[pulseaudio-discuss] [PATCH] echo-cancel: fix the obviously-wrong "buffer+=buffer" logic

[Alexander E. Patrakov]

26.02.2015 09:29, Arun Raghavan wrote:
> On 25 February 2015 at 11:20, Alexander E. Patrakov <patrakov at gmail.com> wrote:
>> 25.02.2015 07:25, Arun Raghavan wrote:
>>>
>>> The falce branch would only be called when recv_counter >
>>> send_counter, so you'd never actually have the zero case. AIUI, this
>>> all just accounting for any "in-flight" rewinds being propagated from
>>> the sink-input to the source-output.
>>
>>
>> OK, this busts the continuity argument, but, as you essentially repeat it
>> below, let's start from that instead.
>>
>>> Also, overwriting the value as you do loses the latency values that
>>> were added in previous lines.
>>
>>
>> No! Look at the first argument of PA_CLIP_SUB. Before the patch, the
>> contribution from the previous lines was double-counted, unless I am missing
>> something.
>
> D'oh! Pushed, thanks.

Could you please also look at the similar patch for module-loopback? 
It's 01/13 in Georg's series, and fixes exactly the same kind of 
double-counting.


-- 
Alexander E. Patrakov