The generator had, i think, never been quite correct with the FIR position of the samples. In 6e0df99 we fixed this sort of, in that we sent the right sample position to allow loops to not click, but by doing so invited a read beyond end of buffer. Whether to include the FIR offset or not in the read data was ambiguous and buggy.
So this commit aims to resolve that by
Clearly documenting the expectation
Adding the FIRoffsets in the appropriate places per the documentation
Testing, and hoping, but definitely improving
This addresses #982 quite nicely and I think just makes everything 'better' but requires some pretty careful testing
The generator had, i think, never been quite correct with the FIR position of the samples. In 6e0df99 we fixed this sort of, in that we sent the right sample position to allow loops to not click, but by doing so invited a read beyond end of buffer. Whether to include the FIR offset or not in the read data was ambiguous and buggy.
So this commit aims to resolve that by
This addresses #982 quite nicely and I think just makes everything 'better' but requires some pretty careful testing