[Ffmpeg-devel-irc] ffmpeg-devel.log.20170716

[burek]

[00:20:36 CEST] <wm4> JEEB: the new API forces refcounting
[00:20:50 CEST] <wm4> JEEB: refcounted_frames has been deprecated for this reason
[00:21:06 CEST] <wm4> if you give the new APIs non-refcounted stuff, they'll probably copy it
[00:24:03 CEST] <iive> what?!
[00:29:30 CEST] <JEEB> wm4: yea it was mostly about the flag you could set it
[00:29:39 CEST] <JEEB> which doesn't make sense wrt encoders
[00:29:43 CEST] <wm4> iive: yes
[00:29:47 CEST] <JEEB> since they don't produce refcounted frames
[00:29:50 CEST] <JEEB> only packets
[00:37:07 CEST] <iive> the new api is not ref-counted?
[00:37:15 CEST] <iive> the threaded api?
[00:44:30 CEST] <wm4> the new API forces refcounting, and it has nothing to do with threading
[00:47:39 CEST] <iive> by threaded i was refereing to the async api, that's kind of most useful with threading
[00:47:55 CEST] <iive> but if there is another api that doesn't implement that....
[00:56:45 CEST] <wm4> there's no async API either
[00:57:31 CEST] <iive> i thought libav made one and we merged it.
[00:58:09 CEST] <atomnuker> no, there are plans to make the new api async but nothing's come of it
[00:58:49 CEST] <atomnuker> and that's something I think I remember back when the patches were originally sent to the ml
[01:00:08 CEST] <wm4> there's no need for an async API currently
[01:00:13 CEST] <wm4> nothing would use it
[01:00:30 CEST] <wm4> but the new API could be made async easily
[01:09:04 CEST] <cone-481> ffmpeg 03Michael Niedermayer 07master:0ef8f03133a0: avcodec/aacdec_template: Fix undefined integer overflow in apply_tns()
[03:43:07 CEST] <cone-481> ffmpeg 03Michael Niedermayer 07master:4de4308d2aa3: avcodec/ffv1dec_template: Fix signed integer overflow
[14:29:57 CEST] <cone-900> ffmpeg 03Ashish Singh 07master:615479d51c6a: avfilter: add LIBVMAF filter
[14:30:06 CEST] <BBB> woohoo \o/
[14:30:50 CEST] <JEEB> did netflix enable building that lib?
[14:30:57 CEST] <JEEB> IIRC that guy was trying to make it buildable as lib
[14:40:14 CEST] <durandal_1707> BBB: how is native vmaf supposed to work if its split into several filters?
[15:01:26 CEST] <wm4> <cone-900> ffmpeg Ashish Singh master:615479d51c6a: avfilter: add LIBVMAF filter
[15:01:29 CEST] <wm4> who pushed this?
[15:01:35 CEST] <wm4> apparently the patch review wasn't finished
[15:12:11 CEST] <durandal_1707> wm4: by who..?
[15:14:24 CEST] <RiCON> BBB, according to the commiter name?
[15:22:00 CEST] <wm4> BBB: <Daemon404> wm4, http://fate.ffmpeg.org/?sort=descdate
[15:22:09 CEST] <wm4> <Daemon404> fate explodes~
[15:22:28 CEST] <wm4> seems like configure enables this filter by default
[15:22:32 CEST] <wm4> what a shit show
[15:27:00 CEST] <RiCON> maybe because it's using vmaf_filter_deps instead of libvmaf_filter_deps
[15:27:21 CEST] <wm4> BBB: can you push a revert?
[15:33:45 CEST] <BBB> wm4: uhm
[15:33:49 CEST] <BBB> can you just fix it?
[15:33:52 CEST] <BBB> review was definitely fixed
[15:33:59 CEST] <BBB> dont push reverts just because
[15:34:05 CEST] <nevcairiel> if reviews were ignored, it should probably just go back to the ML
[15:34:06 CEST] <wm4> I don't know anything about this patch
[15:34:11 CEST] <wm4> but fate says it breaks EVERYTHING
[15:34:17 CEST] <BBB> nevcairiel: all reviews were finished and addressed
[15:34:24 CEST] <BBB> let me fix the deps
[15:34:30 CEST] <BBB> I have to run out though
[15:34:32 CEST] <BBB> darn it
[15:34:34 CEST] <nevcairiel> BBB: the ML claims otherwise
[15:35:28 CEST] <BBB> ...
[15:35:31 CEST] <BBB> I have to run
[15:35:35 CEST] <BBB> revert it if you have to
[15:35:39 CEST] <BBB> I have no time unfortunately
[15:35:54 CEST] <BBB> I think you guys could be more constructive, Ive spent a lot of time with the student on getting this in
[15:35:56 CEST] <BBB> :(
[15:36:13 CEST] <wm4> constructive or not, if it breaks everything, it must be addressed quickly
[15:36:21 CEST] <wm4> that's just a fact of life
[15:39:19 CEST] <durandal_1707> lol, fact of life
[15:40:31 CEST] <durandal_1707> leave it to michaelni 
[16:15:04 CEST] <kierank> wm4: Nvidia cuda crap is enabled by default even with --disable-everything
[16:15:13 CEST] <kierank> That was never fixed
[16:15:19 CEST] <wm4> heh
[16:15:34 CEST] <wm4> well, configure is bad with this
[16:15:42 CEST] <wm4> it's hard to enable things by default in a proper way
[16:15:45 CEST] <wm4> honestly, fuck configure
[16:16:12 CEST] <RiCON> cuda uses the same enable_weak as other system hwdecs
[16:16:55 CEST] <wm4> hwdec autoselection is broken too
[16:17:09 CEST] <RiCON> yeah, you have to explicitely disable dxva/d3d11va
[16:17:09 CEST] <wm4> I had some "special" experiences with this
[16:21:45 CEST] <nevcairiel> the cuda and dxva stuff doesnt really harm anything since it doesnt add link dependencies
[16:22:28 CEST] <RiCON> but it could be argued that --disable-everything doesn't actually disable-everything
[16:23:00 CEST] <wm4> the hwaccels can pull in the proper decoders
[16:34:38 CEST] <RiCON> either way, having a broken git master isn't good
[16:49:01 CEST] <wm4> yeah
[16:49:15 CEST] <wm4> if nothing happens in an hour or so I'll push a simple revert
[16:53:57 CEST] <RiCON> $AUTODETECT_LIBS could be added to --disable-all so there's an easier way to disable auto-detected stuff without doing it explicitely for each one
[17:01:04 CEST] <cone-900> ffmpeg 03Michael Niedermayer 07master:ba4beaf6149f: avcodec/apedec: Fix integer overflow
[17:12:53 CEST] <cone-900> ffmpeg 03Derek Buitenhuis 07master:a27c412795c2: webmdashenc: Fix memory leak
[17:12:54 CEST] <cone-900> ffmpeg 03Matt Oliver 07master:d8f198263975: configure: Fix libvmaf name
[17:18:37 CEST] <durandal_1707> wm4: noo, i will pish fix, wait
[17:19:38 CEST] <wm4> k
[17:22:55 CEST] <durandal_1707> does matt fixes it?
[17:24:01 CEST] <nevcairiel> there is probably still other issues in there, like it being enabled by default in the first place
[17:35:29 CEST] <durandal_1707> nevcairiel: its not enabled by default here
[17:35:46 CEST] <nevcairiel> clearly if that was the case, the error would not have happened
[17:40:42 CEST] <RiCON> it was using the wrong filter name
[17:41:13 CEST] <nevcairiel> yet if it was disabled, that should not have mattered
[17:41:13 CEST] <RiCON> it just wasn't being enabled or disabled either way
[17:41:46 CEST] Action: JEEB has learned today that AVSEEK_FLAG_BACKWARD is required for av_seek_frame to work as expected
[17:51:20 CEST] <BtbN> It fixed the Travis-Builds at least
[17:51:43 CEST] <jamrial> nevcairiel: every module (filters/decoders/muxers/etc) is enabled by default during configure until the dependency checks take place
[17:52:07 CEST] <jamrial> in this case, the dependency entry for this filter was broken
[17:52:29 CEST] <jamrial> so the fact libvmaf isn't available didn't disable it
[17:58:29 CEST] <ashk43712> Hi, sorry for the libvmaf patch. I changed everywhere else (from vmaf to libvmaf) but forgot to change this line: vmaf_filter_deps="libvmaf" .
[17:59:05 CEST] <ashk43712> <jamrial> I am not sure how to do that.
[17:59:39 CEST] <jamrial> ashk43712: it's already fixed
[18:00:14 CEST] <ashk43712> This patch is the one that requires Netflix's vmaf library as an external dependency.
[18:00:20 CEST] <ashk43712> thanks a lot.
[18:02:38 CEST] <JEEB> ashk43712: did they already make it buildable as a library_
[18:02:39 CEST] <JEEB> ?
[18:02:47 CEST] <JEEB> that was the primary issue before I think?
[18:03:40 CEST] <ashk43712> yeah, I opened a pull request for the library integration into their code. That was merged a week ago.
[18:04:18 CEST] <ashk43712> Currently, I am implementing vmaf metrics (ansnr, adm, motion and vif) inside ffmpeg so that it can have an independent implementation (without any library) of vmaf later.
[18:05:39 CEST] <ashk43712> vmaf installation is a three step process after getting Netflix's vmaf repo:
[18:05:50 CEST] <ashk43712> make, make lib and make install
[18:19:24 CEST] <JEEB> ok, cool
[19:01:38 CEST] <thebombzen> Quit: ChatZilla 0.9.86.1 [Firefox 2.0.0.22pre/2010030309]
[19:01:49 CEST] <thebombzen> Wow.
[19:23:06 CEST] <DHE> in fairness, firefox 2 is way better than firefox 3 was. or firefox 4. or...
[19:24:05 CEST] <BtbN> Firefox will soon be a Chrome-Copy anyway
[19:37:26 CEST] <durandal_1707> i will apply dolbye if nobody will
[19:39:54 CEST] <JEEB> I think it was already once requested here?
[19:40:36 CEST] <jamrial> JEEB: he means applying the patchset on the ml
[19:40:43 CEST] <JEEB> yes
[19:40:51 CEST] <JEEB> I think kierank or so already poked him to push it
[19:41:50 CEST] <jamrial> foo86 should be given push access by now. he actively maintains the dts decoder and now dolby_e
[19:41:55 CEST] <jamrial> i told him to send his ssh key to michaelni some time ago but i guess he didn't
[19:42:03 CEST] <JEEB> yea
[19:42:12 CEST] <JEEB> I see no reason not to have him have push access :)
[19:42:37 CEST] <durandal_1707> but he wants to be anon
[19:47:51 CEST] <jamrial> throwing a public ssh key our way wouldn't change that
[19:51:41 CEST] <michaelni> jamrial, i asked him to send a patch to add himself to the MAINTAINERs file over a year ago IIRC, not sure why he didnt 
[20:03:33 CEST] <wm4> was master fixed yet
[20:04:02 CEST] <BtbN> Travis builds fine at least
[20:06:24 CEST] <wm4> at least seems to build here too
[20:56:18 CEST] <thebombzen> looking at MAINTAINERS is a bit amusing. There's all these names and then just "wm4" who was the only one who choose not to use real name
[21:12:42 CEST] <atomnuker> BBB: what's the point of having support for the library if its being ported/
[21:16:46 CEST] <iive> atomnuker: easier to compare both versions?
[21:17:13 CEST] <BBB> atomnuker: whats the point of fdkaac or other aac encoders if were workign on our own? :)
[21:17:28 CEST] <atomnuker> i guess
[21:17:32 CEST] <BBB> atomnuker: thats not meant to be an asshole response btw, sorry if it sounded like one
[21:17:40 CEST] <BBB> but I think the reasons are comparable
[00:00:00 CEST] --- Mon Jul 17 2017