> > On 08.11.2022 15:25, Eran Kornblau wrote: > >> > >> -----Original Message----- > >> From: ffmpeg-devel On Behalf Of > >> Timo Rothenpieler > >> Sent: Tuesday, 8 November 2022 13:33 > >> To: ffmpeg-devel@ffmpeg.org > >> Subject: Re: [FFmpeg-devel] [PATCH] libx265: support ATSC A/53 > >> captions > >> > >>> > >>> The attached patch adds rendering of ATSC A/53 captions as HEVC SEI messages. > >>> The option name/implementation is aligned with the corresponding libx264 feature. > >> > >> What frees the sei data on success? > >> > > Thanks Timo, good point! > > > > I missed the fact that libx264 gets a callback for freeing the SEI payloads (sei_free). > > It seems libx265 does not have this option, so I added some 'for' loop to free the payloads after the call to encoder_encode. > > > > In addition: > > 1. I added av_memdup on the 'user-data-unregistered' side data so that it won't be freed twice. > > 2. I added a 'free_picture' function for freeing the x265_picture in case of error. > > (both changes are aligned with the implementation in libx264.c) > > > > Updated patch attached. > > One nit: splitting the first line of free_picture() in two isn't really neccesary and makes the function a bit uglier. > Fixed > One small problem: libx265 now needs to select atsc_a53 in configure, analog to how libx264 and all the others do. > Otherwise the file containing the a53 code might not get compiled. > Fixed > LGTM otherwise > Thanks! Eran > > While checking the point you raised, I ran some libx265 transcode with valgrind, and it reported errors unrelated to this patch. > > Attaching the full output, in case anyone wants to have a look. > > x265 has a bunch of issues related to SEI data. > See for example https://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftrac.ffmpeg.org%2Fticket%2F9666%23comment%3A1&data=05%7C01%7C%7C5cd31da5382f4946e0c508dac198454d%7C0c503748de3f4e2597e26819d53a42b6%7C1%7C0%7C638035156992542567%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=9AdmTE%2B4dzHED8KLN0ps2GvvynacXfrWjhVJk4SBWJo%3D&reserved=0 > > i.e. if the number of per-frame sei data ever increases past its initial count, it'll just straight up explode. > _______________________________________________