[Ffmpeg-devel-irc] ffmpeg-devel.log.20161011

[burek]

[00:42:03 CEST] <llogan> taking bets on number of comments at end of day
[01:48:40 CEST] <jamrial> llogan: better yet, drama just moved to the ml
[01:53:25 CEST] <llogan> jamrial: crossposted too, although he forgot -cvslog and libav-user
[02:24:02 CEST] <Compn> doh he left
[03:39:48 CEST] <Compn> carl knows his bugs.
[03:39:54 CEST] <Compn> in apple software
[03:39:57 CEST] <Compn> and his workarounds :D
[03:47:01 CEST] <cone-034> ffmpeg 03Moritz Barsnick 07master:efbc37a757d6: lavfi/pan: renormalize negative gain coefficients properly
[04:58:53 CEST] <cone-034> ffmpeg 03James Almer 07master:c44eae157f62: avformat/matroskaenc: fix targets for attachment tags
[05:11:50 CEST] <philipl> BtbN: All good including npp scaling. Needed one fix:
[05:11:52 CEST] <philipl> https://gist.github.com/philipl/ae19e614c4c4acd1e39e1a86124e236e
[09:46:01 CEST] <cone-291> ffmpeg 03Carl Eugen Hoyos 07master:d0c1b9821aa1: lavf/mxfenc: Do not print a useless error message.
[10:03:14 CEST] <cone-291> ffmpeg 03Carl Eugen Hoyos 07master:4147d8efe5cc: lavf/aiffdec: Default to full rate qcelp as QT does.
[10:03:15 CEST] <cone-291> ffmpeg 03Carl Eugen Hoyos 07master:34aa14546853: lavf/aiffenc: Write extradata also for qcelp.
[10:07:38 CEST] <cone-291> ffmpeg 03Carl Eugen Hoyos 07master:5781256c013e: lavf/matroskaenc: Always write V_QUICKTIME extradata.
[12:37:30 CEST] <Chloe> I dont get why that issue was closed? Surely mp4 *should* work?
[12:38:28 CEST] <nevcairiel> its just carls interpretation of working, there is a workaround so it works!
[12:41:51 CEST] <Chloe> mpv/58
[12:41:54 CEST] <Chloe> err
[12:42:20 CEST] <jkqxz> Changing the timing destroys information and therefore shouldn't be done unless the user asks for it.  Maybe it could do something automatically with mov (because they are implicitly asking for that compatibility), but it shouldn't be done with mp4.
[12:47:00 CEST] <Chloe> The warning could be improved, actually suggest an appropriate flag for the user to add
[12:51:23 CEST] <jkqxz> That would probably be fair.
[12:53:43 CEST] <jkqxz> (The tone of the reporter being "not working around bugs in third-party devices is a Critical Bug in ffmpeg of the Highest Possibly Importance" didn't exactly encourage a response like that, though.)
[12:56:20 CEST] <Chloe> Yes, they didn't have the best tone...
[13:49:04 CEST] <michaelni> Timothy_Gu, in fatebeta its not possible to click on for example "freebsd" to get "&query=os:freebsd" added and only the freebsd instance shown
[14:43:15 CEST] <Compn> Chloe : because if you want to output for quicktime/ipod you should be using -f ipod
[14:43:52 CEST] <Compn> or if you want to make -f iphone default for mp4 ?
[14:44:36 CEST] <Compn> does quicktime support 10bit now ?
[14:50:41 CEST] <JEEB> uhh, have you made sure those requirements still exist as for -f ipod/iphone?
[14:50:48 CEST] <JEEB> if anyone here has modern such devices
[14:51:13 CEST] <JEEB> because I have this small feeling in the back of my head that some of that stuff is just changes required for stuff from ~7+ years ago
[14:52:34 CEST] <Mavrik> Yeah
[14:52:42 CEST] <Mavrik> Modern stuff will take anything really
[14:58:01 CEST] <JEEB> same for the psp movenc stuff methinks
[14:58:31 CEST] <JEEB> since I've never actually used the PSP specific muxing mode and all of my stuff used to work (I know the limitation of the sony parser, sure, but I think the psp-specific mode does some extra in addition to that)
[16:00:53 CEST] <philipl> BtbN: you saw my comment and fix?
[16:01:00 CEST] <BtbN> yes
[16:49:04 CEST] <philipl> BtbN: then looks good to me.
[17:02:31 CEST] <cone-399> ffmpeg 03Moritz Barsnick 07master:99d68d462fbd: doc: fix various typos and grammar errors
[18:21:56 CEST] <cone-399> ffmpeg 03Marton Balint 07master:24022402be4c: lavfi/metadata: fix metadata deletion if comparison returns false
[19:07:54 CEST] <j-b> Subject: [FFmpeg-devel] webvtt in dash
[19:08:01 CEST] <j-b> I would like to convert dvb_subtitle to TTML and wrap in m4s to be muxed in
[19:08:07 CEST] <j-b> DON"T :D
[19:08:49 CEST] <nevcairiel> do we even have the ability to OCR dvb bitmaps subs
[19:12:10 CEST] <j-b> nevcairiel: you fool: TTML can be image-based!
[19:12:17 CEST] <nevcairiel> o.o
[19:12:19 CEST] Action: j-b runs from nevcairiel 
[19:12:34 CEST] <j-b> nevcairiel: and for the record, I am NOT joking.
[19:13:23 CEST] <nevcairiel> i feared as much
[19:14:10 CEST] <ritsuka> but but, TTML can't be muxed in a mp4, it needs to be converted to tx3g or webvtt, so how would one keep the bitmaps?
[19:17:51 CEST] <JEEB> j-b: jesus christ... did they seriously add img tags with base64 images or something?
[19:18:08 CEST] <nevcairiel> doesnt ass also have that
[19:19:06 CEST] <JEEB> yes
[19:19:45 CEST] <JEEB> meanwhile I note to myself: do not search "images ass" to refresh yourself on ASS
[19:20:58 CEST] <nevcairiel> thats probably wise
[19:23:42 CEST] <kierank> JEEB: you should go to demuxed
[19:23:45 CEST] <kierank> j-b: ^
[19:23:49 CEST] <kierank> they love ttml and webvtt
[19:28:24 CEST] Action: JEEB was thinking of demuxed but then started moving house
[19:34:15 CEST] Action: TD-Linux will be there
[19:34:52 CEST] <kierank> at least derek is there
[19:34:57 CEST] <kierank> someone who isn't trying to play politics
[19:34:59 CEST] <kierank> and will just tell teh truth
[19:36:52 CEST] <JEEB> yup
[19:39:32 CEST] <JEEB> meanwhile ARIB got their shop done and now they're selling the specs instead of just providing them :<
[19:42:03 CEST] <TD-Linux> do they sell limited editions with artbook and wall scroll?
[20:11:38 CEST] <j-b> JEEB: base64, of course.
[20:12:17 CEST] <j-b> JEEB: and fun to be added, it is even in the "simple version" of TTMl
[20:13:23 CEST] <JEEB> :<
[20:13:51 CEST] <JEEB> it kind of makes sense since image-based subs is something they want to pass on from subpicture formats
[20:13:57 CEST] <JEEB> but that still makes me a bit of a sad panda
[20:17:01 CEST] Action: j-b injects fun into JEEB 
[20:17:09 CEST] <j-b> sad panda -> happy panda
[20:17:19 CEST] <JEEB> :D
[20:18:27 CEST] <rcombs> nevcairiel: there's a filter for OCRing with Tesseract, but it takes video input (not subtitles, so it'd require sub2video which is bad) and output is in the form of frame metadata
[20:18:28 CEST] <rcombs> both because lavfi doesn't know what subtitles are
[20:18:28 CEST] <rcombs> once it does, I can imagine an image->text sub filter, and vice versa
[20:24:39 CEST] <atomnuker> nice, patches from andreas again, the debian package maintainer
[20:46:51 CEST] <Chloe> Compn: will that fix the timescale as well?
[20:53:09 CEST] <Chloe> unsigned int timescale_new = (unsigned int)av_rescale(st->time_base.den, 1000, st->time_base.num); should be better than unsigned int timescale_new = (unsigned int)((double)(st->time_base.den) * 1000 / (double)(st->time_base.num)); right?
[20:53:32 CEST] <nevcairiel> yes thats what the rescale functions are for
[20:53:41 CEST] <nevcairiel> avoiding overflow
[20:54:25 CEST] <Chloe> okie dokie, I was more concerned with the unsigned int cast but I dont think there's a way around it
[20:57:15 CEST] <Chloe> also this guy just isnt listening
[21:13:19 CEST] <jkqxz> Chloe:  Where do the magic numbers (100000, 1000) come from?  If the 100000 is fixed, it should probably work out the divider more carefully: 1000 looks like too big for, say, NTSC-style 120fps (120000/1001).
[21:15:26 CEST] <jkqxz> (tb.den + 99999) / 100000, maybe?
[21:16:33 CEST] <Chloe> I just got them from carl's post on the ticket (5882)
[21:17:50 CEST] <Chloe> I dont know what it's meant to be
[21:18:22 CEST] <Compn> Chloe i think so
[21:24:15 CEST] <jkqxz> Hmph, very useful commit message for where the 100000 came from: "git show 911f9b714".
[21:27:49 CEST] <jkqxz> Right, it's a 32-bit field; 2^32 / 12*60*60 ~ 100000.  Who would want to make a stream more than 12 hours long anyway?
[21:28:39 CEST] <Compn> what kind of stream
[21:28:46 CEST] <Compn> :D
[21:29:17 CEST] <Compn> hmm mp4 ? 
[21:29:32 CEST] <Compn> wouldnt trust long mp4 file to write index properly
[21:29:44 CEST] <Compn> 12 hour long file that doesnt play lol
[21:39:55 CEST] <Chloe> jkqxz: shall I resend with (tb.den + 99999) / 100000 instead?
[21:45:04 CEST] <RiCON> what's wrong with just using pkg-config to check for libopenjpeg?
[21:46:22 CEST] <RiCON> oh right, the HAVE_OPENJPEG_x_x_OPENJPEG_H macros.
[21:46:45 CEST] <jkqxz> Chloe:  The warning is just generally flaky: you really want it to tell the user to work out the Right Answer, and much of the time that answer is going to be to do nothing because the content isn't long enough for it to matter.
[21:48:05 CEST] <Chloe> yeah it's in the warning as a suggestion anyway, since we cant actually know for sure
[21:48:07 CEST] <jkqxz> (The NTSC 120fps case is one where you actually can't reduce it, so the result if you do will be subtly wrong.  If your video is under 10 hours, though, doing nothing is definitely the best answer.)
[21:57:06 CEST] <jkqxz> How about actually telling the user how long their video can be with that timebase, rather than suggesting a possibly-subtly-wrong lower value?
[22:01:32 CEST] <Chloe> what's the difference between timescale and timebase?
[22:06:11 CEST] <jkqxz> The timebase is the length of one tick from the point of view of the codec (for CFR, that is the exactly the time between two consecutive frames; for VFR, it is the input clock rate).  The timescale is the smallest fraction of a second that you can represent, and you want to be able to represent the timebase exactly in it - that is most easily just the denominator of the timebase, but need not be.
[22:10:14 CEST] <JEEB> Chloe: timebase is the general word and timescale is used in ISOBMFF
[23:25:16 CEST] <cone-125> ffmpeg 03Michael Niedermayer 07master:51f3278089bb: doc/developer: Mention mime type and patchwork in "Submitting patches"
[00:00:00 CEST] --- Wed Oct 12 2016