[alsa-devel] How exactly snd_pcm_hw_rule_add() should to be used?

[Takashi Iwai]

At Tue, 5 Feb 2013 11:24:45 +0100,
Peter Ujfalusi wrote:
> 
> Hi,
> 
> I got confused (again) regarding to how to use correctly the
> snd_pcm_hw_rule_add(). Looking at the documentation [1] and to the code in
> kernel seams to be suggesting the opposite.
> 
> [A]
> using the 5.6. example from [1]:
> snd_pcm_hw_rule_add(substream->runtime, 0, SNDRV_PCM_HW_PARAM_CHANNELS,
> 		    hw_rule_channels_by_format, 0, SNDRV_PCM_HW_PARAM_FORMAT,
> 		    -1);
> 
> The [1] tells me that the SNDRV_PCM_HW_PARAM_FORMAT will be refined based on
> the SNDRV_PCM_HW_PARAM_CHANNELS.
> 
> In 5.7. of [1] the reverse rule is placed:
> snd_pcm_hw_rule_add(substream->runtime, 0, SNDRV_PCM_HW_PARAM_FORMAT,
> 		    hw_rule_channels_by_format, 0,
> 		    SNDRV_PCM_HW_PARAM_CHANNELS, -1);

It's:
  snd_pcm_hw_rule_add(substream->runtime, 0, SNDRV_PCM_HW_PARAM_FORMAT,
                      hw_rule_format_by_channels, 0,
		       SNDRV_PCM_HW_PARAM_CHANNELS -1);

> Which according to the example will refine the SNDRV_PCM_HW_PARAM_CHANNELS
> based on the SNDRV_PCM_HW_PARAM_FORMAT.

The definitions of hw_rule_channels_by_format and
hw_rule_format_by_channels are placed wrongly.  And the description
isn't clear, which function is for what, indeed.  So, overall it's
confusing, yes.

But the code written there is correct, at least :)

> However when looking at the code in the kernel the opposite seams to be what
> is actually happening.
> 
> [B]
> using the same 5.6. example from [1]:
> snd_pcm_hw_rule_add(substream->runtime, 0, SNDRV_PCM_HW_PARAM_CHANNELS,
> 		    hw_rule_channels_by_format, 0, SNDRV_PCM_HW_PARAM_FORMAT,
> 		    -1);
> 
> It looks like the SNDRV_PCM_HW_PARAM_CHANNELS should be refined based on the
> SNDRV_PCM_HW_PARAM_FORMAT.

This is correct.

The third argument of snd_pcm_hw_rule_add() is the parameter to be
refined, following the function to call, the private data pointer, and
the dependent parameters.  In the case above,

I'll try to improve the text (and correct the definitions).


thanks,

Takashi