[PATCH v3 6/6] winegstreamer: Use an atomic queue for wg_transform input buffers.

[Zebediah Figura]

On 6/23/22 12:12, Rémi Bernon wrote:
> @@ -802,17 +803,14 @@ NTSTATUS wg_transform_read_data(void *args)
>       InterlockedIncrement(&sample->refcount);
>       InterlockedExchangePointer((void **)&transform->output_wg_sample, sample);
>   
> -    if (!gst_buffer_list_length(transform->input))
> -        GST_DEBUG("Not input buffer queued");
> -    else if ((input = gst_buffer_list_new()))
> +    while (!transform->output_sample && ret == GST_FLOW_OK)
>       {
> -        ret = gst_pad_push_list(transform->my_src, transform->input);
> -        transform->input = input;
> -    }
> -    else
> -    {
> -        GST_ERROR("Failed to allocate new input queue");
> -        ret = GST_FLOW_ERROR;
> +        if ((input_buffer = gst_atomic_queue_pop(transform->input_queue)))
> +            ret = gst_pad_push(transform->my_src, input_buffer);
> +        transform->output_sample = gst_atomic_queue_pop(transform->output_queue);
> +
> +        if (!input_buffer)
> +            break;
>       }

This is correct anyway, but the "ret == GST_FLOW_OK" feels wrong, as 
does continuing to process output samples even after we've hit an error 
condition. Would you mind moving the error check and break into the loop?