#277 pulseaudio sound handler is buggy/inconsistent

[Stefano Guidoni]

There is a bug in the SoX pulseaudio handler, that leads to an inconsistent behaviour of SoX: with ALSA, OSS etc. when a write of zero length is performed nothing happens, with PulseAudio instead SoX quits when a zero length write is performed. This is because pa_simple_write() returns an error whenever the input buffer length is zero.

As a result, example3.c from the SoX source code fails if one tries to substitute alsa with pulseaudio (tested with an ogg/vorbis file). The solution is to modify write_samples from pulseaudio.c so that it behaviours exactly like ALSA, i.e. it returns 0 when 0 is the length of the write. This can be achieved by simply adding something like if (nsamp == 0) return 0; before the call to pa_simple_write(). As far as I tested it there are not downturns to this approach.

NOTE: if one tries to perform a zero length write (i.e. len is zero) with the ALSA handler, the for cycle never executes its code, because the condition is (done = 0; done < len; done += n), so it reaches the end of the function where the line return len; is executed, i.e. it just returns zero when len is zero.

[Stefano Guidoni]

A patch that solves the problem:

--- old/src/pulseaudio.c        2015-02-27 04:48:40.000000000 +0100
+++ new/src/pulseaudio.c        2016-07-08 20:33:45.608428058 +0200
@@ -126,6 +126,9 @@
   size_t len;
   int rc, error;


+  if (nsamp == 0)
+    return 0;
+  
   /* Pulse Audio buffer lengths are true buffer lengths and not
    * count of samples. */
   len = nsamp * sizeof(sox_sample_t);

[Mans Rullgard]

Fixed.