I cannot reproduce the error. When I run pa_devs.c I get non-zero values for latency. From pa_mac_core.c, these default values are set:

deviceInfo->defaultLowInputLatency = .01;
deviceInfo->defaultHighInputLatency = .10;
deviceInfo->defaultLowOutputLatency = .01;
deviceInfo->defaultHighOutputLatency = .10;

then the latency in frames is queried:
AudioDeviceGetProperty( ..., kAudioDevicePropertyLatency, ... )

that is converted to seconds:
double secondLatency = frameLatency / deviceInfo->defaultSampleRate;

then the default latencies are updated based on that:
deviceInfo->defaultLowOutputLatency = 3 * secondLatency;
deviceInfo->defaultHighOutputLatency = 3 * 10 * secondLatency;

This seems fairly reasonable. I cannot explain why a latency of zero would be returned unless frameLatency was zero. So I added a trap for frameLatency==0 and use the original defaults in that case.

Some useful info from Jeff Moore on latency at:
http://osdir.com/ml/coreaudio-api/2010-01/msg00046.html

http://osdir.com/ml/coreaudio-api/2009-07/msg00140.html

I added some calls to query the properties that Jeff mentions. On my MacBook:

kAudioDevicePropertyLatency had values of 0,10,15. o was for aggregate device.
kAudioStreamPropertyLatency had values of 0,10,15,48. 48 was for a USB headset.
kAudioDevicePropertyBufferFrameSize was always 512.
kAudioDevicePropertySafetyOffset was 32,50,65,134.

Another useful link from Jeff Moore:

>In software play through, the latency (in sample frames) is determined by the following formula (presuming the play through is happening on the same device in the same IO proc):
>
>Input Latency + Input Safety Offset + (2* IO buffer Size) + Output Safety Offset + Output Latency

http://lists.apple.com/archives/coreaudio-api/2004/Jan/msg00445.html

Default latencies are now calculated like Jeff Moore suggested. Need to verify and review.

Looks very close for full duplex. It is hard to measure latency for half duplex.

TRAC migration: propagating ticket status from TRAC