proposal: pa_linux_alsa.c: do we want to pass `dir` (direction) param to `alsa_snd_pcm_hw_params_set_period_size_near`?

[RossBencina]

As part of our review of #524 we notice that GropeDevice is passing NULL for the alsa_snd_pcm_hw_params_set_period_size_near direction parameter (lower/higher/match). We're not sure about that. Was changed from uninitialized var to NULL in #268

Possibly we should change it to pass the address of an int set to zero (see patch below) -- rationale being to match what we do when opening the stream. However it is unclear whether this would give the same behavior. The ALSA docs don't mention passing NULL: https://www.alsa-project.org/alsa-doc/alsa-lib/pcm_2pcm_8c.html#a9162045265f283c532634506456cab09 however the test program linked by the ALSA docs does pass NULL (written as 0): https://www.alsa-project.org/alsa-doc/alsa-lib/_2test_2latency_8c-example.html#a14

Note that alsa-lib seems to treat passing NULL the same as passing &int where int dir=0;

• https://github.com/alsa-project/alsa-lib/blob/62c24074ae2a39613b27028cde0bac2a20635ea1/src/pcm/pcm.c#L5494
• https://github.com/alsa-project/alsa-lib/blob/62c24074ae2a39613b27028cde0bac2a20635ea1/src/pcm/pcm_params.c#L822

F:\PortAudio\portaudio>git diff
diff --git a/src/hostapi/alsa/pa_linux_alsa.c b/src/hostapi/alsa/pa_linux_alsa.c
index d642ba0..e23a423 100644
--- a/src/hostapi/alsa/pa_linux_alsa.c
+++ b/src/hostapi/alsa/pa_linux_alsa.c
@@ -850,6 +850,7 @@ static PaError GropeDevice( snd_pcm_t* pcm, int isPlug, StreamDirection mode, in
     PaError result = paNoError;
     snd_pcm_hw_params_t *hwParams;
     snd_pcm_uframes_t alsaBufferFrames, alsaPeriodFrames;
+    int nearDir = 0;
     unsigned int minChans = 0;
     unsigned int maxChans = 0;
     int* minChannels, * maxChannels;
@@ -960,7 +961,8 @@ static PaError GropeDevice( snd_pcm_t* pcm, int isPlug, StreamDirection mode, in
     alsaBufferFrames = 512;
     alsaPeriodFrames = 128;
     ENSURE_( alsa_snd_pcm_hw_params_set_buffer_size_near( pcm, hwParams, &alsaBufferFrames ), paUnanticipatedHostError );
-    ENSURE_( alsa_snd_pcm_hw_params_set_period_size_near( pcm, hwParams, &alsaPeriodFrames, NULL ), paUnanticipatedHostError );    
+    nearDir = 0;
+    ENSURE_( alsa_snd_pcm_hw_params_set_period_size_near( pcm, hwParams, &alsaPeriodFrames, &nearDir ), paUnanticipatedHostError );
     *defaultLowLatency = (double) (alsaBufferFrames - alsaPeriodFrames) / defaultSr;

     /* Base the high latency case on values four times larger */
@@ -971,7 +973,8 @@ static PaError GropeDevice( snd_pcm_t* pcm, int isPlug, StreamDirection mode, in
     ENSURE_( SetApproximateSampleRate( pcm, hwParams, &approximateSampleRate ), paUnanticipatedHostError );
     defaultSr = (double) approximateSampleRate;
     ENSURE_( alsa_snd_pcm_hw_params_set_buffer_size_near( pcm, hwParams, &alsaBufferFrames ), paUnanticipatedHostError );
-    ENSURE_( alsa_snd_pcm_hw_params_set_period_size_near( pcm, hwParams, &alsaPeriodFrames, NULL ), paUnanticipatedHostError );    
+    nearDir = 0;
+    ENSURE_( alsa_snd_pcm_hw_params_set_period_size_near( pcm, hwParams, &alsaPeriodFrames, &nearDir ), paUnanticipatedHostError );
     *defaultHighLatency = (double) (alsaBufferFrames - alsaPeriodFrames) / defaultSr;

     *minChannels = (int)minChans;

[philburk]

Experiments on Raspberry Pi 4.

In pa_linux_alsa.c GropeDevice, set initial alsaBufferFrames to 0.
Groping for a USB device led to alsaBufferFrames=4 and alsaPeriodFrames=2.
Both functions returned 0.

initial alsaBufferFrames	final alsaBufferFrames	alsaPeriodFrames
0	4	2
8	8	2
16	16	4
512	512	128
1024	1024	256
2048	2048	512
4096	4096	1024
8192	8192	2048
16384	16384	4096

[RossBencina]

Maybe rename GropeDevice to ProbeDeviceDefaults