// Vaughan, 2011-02-04: Now that we're updating all meters from audacityAudioCallback,
// this causes an assert if you click Mute while playing, because ResetMeter() resets
// the timer, and wxTimerbase says that can only be done from main thread --
// but it seems to work fine.
This should fix the detection problems on the Mac and may even help with issues
on Windows and Linux.
On any of the platforms, the main issue is the search path for libraries and
how absolute path names are handled. The Mac seems to be especially
susceptible since there isn't one concrete location where libraries are stored.
Windows handles absolute paths differently and allows runtime updates to the
environment variables (and if push comes to shove, provides the
SetDllDirectory() function), so if problems still exist, they should be easy to
circumvent.
This patch does three things:
1) It adds a shell script on OSX that takes care of starting Audacity after
reassigning and clearing the DYLD_LIBRARY_PATH environment variable. This will
allow loading of libraries from their absolute path rather than searching
DYLD_LIBRARY_PATH first. This script should be transparent to the user, but it
will affect people running Audacity with gdb as they will have to specifically
target Audacity.app/Contents/MacOS/Audacity instead of the Audacity.app bundle.
Not big deal really. If ppl no enough to use gdb from the command line, they
should be able to figure it out.
2) It corrects detection of a monolithic FFmpeg library. This is one where
avformat, avcodec, and avutil have all been linked into one large library. The
main issue here was that under OSX and Linux, looking for symbols that should
reside in avutil and avcodec, would always succeed since dlsym() on these
platforms not only scans the requested library, but any dependent libraries as
well. And when avformat was loaded, it would pull in it's dependent libraries
as well.
Another issue here was the if it was determined that the library was not
monolithic, the library was never unloaded, so those dependent libraries may
have come from the wrong location since they would have been loaded via the
normal search paths and not by absolute path name.
3) It adds a method to FileNames which returns the full path name of a loaded
module that contains an address. In the case of FFmpeg, it is used to verify
that a routine from a specific library is actually being used from that library
or not. It is also used to show (in Help->Show log) the directory from which a
library was actually loaded.
This is a workaround for the portaudio issue where changing the default device in xp will corrupt the portaudio device indecies.
This combined with the portmixer fix (earlier today) should address bug 29.
note that this change may require a solution clean and rebuild because it is in libsrc.
Also affects bug 29, as before portmixer was fetching the wrong mixer when many devices were available, but this does not address all issues.
Make OnChoice only do the action associated with the combo box being interacted with.
Also refactor associated input and output device code into new function.
This is kind of a hack that relies upon portaudio to make the first index the mapper for these apis. I verified this for MME in portaudio src, but could not for DirectSound, so this may need to be reverted.
Should address gale's issues on win xp.
Bug 148 (P3) - Label keystrokes painted at playback position
Roger's patch for Bug 255 is to Refresh the whole TrackPanel. He says this should also fix Bug 148.
We'll see if it's too much of a performance hit.
