[FFmpeg-devel] [PATCH] avdevice/decklink: add option to drop frames till timecode is seen

[Marton Balint]

On Wed, 13 Nov 2019, Gyan wrote:

> As suggested in the review of an earlier patch*, this one drops frames till a 
> TC is seen.
>

> From 05555932db9d14dea215c1b0e5dcac4eac16cd6a Mon Sep 17 00:00:00 2001
> From: Gyan Doshi <ffmpeg at gyani.pro>
> Date: Mon, 9 Sep 2019 18:33:09 +0530
> Subject: [PATCH] avdevice/decklink: add option to drop frames till timecode is
>  seen
> 
> Option wait_for_tc only takes effect if tc_format is set
> ---
>  libavdevice/decklink_common.h   |  1 +
>  libavdevice/decklink_common_c.h |  1 +
>  libavdevice/decklink_dec.cpp    | 12 ++++++++++++
>  libavdevice/decklink_dec_c.c    |  1 +
>  4 files changed, 15 insertions(+)

Docs update is missing for the new option as well as micro version bump.

> 
> diff --git a/libavdevice/decklink_common.h b/libavdevice/decklink_common.h
> index 921818ba41..35422a300b 100644
> --- a/libavdevice/decklink_common.h
> +++ b/libavdevice/decklink_common.h
> @@ -149,6 +149,7 @@ struct decklink_ctx {
>
>      int channels;
>      int audio_depth;
> +    unsigned long tc_seen;    // used with option wait_for_tc
>  };
>
>  typedef enum { DIRECTION_IN, DIRECTION_OUT} decklink_direction_t;
> diff --git a/libavdevice/decklink_common_c.h b/libavdevice/decklink_common_c.h
> index ca85ec2504..b78630b5fc 100644
> --- a/libavdevice/decklink_common_c.h
> +++ b/libavdevice/decklink_common_c.h
> @@ -58,6 +58,7 @@ struct decklink_cctx {
>      int copyts;
>      int64_t timestamp_align;
>      int timing_offset;
> +    int wait_for_tc;
>  };
>
>  #endif /* AVDEVICE_DECKLINK_COMMON_C_H */
> diff --git a/libavdevice/decklink_dec.cpp b/libavdevice/decklink_dec.cpp
> index 4da9122bff..69be37ed64 100644
> --- a/libavdevice/decklink_dec.cpp
> +++ b/libavdevice/decklink_dec.cpp
> @@ -784,6 +784,8 @@ HRESULT decklink_input_callback::VideoInputFrameArrived(
>                              if (packed_metadata) {
>                                  if (av_packet_add_side_data(&pkt, AV_PKT_DATA_STRINGS_METADATA, packed_metadata, metadata_len) < 0)
>                                      av_freep(&packed_metadata);
> +                                else if (!ctx->tc_seen)
> +                                    ctx->tc_seen = ctx->frameCount;
>                              }
>                          }
>                      }
> @@ -793,6 +795,14 @@ HRESULT decklink_input_callback::VideoInputFrameArrived(
>              }
>          }
> 
> +        if (ctx->tc_format && cctx->wait_for_tc && !ctx->tc_seen) {
> +
> +            av_log(avctx, AV_LOG_WARNING, "No TC detected yet. wait_for_tc set. Dropping. \n");
> +            av_log(avctx, AV_LOG_WARNING, "Frame received (#%lu) - "
> +                        "- Frames dropped %u\n", ctx->frameCount, ++ctx->dropped);
> +            return S_OK;
> +        }
> +
>          pkt.pts = get_pkt_pts(videoFrame, audioFrame, wallclock, abs_wallclock, ctx->video_pts_source, ctx->video_st->time_base, &initial_video_pts, cctx->copyts);
>          pkt.dts = pkt.pts;
> 
> @@ -880,6 +890,8 @@ HRESULT decklink_input_callback::VideoInputFrameArrived(
>              videoFrame->AddRef();
>
>          if (avpacket_queue_put(&ctx->queue, &pkt) < 0) {
> +            if (ctx->tc_seen == ctx->frameCount)
> +                ctx->tc_seen = 0;

I don't think this is needed, this can only happen in an ENOMEM case. Keep 
it if you feel strongly about it, but it looks a rather useless corner 
case to me.

LGTM if docs is added.

Thanks,
Marton