On Tue, Nov 18, 2014 at 5:14 PM, David Henningsson <david.henningsson at canonical.com> wrote: > > > On 2014-11-18 14:46, Alexander E. Patrakov wrote: >> >> Add support for libsoxr resampler: David's objection about overriding >> pa_resampler_request is 100% valid, and the patchset cannot be merged >> without taking it into account. > > > Well, the result will be inoptimal rather than completely not working > without a working pa_resampler_request (especially given that Andrey seems > to be satisfied with the current behaviour). If we're given fewer samples > back than we expected, we'll just go through another round of > resampling/mixing/etc, which I assume is what happens here. > > We could also statically add "20 ms" to pa_resampler_request (only in case > soxr is selected, of course) and see if that gives a better result from an > optimality perspective. > >> So, it looks like we are waiting for a >> new version - not only of the patchset, but also of libsoxr. Otherwise >> there is no API to implement pa_resampler_request. I'm not sure how pa_resampler_request is used in PA. If it's just an estimate (e.g. to allocate buffers), and the code that uses the resampler actually uses the sample counts consumed and produced by the resampler, everything should work fine. Is there a context where the resampler absolutely must produce samples according to pa_resampler_request?
