Update audio code to use size_t more correctly, webrtc/common_audio/resampler/

webrtc/common_audio/resampler/sinusoidal_linear_chirp_source.h

Index: webrtc/common_audio/resampler/sinusoidal_linear_chirp_source.h
diff --git a/webrtc/common_audio/resampler/sinusoidal_linear_chirp_source.h b/webrtc/common_audio/resampler/sinusoidal_linear_chirp_source.h
index 1bddfc907472c86dc7ac2659ae4e563bf14835ed..eff36db43fdc66033af0e38fa886afb8c5a552f0 100644
--- a/webrtc/common_audio/resampler/sinusoidal_linear_chirp_source.h
+++ b/webrtc/common_audio/resampler/sinusoidal_linear_chirp_source.h
@@ -24,16 +24,16 @@ namespace webrtc {
 // resampler for the specific sample rate conversion being used.
 class SinusoidalLinearChirpSource : public SincResamplerCallback {
  public:
-  // |delay_samples| can be used to insert a fractional sample delay into the
-  // source.  It will produce zeros until non-negative time is reached.
-  SinusoidalLinearChirpSource(int sample_rate, int samples,
-                              double max_frequency, double delay_samples);
+  // |delay_samples| can be used to insert a sample delay into the source.  It
+  // will produce zeros until non-negative time is reached.
+  SinusoidalLinearChirpSource(int sample_rate, size_t samples,
+                              double max_frequency, size_t delay_samples);

[Andrew MacDonald, 2015/07/23 01:07:03]

I see that we don't make use of the fractional delay, but why remove that
possibility? It doesn't make the implementation any simpler.

[Peter Kasting, 2015/07/23 06:02:16]

I think it eliminates a cast (to avoid signed-vs.-unsigned when comparing size_t
against double) but at the cost of adding a different one, so that's a wash.

It seems strange and misleading to allow for fractional samples when we don't
use that.  Maybe it's just me, but fractional samples seem like a weird thing,
and even if the implementation is not actually simpler, it feels less complex
that we implement dealing with integral samples using some floating-point math
under the hood, than that we go ahead and expose an API dealing in fractional
samples.

Semantically, I'm using size_t to represent "count of samples".  So if
|delay_samples| is a count of samples, but it's using a different type, that
feels weird.

There's a lot of "feels" in the justification above, which is probably another
way of saying my rationale isn't very strong.  If you still want me to restore
this, I will.

[Andrew MacDonald, 2015/07/24 03:28:33]

It might seem weird at first blush, but this whole resampling thing is about
interpolating fractional sample values, so it's not odd at all in this context.
I have a small preference for restoring it.

[Peter Kasting, 2015/07/27 23:35:22]

Restored.

 
   virtual ~SinusoidalLinearChirpSource() {}
 
-  void Run(int frames, float* destination) override;
+  void Run(size_t frames, float* destination) override;
 
-  double Frequency(int position);
+  double Frequency(size_t position);
 
  private:
   enum {
@@ -41,11 +41,11 @@ class SinusoidalLinearChirpSource : public SincResamplerCallback {
   };
 
   int sample_rate_;
-  int total_samples_;
+  size_t total_samples_;
   double max_frequency_;
   double k_;
-  int current_index_;
-  double delay_samples_;
+  size_t current_index_;
+  size_t delay_samples_;
 
   DISALLOW_COPY_AND_ASSIGN(SinusoidalLinearChirpSource);
 };