[Ffmpeg-devel-irc] ffmpeg-devel.log.20150621

burek burek021 at gmail.com
Mon Jun 22 02:05:02 CEST 2015


[00:44:07 CEST] <cone-285> ffmpeg 03Rodger Combs 07master:631d56ffc823: lavc/adpcm: THP: fix indentation
[01:27:21 CEST] <J_Darnley> Can I bitch about libavfilter continuing to request frames even though I returned -1 from filter_frame?
[01:32:08 CEST] <nevcairiel> as long as its not EOF it will continue doing things
[01:32:24 CEST] <nevcairiel> a generic error code is not go ing to abort the entire thing usually
[01:33:33 CEST] <cone-285> ffmpeg 03James Zern 07master:9b747500f361: libvpxenc: support setting colorspace for vp9
[01:33:52 CEST] <J_Darnley> Would that be one of the "regular" errors or one of the "macro" errors?
[01:34:03 CEST] <J_Darnley> EOF, I mean?
[01:41:45 CEST] <J_Darnley> AVERROR_EOF
[01:41:57 CEST] <J_Darnley> Well, that was a slight improvement
[01:43:27 CEST] <J_Darnley> ffmpeg doesn't spam an alternating two messages now (just the one, 3300 times)
[01:45:36 CEST] <nevcairiel> you probably cant trigger EOF from a filter in the middle of the chain, EOF needs to trickle down from the source
[01:50:09 CEST] <J_Darnley> At least I can now see the error I am printing
[01:54:47 CEST] <J_Darnley> Anyway, thanks.
[02:54:10 CEST] <cone-285> ffmpeg 03Michael Niedermayer 07master:dde6b2a35545: avcodec/mjpegdec: dont try to combine fields for decimated multiscope 2 material
[03:30:08 CEST] <BBB> so I seem to recall that there was a way to seek to seek only in one direction of a timestamp, e.g. seek to a keyframe on or before, but not after, ts=X
[03:30:20 CEST] <BBB> I dont think av_seek_frame() does that; is that avformat_seek_file()?
[04:33:42 CEST] <gchristensen> I doubta nyone is around ... but: I'm looking to setup FATE in my build pipeline and can contribute results per commit for debian (oldstable, stable, testing, unstable, experimental), using the current GCC and other build toolchains. thislooks like it'd be helpful, but I just want to make sure.
[04:41:04 CEST] <jamrial> gchristensen: as you were told the other day, setting up fate clients is of course helpful
[04:42:07 CEST] <jamrial> although i'd say to set cron to run them one to four times a day rather than per commit
[04:48:32 CEST] <gchristensen> jamrial: are the commits too frequent for the more regular feedback? I have plenty of room in the build cluster. I can also run on various other distributions, centos, ubuntu, and fedora fairly easily. I suppose I'd be happy to set this up and go to town as-is, but I suppose the reason I keep looking for feedback is because I'm happy to set it up in a way which most helps the project.
[04:50:42 CEST] <jamrial> some days the repository gets like 20 commits in an hour, others it only gets five in a day. it varies
[04:52:03 CEST] <jamrial> most clients are setup to run every x amount of hours, so that's probably your best choice
[04:53:03 CEST] <jamrial> and yeah, the more variety (distributions, configurations, compiler versions, architectures, etc) the better
[04:53:56 CEST] <gchristensen> I'm actually not sure our build cluster supports cron, but I'll check in to that. OK.
[04:54:10 CEST] <gchristensen> I'll go ahead and setup what I have, get a public key to the mailing list, and we can change course from there.
[04:56:02 CEST] <jamrial> gchristensen: not to the mailing list, send it to the email address mentioned in doc/fate.texi, fate-admin at ffmpeg.org
[04:56:21 CEST] <gchristensen> yep that is what I meant, I just assumed it was a mailing list.
[05:08:33 CEST] <cone-285> ffmpeg 03Michael Niedermayer 07master:f98c3ec5c68d: avformat/movenc: Check return code of ff_iso8601_to_unix_time()
[05:08:34 CEST] <cone-285> ffmpeg 03Michael Niedermayer 07master:54649cbda890: ffmpeg_opt: Fix handling of creation_time now
[11:05:27 CEST] <halfie> hi! can the muxer / demuxer parse the file format, and pass on the extra data (needed by the bitstream filter) in the AVCodecContext object?
[11:06:04 CEST] <halfie> (I have written a new bitstream filter which I am trying to make non-hacky now)
[11:29:21 CEST] <durandal_1707> News on web page still not list 2.7
[11:37:14 CEST] <cone-874> ffmpeg 03Paul B Mahol 07master:95ee0fbacfad: Support demuxing 4gv codec stored in qcp files
[12:25:26 CEST] <cone-874> ffmpeg 03Michael Niedermayer 07master:6c8a05268cfe: avformat/rawenc: Store sample number for ADX
[13:18:03 CEST] <cone-874> ffmpeg 03Rodger Combs 07master:6c56827e629b: lavf/brstm: move bfstm var to a local
[13:18:04 CEST] <cone-874> ffmpeg 03Rodger Combs 07master:70a39bcf1c95: lavf/brstm: cleanup; fix short-block demuxing
[13:18:05 CEST] <cone-874> ffmpeg 03Rodger Combs 07master:9c9cf3956cd6: lavf/brstm: add support for seeking
[13:18:06 CEST] <cone-874> ffmpeg 03Rodger Combs 07master:d2ce10093e37: lavf/brstm: handle a BFSTM endianness oddity
[13:18:07 CEST] <cone-874> ffmpeg 03Rodger Combs 07master:0f6735194459: lavc/adpcm: THP: don't use the ADPC/SEEK table when not seeking
[13:18:08 CEST] <cone-874> ffmpeg 03Paul B Mahol 07master:a49154e9d79a: avformat/brstm: support little-endian s16 PCM
[14:00:05 CEST] <cone-874> ffmpeg 03Michael Niedermayer 07master:3bd46914fbf2: avcodec/jpeg2000dwt: Allow testing with a maximum allowed difference
[14:00:06 CEST] <cone-874> ffmpeg 03Michael Niedermayer 07master:d23b3f3940a9: avcodec/jpeg2000dwt: Also test 9/7 integer DWT
[14:00:07 CEST] <cone-874> ffmpeg 03Michael Niedermayer 07master:aa594415a0a2: avcodec/jpeg2000dwt: increase the number of test iterations
[14:00:08 CEST] <cone-874> ffmpeg 03Michael Niedermayer 07master:6528b90dea7c: avcodec/jpeg2000dwt: List differences for DWT test
[16:10:30 CEST] <wm4> michaelni: in which format does libswresample resample and down/upmix? is it always float like in libavresample?
[16:13:21 CEST] <michaelni> libswresample supports float, double, int16 and int32 as internal formats, see internal_sample_fmt
[16:13:40 CEST] <wm4> so if input is 16 bit, it gets mixed in 16 bits?
[16:15:28 CEST] <wm4> I'm having a user who complains that with 16 bit in and float out, the result (24 bit on the audio device) has only 16 bits set, although logically the lower bits should not be completely 0
[16:15:38 CEST] <wm4> and I don't know where these bits get lost on the way
[16:16:58 CEST] <michaelni> see swresample.c line 212, thats the code that chooses the internal format when its not set by the user
[16:17:32 CEST] <wm4>     213         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
[16:17:32 CEST] <wm4>     214             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
[16:17:34 CEST] <wm4> that's really dumb
[16:17:46 CEST] <wm4> giess I'll disable libswresample for these builds
[16:18:46 CEST] <wm4> michaelni: how does it make sense to mix in 16 bit when the output is float?
[16:20:10 CEST] <Daemon404> wm4, didnt swr have a sox backend
[16:20:11 CEST] <Daemon404> or something
[16:20:17 CEST] <Daemon404> (optional)
[16:20:21 CEST] <wm4> yes, it breaks the API
[16:20:33 CEST] <Daemon404> wut
[16:20:49 CEST] <wm4> some essential API functions return an error if soxr is used
[16:20:58 CEST] <kierank> wm4: because swr is swscale for audio =p
[16:21:07 CEST] <kierank> everything goes through 4:2:0 planar
[16:21:13 CEST] <Daemon404> wm4, i'd assume sox is too slow to use for playback anyway
[16:21:14 CEST] <wm4> and libavresample is "unsupported"
[16:21:31 CEST] <rcombs> kierank: wat
[16:21:40 CEST] <wm4> Daemon404: no idea, I think it's actually fast enough
[16:21:53 CEST] <wm4> rcombs: aka 16 bit integer planar
[16:22:01 CEST] <Daemon404> kierank, for fun i tried to use OCIO yesterday
[16:22:06 CEST] <Daemon404> i ended that experiment pretty quickly
[16:22:12 CEST] <kierank> never heard of it
[16:22:19 CEST] <Daemon404> OpenColorIO
[16:22:24 CEST] <Daemon404> it's pretty prevalent
[16:22:41 CEST] <Daemon404> terrible C++ API with terrible SWIG bindings
[16:23:46 CEST] <rcombs> waitasec
[16:24:29 CEST] <rcombs> wm4: doesn't that code only change u8 to s16p?
[16:24:40 CEST] <wm4> yes
[16:25:09 CEST] <wm4> so if input is 16 bit, and output is float, all processing is done in 16 bit, and then converted to float
[16:25:13 CEST] <rcombs> ah
[16:25:18 CEST] <michaelni> wm4 no
[16:26:18 CEST] <michaelni> if you do not set it to float it will pick some format automatically and ATM that is 16bit if the input has only 16bit as theres really not more precission but we could increase this default if people really want
[16:27:16 CEST] <michaelni> the AVOption to force a specific format is "internal_sample_fmt"
[16:30:20 CEST] <wm4> it doesn't make much sense to perform a downmix in 16 bit samples
[16:35:07 CEST] <michaelni> down/upmix sums the scaled 16bit samples in 32bit before returning to 16bit
[16:42:17 CEST] <wm4> so 16 bit -> 32 bit -> 16 bit -> float
[16:42:20 CEST] <wm4> sense this makes none
[16:44:27 CEST] <wm4> maybe you should just copy the libavresample logic
[16:44:30 CEST] <wm4> or code
[16:44:41 CEST] <wm4> (why the fuck does libswresample even exist)
[16:45:33 CEST] <kierank> wm4: seriously use libavresample
[16:45:48 CEST] <kierank> this is exactly why it was written
[16:54:35 CEST] <durandal_1707> what is faster?
[16:57:38 CEST] <Daemon404> theyre both sufficiently fast
[16:57:49 CEST] <Daemon404> the better question is: which is more correct
[16:58:00 CEST] <durandal_1707> IIRC swscale doesn't use 420 for everything
[16:58:10 CEST] <nevcairiel> guess i'm lucky that i used avresample before swr was even a thing, and never even thought about switching
[16:58:20 CEST] <Daemon404> nevcairiel, swr was pushed before avr
[16:58:32 CEST] <Daemon404> durandal_1707, the problem is when it does, it does so silently
[16:58:35 CEST] <Daemon404> as an internal step
[16:58:39 CEST] <Daemon404> it used to be a much larger problem.
[17:03:18 CEST] <durandal_1707> for audiophiles mixing should use higher bitdepth format, no?
[17:03:52 CEST] <Daemon404> audiophiles would use libsox
[17:08:57 CEST] <iive> nevcairiel: wasn't avresample written a few months after swr ?
[17:10:11 CEST] <Daemon404> avr was started before swr, swr was pushed first
[17:10:14 CEST] <iive> michaelni: imho, you have input and output formats set, the routine must use the one of them that have highest precision.
[17:10:37 CEST] <iive> i doubt that...
[17:11:04 CEST] <Daemon404> 1. avr starts
[17:11:10 CEST] <Daemon404> 2. youtube gets impatient and pays michael
[17:11:13 CEST] <Daemon404> 3. swr is born
[17:12:00 CEST] <michaelni> iam not sure this is true
[17:12:19 CEST] <michaelni> when did avr and swr start ?
[17:12:39 CEST] <Daemon404> it involves ffmtech or whatever
[17:12:49 CEST] <Daemon404> it doesn't really matter to me who was first or why stuff was written
[17:12:55 CEST] <Daemon404> it's all sausage to me.
[17:13:21 CEST] <iive> avr was paid by ffmtech. and i clearly remember that the decision for that was made after swr was committed.
[17:13:36 CEST] <iive>  i might be wrong...
[17:14:23 CEST] <iive> that's why i usually check the history. but it takes time, to search back.
[17:14:24 CEST] <michaelni> IIRC i heared about ffmtech deciding to fund avr long after i started writing swr
[17:14:38 CEST] <Daemon404> doesnt really matter in the end.
[17:14:42 CEST] <Daemon404> it's 100% irrelevant to a user.
[17:15:47 CEST] <iive> apology accepted.
[17:15:59 CEST] <iive> ;)
[17:16:20 CEST] <Daemon404> there is no apology
[17:17:48 CEST] <iive> just kidding.
[17:34:08 CEST] <rcombs> wm4: how about expressing seek points as metadata
[17:34:13 CEST] <rcombs> erm, loop points
[17:34:46 CEST] <rcombs> (wm4 or other interested parties, though I'm not sure if there actually are any)
[17:35:57 CEST] <wm4> probably fine
[17:37:32 CEST] <ubitux> Daemon404: what did you need opencolorio for?
[17:42:39 CEST] <Daemon404> ubitux, i didnt need it
[17:42:41 CEST] <Daemon404> i tried it.
[17:42:54 CEST] <ubitux> try what in particular?
[17:43:18 CEST] <Daemon404> gamma crap
[17:44:13 CEST] <rcombs> wm4: think a float, or an AV_TIME_BASE int
[17:52:06 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:4299f085f45c: avcodec/jpeg2000dwt: also test 9/7 float DWT
[17:52:07 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:4977692461f0: swresample: Choose 16bit internally only if input and output is 16bit or less
[17:55:05 CEST] <rcombs> michaelni: why 2 branches doing the same thing, instead of an ||? 
[17:55:48 CEST] <michaelni> thought its more readable but feel free to change it
[17:58:10 CEST] Action: rcombs shrugs
[17:58:21 CEST] <rcombs> as long as it was deliberate
[18:00:23 CEST] <wm4> michaelni: can you add a log call printing which format is used?
[18:05:39 CEST] <michaelni> sure, will do
[18:38:51 CEST] <jamrial> michaelni: don't merge 8655c54 from libav. our implementation is too different and apparently better
[18:40:27 CEST] <jamrial> at most merge the configure changes that bump the version requirement to vpx 1.3.0, if that's ok with people
[18:40:44 CEST] <jamrial> i can then look into removing some of the related ifdeffery mess
[18:56:40 CEST] <BBB> I dont understand why they dont just merge our changes
[18:56:41 CEST] <BBB> I mean
[18:56:48 CEST] <BBB> we didnt write them, google did, and they own libvpx
[18:57:01 CEST] <BBB> so who could possibly know better whats good for users than them? at least for that one module
[18:57:50 CEST] Action: jamrial shrugs
[19:00:40 CEST] <Compn> BBB : libav is upstream of google too. :P
[19:27:40 CEST] <ubitux> BBB: why would they care about irrelevant downstreams ;)
[19:27:55 CEST] <BBB> so you dont have to do everything yourself?
[19:28:09 CEST] <ubitux> :)
[19:34:10 CEST] <kierank> they like doing things themselves
[19:34:15 CEST] <kierank> it makes them pretend they have relevance
[19:34:21 CEST] <kierank> sad but true
[19:36:05 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:5de3a589f16e: swresample/swresample: Print used int_sample_fmt
[19:36:06 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:3bcf61f495dc: avcodec/jpeg2000dwt: Fix value of F_LFTG_X and I_LFTG_X
[19:40:25 CEST] <BBB> ubitux: oh, side-note, the audience at webm summit last week was positively surprised about ffvp9, so GreatJob[tm]!
[19:40:42 CEST] <wm4> "surprised"?
[19:40:44 CEST] <ubitux> not like i've done much :p
[19:40:47 CEST] <wm4> so they didn't even know it existed=?
[19:40:53 CEST] <BBB> half of them probably didn't
[19:41:04 CEST] <wm4> michaelni: thanks
[19:41:13 CEST] <michaelni> np
[19:41:14 CEST] <ubitux> damn i need a photoshop outside of my work to get done with this !@#$ filter
[19:41:19 CEST] <ubitux> that shit is fucking expensive
[19:41:21 CEST] <ubitux> :(
[19:41:35 CEST] <wm4> use gimp
[19:41:45 CEST] <ubitux> seriously, you need a monthly subscription (like $50/m or so)
[19:41:48 CEST] <wm4> or a >100 characters ffmpeg command line involving libavfilter
[19:41:53 CEST] <ubitux> yeah but gimp doesn't have the feature i'm trying to replicate
[19:42:25 CEST] <wm4> did photoshop make itself unpiratably by becoming a web service?
[19:43:10 CEST] <ubitux> well, even if i pirate it (evil! :o), i'll need to setup a no-internet vm with linux and hope wine will works
[19:43:15 CEST] <ubitux> so it's still fucking annoying
[19:43:32 CEST] <BBB> you just want to test?
[19:43:43 CEST] <BBB> or you need photoshop for the filter itself?
[19:44:05 CEST] <ubitux> i just want to test
[19:44:13 CEST] <BBB> send me image and tell me what sequence of menu items to press
[19:44:16 CEST] <ubitux> i mean i need to look how it behaves
[19:44:18 CEST] <BBB> Ill send result back
[19:44:26 CEST] <kierank> ubitux: use the 30 day trial
[19:44:31 CEST] <BBB> or that
[19:44:34 CEST] <ubitux> BBB: well, i have a bunch of settings to try
[19:44:39 CEST] <ubitux> kierank: ah? sounds good
[19:44:44 CEST] <ubitux> works with wine?
[19:44:48 CEST] <kierank> unlikley
[19:44:49 CEST] <Daemon404> i somehow doubt it
[19:44:55 CEST] <ubitux> will try anyway
[19:50:09 CEST] <Compn> why refuse to run w7 vm for pshop? :P
[19:50:19 CEST] <Compn> wonder how easy it is to get a w7 vm haha
[19:51:17 CEST] <ubitux> because it will require to pirate yet another sw
[19:51:34 CEST] <ubitux> just to play with 2 spin menu in a filter :p
[19:52:59 CEST] <ubitux> fuck i need to register for the trial as well
[19:52:59 CEST] <Daemon404> windows has a 3 month trial
[19:53:08 CEST] <Daemon404> oh muh privacy?
[19:53:18 CEST] <ubitux> more like pita
[19:54:34 CEST] <ubitux> https://appdb.winehq.org/objectManager.php?sClass=version&iId=31467
[19:54:46 CEST] <ubitux> should be good enough
[19:55:10 CEST] <ubitux> What does not
[19:55:11 CEST] <ubitux> -creating a new psd file (loading works fine)
[19:55:13 CEST] <ubitux> lol
[19:56:15 CEST] <Daemon404> saving is for newbs
[20:21:21 CEST] <kierank> nevcairiel: renaming :(
[20:28:34 CEST] <kierank> durandal_1707: does that patch work if seeking fails
[20:28:35 CEST] <kierank> https://git.videolan.org/?p=ffmpeg.git;a=commitdiff;h=6c8a05268cfe2235e37a28a5cc5b20f6c16b90ad
[20:28:45 CEST] <kierank> ah
[20:28:46 CEST] <kierank> ignore me
[20:28:50 CEST] <kierank> if (pb->seekable)
[20:54:15 CEST] <cone-554> ffmpeg 03nu774 07master:6ec688e1bc76: mp3: enable packed main_data decoding in MP4
[20:54:16 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:ae215e2b42dd: Merge commit '6ec688e1bc76dd93151cbca1c340162ae4b10d77'
[21:02:07 CEST] <cone-554> ffmpeg 03Luca Barbato 07master:c060d046aa2f: af_resample: Set the number of samples in the last frame
[21:02:08 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:f399f826ee2d: Merge commit 'c060d046aa2f89c0e601a2dcfbce53f0e36cf498'
[21:16:39 CEST] <cone-554> ffmpeg 03Luca Barbato 07master:05bf3f54e607: libvpx: Do not set vp8 only parameters when encoding in vp9
[21:16:40 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:30e9ebb14f19: Merge commit '05bf3f54e6078716f6267df530bf1d474ca3d606'
[21:35:00 CEST] <J_Darnley> ubitux: About a year ago you let me have ssh access to your server (I think you said it was really a laptop)...
[21:35:24 CEST] <J_Darnley> Can I still access it and will you still let me use it?
[21:39:41 CEST] <ubitux> ah mmh i disconnected it a while back bc i found a way to replace the part of the blackbone with the shitty clickpad
[21:39:57 CEST] <ubitux> so i use it as a home laptop nowadays
[21:40:26 CEST] <ubitux> i don't have time to replug it right now but i can make it available for tmr or so
[21:45:28 CEST] <J_Darnley> No, don't bother
[23:07:22 CEST] <cone-554> ffmpeg 03Luca Barbato 07master:8655c5416076: libvpx: Support the vp9 extended profiles
[23:07:23 CEST] <cone-554> ffmpeg 03Michael Niedermayer 07master:1e7008675b2d: Merge commit '8655c54160767de1a6b96f8bc310d6e4eaceff48'
[23:59:40 CEST] <cone-554> ffmpeg 03Rob Sykes 07master:c70c6be22597: swresample: soxr implementation for swr_get_out_samples()
[00:00:00 CEST] --- Mon Jun 22 2015


More information about the Ffmpeg-devel-irc mailing list