[Audacity-devel] Bug 200: Rounding error in Dither.cpp

[Matt B. <mbr...@cs...>]

Markus Kuhn pointed out a problem in Dither.cpp that causes noise to be
introduced in conversion from 16 bits to 32 bits and back even when
dithering is turned off.  He also suggested a fix.

Can someone with floating-point conversion knowledge review this patch?
  http://audacity.sourceforge.net/cgi-bin/bugzilla/show_bug.cgi?id=200

I've verified that it works with the testcase, but I'm worried that it
may cause problems when dithering is turned on, or in other situations.
(For example, won't PROMOTE_TO_INT16(1.0f) cause overflow now?)

If this patch isn't correct, what's a better way to fix this problem?