be a non-PA-specific UCM Value for that.

Anyway, this patch is currently marked as "needing review", I suppose
this should be moved to the "waiting for new version" section.

-- 
Tanu