Fix #2032: ABSN extrapolates the last output - #2256

cc @karlt.

This looks sensible, but @karlt knows more about this.

I understand that @rtoy and I see this somewhat differently.
sub-sample start timing and upsampling are both situations where extrapolation would be just as useful before the first buffer sample as after the last sample, so it seems undesirable to me to describe extrapolation only after the last sample.

I would suggest just removing "between" from the text in the previous paragraph "UA-supplied algorithm that interpolates -between- sample frames in the neighborhood of position". If "interpolates" is not considered generic enough to support values outside the provided buffer samples, then it can be replaced with "interpolates or extrapolates". I would be happy that using "extrapolates" in this way would also include UA-specific sinc function approximations.

Sorry for the delay @karlt

I understand that @rtoy and I see this somewhat differently.
sub-sample start timing and upsampling are both situations where extrapolation would be just as useful before the first buffer sample as after the last sample, so it seems undesirable to me to describe extrapolation only after the last sample.

I would suggest just removing "between" from the text in the previous paragraph "UA-supplied algorithm that interpolates -between- sample frames in the neighborhood of position". If "interpolates" is not considered generic enough to support values outside the provided buffer samples, then it can be replaced with "interpolates or extrapolates". I would be happy that using "extrapolates" in this way would also include UA-specific sinc function approximations.

I'm fine with removing "between" as you suggest. I'm not sure what you're suggesting for the rest of the new text, though.

I don't think sinc interpolators are forbidden. It might be unexpected if users say start(t) and there's non-zero and possibly audible output before time t. For all the DSP work I've done, we always used casual systems and had to account for any filter delays appropriately.

I'm fine with removing "between" as you suggest. I'm not sure what you're suggesting for the rest of the new text, though.

If "between" is removed, I'm suggesting that there is no need to add a new paragraph. i.e. modification to the preceding paragraph is sufficient to remove implication that interpolation (or extrapolation) happens only between samples.

I'm fine with removing "between" as you suggest. I'm not sure what you're suggesting for the rest of the new text, though.

If "between" is removed, I'm suggesting that there is no need to add a new paragraph. i.e. modification to the preceding paragraph is sufficient to remove implication that interpolation (or extrapolation) happens only between samples.

That's fine with me. I've made the proposed changes.

@karlt says these changes are ok. I'm going to land this now.