[FFmpeg-devel-irc] IRC log for 2010-03-27

irc at mansr.com irc at mansr.com
Sun Mar 28 01:00:03 CET 2010


[01:30:47] <astrange> mru: i used cortex-a8
[01:30:59] <astrange> gcc 4.5 can generate rev but not rev16... bug filed in a sec
[01:46:28] <mru> can it also do unaligned loads properly?
[02:28:48] <ramiro> mru: what odcctools must be used to compile ffmpeg for the iPhone?
[02:53:21] <mru> ramiro: what what must be used?
[02:57:36] <spyfeng> Hi~ if a code line exceed 80 characters in FFmpeg, how to spllit it into two more lines?
[02:57:53] <spyfeng> any rules?
[02:58:39] <spyfeng> for example:  memcpy(mms->asf_header + mms->asf_header_size, mms->pkt_read_ptr, mms->pkt_buf_len);
[02:59:44] <ramiro> mru: isn't odcctools the equivalent of binutils from apple?
[03:01:25] <ramiro> or, more simply, what toolchain must be used to compile FFmpeg for the iPhone?
[03:02:59] <mru> I've never heard of odcctools before
[03:03:12] <mru> use whatever the latest iphone sdk comes with
[03:03:33] <mru> it probably won't work, but it's silly to attempt making things work with an old version
[03:03:48] <ramiro> hm, I had thought the apple tools were buggy and we were supposed to use some "community" tools
[03:04:04] <mru> oh, the apple tools are beyond buggy
[03:04:12] <mru> but there are no alternatives
[03:04:42] <mru> you have to use yuvi's asm preprocessor though
[03:19:52] <astrange> odcctools is just a more portable packaging of the apple cctools
[03:19:56] <astrange> but it's not updated that often
[03:20:32] <ramiro> what's not updated that often? the cctools or the odcctools?
[03:44:26] <Dark_Shikari> astrange: does gcc automatically merge constants in the same file?
[03:44:30] <Dark_Shikari> i.e. multiple identical const strings
[03:44:42] <Dark_Shikari> e.g. a = "apple"; b = "apple", will it store only one "apple"?
[03:45:35] <peloverde> Dark_Shikari, I was under the impression it does
[03:46:12] <astrange> it's supposed to
[03:46:16] <Dark_Shikari> k
[03:48:50] <saintdev> but, we all know gcc...
[03:49:17] <drv> from what i've heard (not actually tested), it's even smart enough to merge things like "abcdef" and "def"
[03:50:49] <mru> it's allowed to do that
[03:51:08] <mru> there is no promise
[04:02:20] <Dark_Shikari> oh god.
[04:02:21] <Dark_Shikari> oh god.
[04:02:21] <Dark_Shikari> oh god.
[04:02:28] <Dark_Shikari> saintdev: thanks for making me lol hard
[04:02:31] <Dark_Shikari> peloverde: http://saintdevelopment.com/fies/tests.zip
[04:02:45] <Dark_Shikari> One is FAAC at 160kbps, one is FFAAC at 160kbps, one is source.
[04:02:47] <saintdev> Dark_Shikari: you didn't see that the other day?
[04:02:51] <Dark_Shikari> nope
[04:03:03] <Dark_Shikari> I didn't think it was possible for an encoder to be this bad
[04:03:20] <Dark_Shikari> Oh hey, I'll go try celt on these.
[04:03:29] <saintdev> I told you about it in #handbrake, thought you would find it humorous.
[04:03:40] <Dark_Shikari> oh yeah, I never looked at it.  now you can see I did =p
[04:03:46] <astrange>  /fies/?
[04:03:50] <saintdev> peloverde: correct URL is http://saintdevelopment.com/files/tests.zip
[04:04:00] <Dark_Shikari> er, how did that happen
[04:04:10] <Dark_Shikari> /fies/ works here
[04:04:15] <saintdev> astrange: should be /files/
[04:04:29] <saintdev> 1and1 name adjustment ftw
[04:04:53] <saintdev> if you type in a name that's a letter off it'll still work
[04:04:55] <Dark_Shikari> btw, the "good" 160kbps is definitely transparent on that sample.
[04:05:44] <saintdev> what's funny, is quite a few people have picked the "good" sample as the original :P
[04:05:51] <saintdev> ...and that's with faac
[04:06:03] <Dark_Shikari> at 160kbps I would expect transparency
[04:06:19] <saintdev> fixed url in the post
[04:06:22] <peloverde> There is a lowpass that's applied in a stupid manner
[04:08:04] <Dark_Shikari> it lowpasses that much at 160kbps?
[04:08:10] <Dark_Shikari> would be nice to fix that...
[04:08:18] <peloverde> Also perhaps you remember the e-mail where I questioned working on this codebase at all and going and replacing the bad code in faac instead
[04:08:27] <Dark_Shikari> "bad code"?
[04:08:36] <saintdev> might actually be usable without that lowpass, if that's the case.
[04:08:37] <peloverde> Not (L)GPL compatible
[04:08:41] <Dark_Shikari> getting rid of the badly licensed code in faac doesn't make it not shit
[04:08:54] <mru> rotfl, that is AWFUL
[04:08:57] <peloverde> It makes it a better jumping off point
[04:09:15] <Dark_Shikari> mru: the part around 1.5-2.5 seconds is particularly bad
[04:09:25] <Dark_Shikari> er, 0.5-1.5
[04:09:28] <mru> one of the samples is just disgusting
[04:09:34] <Dark_Shikari> mru: that's ffaac
[04:09:38] <mru> I won't say which in case people want to try
[04:09:44] <Dark_Shikari> It's kinda obvious.
[04:09:48] <mru> well, yeah
[04:09:58] <mru> the other two are hard to tell apart
[04:10:02] <mru> I'll have to listen again
[04:10:07] <Dark_Shikari> I can't tell them apart after 10 listens
[04:10:11] <peloverde> Though I'm curious as to the cause of the big spectral holes at the end
[04:10:15] <mru> tomorrow, without waking neighbours
[04:10:18] <Dark_Shikari> spectral holes?
[04:10:21] <Dark_Shikari> mru: headphones?
[04:10:24] <mru> no good ones
[04:11:00] <Dark_Shikari> in some sense, the more complex the music, the harder it is to hear artifacts
[04:11:03] <Dark_Shikari> just like in video
[04:11:07] <Dark_Shikari> complexity hides error
[04:11:09] <peloverde> big holes in the spectrum
[04:11:14] <Dark_Shikari> peloverde: where in the spectrum?
[04:11:21] <Dark_Shikari> (you ran it through a spectrum analyzer?)
[04:11:26] <saintdev> all over, it's swiss cheese at this point.
[04:11:34] <peloverde> between 20 and 24.5
[04:11:38] <Dark_Shikari> khz?
[04:11:41] <Dark_Shikari> or hz
[04:11:47] <saintdev> seconds?
[04:11:50] <Dark_Shikari> oh
[04:11:57] <peloverde> seconds
[04:11:58] <Dark_Shikari> oh holy shit
[04:12:03] <Dark_Shikari> Wow
[04:12:28] <saintdev> once again touhou comes in handy :D
[04:12:45] <astrange> you probably can't tell faac at 160 from the original
[04:13:19] <astrange> if you can, bad headphones might help because of the unexpected response levels
[04:14:52] <peloverde> Also is that using TLS?
[04:15:05] <saintdev> TLS?
[04:15:15] <peloverde> Two Loop Search?
[04:15:33] <peloverde> There are 3 coefficient coders in the aac "encoder"
[04:15:48] <saintdev> dunno, whatever the defaults are.
[04:16:00] <peloverde> The defaults are insanely bad
[04:16:47] <saintdev> so why are the the defaults then?
[04:17:06] <saintdev> tell me what I should be using and I'll do another test.
[04:18:05] <peloverde> The cascaded trellis is supposed to be the best but I'm not sure of the sate of the code that's actually checked in
[04:19:46] <peloverde> how does ffwma do?
[04:23:07] <saintdev> gotta rebuild for that. v2?
[04:24:14] <peloverde> I think v2 is the only version supported
[04:24:28] <saintdev> --list-encoders shows v1 and v2
[04:31:27] <saintdev> it does decent.
[04:35:01] <saintdev> ffwmav2 @ 184kbps (160kbps was what I requested) http://saintdevelopment.com/files/test_4.wav
[04:35:28] <Dark_Shikari> btw, I teseted other encoders for you
[04:35:43] <Dark_Shikari> vorbis sounds amazing down to 64kbps
[04:35:46] <Dark_Shikari> and still quite good at 48kbps (with aotuv)
[04:35:56] <Dark_Shikari> it completely falls apart at 32.
[04:36:15] <Dark_Shikari> nero sounds slightly worse than vorbis at 48, but performs admirably all the way down to 24
[04:36:35] <saintdev> Dark_Shikari: he-aac or just lc?
[04:37:28] <Dark_Shikari> whatever it used.  probably he-aacv2
[04:38:05] <saintdev> ok, just checking if you forced it to use lc
[04:44:12] <peloverde> makes me want to just adapt the WMA encoder to do AAC
[04:47:35] <Dark_Shikari> lol
[04:47:38] <peloverde> Here is the ABR cutoff equation "cutoff_coeff = avctx->bit_rate / (4.0f * avctx->sample_rate * avctx->channels)" lol
[04:47:45] <Dark_Shikari> for wma?
[04:48:09] <peloverde> for AAC
[04:48:21] <Dark_Shikari> ah
[04:50:46] <astrange> the wma encoder is only 400 lines, i'm surprised it sounds good at 128kbit
[04:51:09] <astrange> well, requesting 128kbit gets 140kbit
[04:51:27] <saintdev> astrange: yeah like my requested 160kbit = 184kbit
[04:52:00] <saintdev> peloverde: how does cutoff_coeff relate to the actual cutoff frequency?
[04:54:18] <aaronl> holy shit
[04:54:18] <peloverde> The actual cutoff frequency should be the sample rate * cutoff_ratio / 2, I think
[04:54:21] <aaronl> this is pretty far from transparent
[04:54:28] <peloverde> iirfilter.c is not documented at all
[04:56:10] <saintdev> so that gives a cutoff of 10kHz? lol
[04:56:39] <saintdev> no wonder it sounds like crap
[04:59:38] <peloverde> Sadly I'm getting segfaults if I turn the filter off
[04:59:59] <Dark_Shikari> >_>
[05:00:42] <peloverde> nevermind it's not related to the filter
[05:03:27] <peloverde> There IIRC were some issues with the way ff_aac_pow2sf_tab was structured and my first proposed fix was rejected and I never got around to a second attempt
[11:31:54] <CIA-24> ffmpeg: diego * r22697 /trunk/LICENSE: Clarify which parts of libswscale remain GPL.
[12:36:28] <CIA-24> ffmpeg: stefano * r22698 /trunk/ffprobe.c:
[12:36:29] <CIA-24> ffmpeg: 10l: Initialize tag to NULL in show_stream, fix metadata showing in
[12:36:29] <CIA-24> ffmpeg: streams.
[12:39:52] <pengvado> wow, my optimized init_vlc() managed to break vorbis in such a way that it adds noticeable distortion but doesn't completely desync nor even register any decoding errors.
[12:44:54] <kshishkov> something wrong with symbols?
[12:46:15] <pengvado> could be
[12:46:36] <kshishkov> and I remember that init_vlc() calls in lavc/mpc8.c triggered internal compiler error on some platforms
[12:48:15] <CIA-24> libswscale: diego * r30965 /trunk/libswscale/ (5 files):
[12:48:15] <CIA-24> libswscale: libswscale: Relicense almost all x86 assembler optimizations as LGPL.
[12:48:15] <CIA-24> libswscale: This is of course done with permissions from the authors. The only GPL
[12:48:15] <CIA-24> libswscale: component left are MMX optimizations for YUV to RGB conversion.
[13:18:44] <CIA-24> ffmpeg: diego * r22699 /trunk/doc/developer.texi: Clarify where the list of supported formats is.
[13:20:57] <elenril> name = "avi", name_long = "AVI format" << what's the point of this
[13:21:34] <ramiro> elenril: for ffmpeg - formats
[13:21:40] <ramiro> oops, ffmpeg -formats
[13:22:06] <elenril> i mean name_long contains exactly the same information as name
[13:22:21] <ramiro> elenril: not in all cases
[13:22:58] <CIA-24> ffmpeg: reimar * r22700 /trunk/libavcodec/avcodec.h: Clearer CODEC_CAP_DR1 documentation.
[13:23:06] <elenril> yeah, but in this case (and many other) name_long is redundant
[13:23:34] <ramiro> if it bothers you so much, use --enable-small
[13:23:39] <ramiro> or whatever the configure flag is
[13:24:46] <elenril> no, i just thought it's weird
[13:29:29] <kshishkov> elenril: isn't that a bit idiotic translation here? http://cs.wikipedia.org/wiki/Soubor:Skoda_tabl_%D0%A7%D0%A12.jpg
[13:31:11] <CIA-24> ffmpeg: diego * r22701 /trunk/doc/APIchanges: misc wording and grammar fixes
[13:31:34] <elenril> not really
[13:31:40] <elenril> sounds only a little weird to me
[13:32:27] <elenril> then agains my russian is a bit rusty
[13:34:32] <kshishkov> still, translating "Skoda" as "Mechanical works named after Lenin" is a bit idiotic
[13:54:04] <DonDiego> ramiro, Kovensky: what's the yearly income for a beginner software engineer in brazil? (just a ballpark figure)
[13:54:21] * Kovensky has no idea
[13:54:33] <Kovensky> ramiro would know, probably
[13:55:10] <iive> kshishkov: are you sure it doesn't say that the factory name is called lenin?
[13:58:12] <kshishkov> iive: look closely at small print word ;)
[14:01:24] <kshishkov> DonDiego: some googling suggests it may be <= 24000 USD
[14:01:57] <iive> i don't see any small print. and my cz is nonexistent.
[14:02:12] <kshishkov> it has only one word in Chezh - logo
[14:03:50] <ramiro> DonDiego: yes, certainly below 24000 usd
[14:05:17] <ramiro> I'm not sure since the incomes vary a lot and I don't go around asking (and I'm still far from graduating =), but something around 2k reais per month...
[14:05:54] <kshishkov> some page said it's ~4600 brazil tokens per month for IBM software dev, googling for exchange rate helped me estimate that threshold
[14:06:20] <DonDiego> so something like 20.000 EUR as ballpark?
[14:07:02] <ramiro> ah, ibm... they certainly pay a reasonable income. 4600 seems about right.
[14:07:28] <DonDiego> what's the exchange rate to euro?
[14:07:57] <ramiro> DonDiego: are you thinking big companies or just the regular neighborhood software houses?
[14:08:11] <DonDiego> both
[14:08:21] <kshishkov> 0.41 EUR per king
[14:08:31] <DonDiego> i'm thinking about a very rough average figure
[14:09:17] <DonDiego> for example in germany it's roundabout 40k eur per year at entry level, ronald tells me it's 100k$ in usa
[14:10:08] <kshishkov> here it's probably all food you can steal
[14:10:35] <ramiro> DonDiego: from 10keur for smaller companies and up to some 25keur...
[14:11:11] <DonDiego> so roughly 20k eur
[14:11:16] <mru> DonDiego: 40k eur for a graduate???
[14:11:26] <mru> since when?
[14:11:31] <kshishkov> several years ago I was offered a job at the most notorious sweatshop here for ~4000 USD a year
[14:11:46] <ramiro> but people don't bother working as software engineers here. everyone wants to become manager so they can earn at least double. it is the interns that do all the work (hence poor quality software)
[14:11:51] <DonDiego> mru: that's the average number i hear, 37k - 42k being the standard i think
[14:11:56] <DonDiego> mru: what did you expect?
[14:12:18] <kshishkov> ramiro: also you need to train only your tongue as a manager, not brains
[14:12:52] <ramiro> kshishkov: indeed =)
[14:14:39] <mru> I don't believe those numbers
[14:14:57] <mru> I wouldn't expect a cent above 35
[14:16:12] <ramiro> DonDiego: it also depends on the city you live. in sao paulo you can easily get some 30k or more, but you spend most of it on housing, transportation, and eating.
[14:16:40] <jai> kshishkov: $4k/yr is common here as well
[14:18:01] <DonDiego> i have credible sources, i.e. people i know, but since i'm just looking for ballpark figures, 20% error is ok
[14:18:31] <ramiro> DonDiego: just curious, are you planning to move here? =)
[14:19:22] <DonDiego> haha :)
[14:19:26] <DonDiego> no, not yet..
[14:24:55] <kshishkov> jai: I told you that Ukraine =~ European India, didn't I?
[14:25:15] <kshishkov> ramiro: he's from Argentina anyway
[14:25:51] <janneg> DonDiego: öffentlicher dienst is around 33k
[14:26:27] <ramiro> kshishkov: I know, but argentinians and brazilians hate each other =)
[14:26:40] <DonDiego> only in football :)
[14:27:16] <kshishkov> ramiro: I've heard that argentinians are viewed like people with superiority complex in the rest of South America
[14:27:33] <DonDiego> do we have soc candidates for vc-1 interlaced and dts encoder?
[14:27:38] <ramiro> DonDiego: ha, that's because you don't live in a tourist city... during the summer the city is clogged with rich and filthy argentinians and uruguayians...
[14:28:21] <ramiro> there's one beach here where people won't understand if you speak portuguese during the summer! the whole commerce is in spanish.
[14:28:44] * mru doesn't speak either
[14:28:49] <kshishkov> ramiro: here people won't understand official language too
[14:29:10] <kshishkov> mru: same here but unlike you I also don't speak Swedish and German
[14:29:32] <ramiro> kshishkov: yes, argentinians think of themselves as europeans...
[14:29:52] <mru> yeah, look at DonDiego, he even _lives_ in europe
[14:30:01] <ramiro> kshishkov: hmm I thought you would have learned swedish by now
[14:30:03] <kshishkov> ramiro: here we use term "Moscow denizens"
[14:30:09] <DonDiego> there's hardly an argentinian without a european grand{dad,ma}
[14:30:32] <kshishkov> of German origin?
[14:30:41] <mru> iirc ramiro is some kind of italian...
[14:30:57] <kshishkov> ramiro: yes and I can say some words too but it's too far from claiming that I know it
[14:31:03] <DonDiego> i'm of spanish/basque origin
[14:31:16] <kshishkov> mru: Italian and Spanish are closely related
[14:31:39] <kshishkov> that's kinda feuding origin
[14:31:56] <mru> kshishkov: yes, I understand them about equally well
[14:32:07] <DonDiego> ramiro: you live in florianopolis, right?
[14:32:11] <ramiro> DonDiego: yes
[14:32:28] <kshishkov> mru: same here, I understand most languages equally anyway
[14:32:35] <DonDiego> your city lives off of tourism, so don't complain :)
[14:32:46] <mru> I understand spanish and italian much better than most languages
[14:32:47] <ramiro> kshishkov: me too. I don't understand them =)
[14:32:52] <mru> most languages I don't understand at all
[14:33:07] <mru> spanish I can usually pick out a few %
[14:33:26] <ramiro> DonDiego: well, it's not "my city". I'm only studying here because the university is good. I'm getting out of here as soon as possible.
[14:33:49] <kshishkov> mru: you can always try Lithuanian, Hungarian or Finnish
[14:34:33] <jai> ramiro: when do you finish?
[14:34:56] <mru> kshishkov: I know maybe 5 words in finnish
[14:35:08] <mru> and I can usually recognise hungarian, but not more
[14:35:20] <kshishkov> same here
[14:35:41] <kshishkov> for Finnish because some words are finished Swedish
[14:35:42] <ramiro> jai: good question. I was supposed to graduate last year, but it will probably be in 2012.
[14:36:28] <kshishkov> mru: it was funny when Google translate (sv->en) translated two instances of Hamngatan as "Port street" and once as "Satamakatu"
[14:36:37] <jai> ramiro: oh :|
[14:36:56] <mru> kshishkov: wtf
[14:39:45] <kshishkov> mru: translate.google.com, word "Styrmansgatan"
[14:40:20] <mru> that's finnish, not english
[14:40:23] <mru> wtf is going on
[14:41:32] <kshishkov> dunno but it was funny
[14:42:22] <kshishkov> maybe dictionary compilator walked at Helsingfors and thought that top name on street plaques are in English...
[14:42:34] <mru> hehe
[14:46:40] <kshishkov> I remember that once Google Translate when using en -> ru replaced "Peter Norton" with some Russian antivirus maker name
[14:46:55] <mru> :-)
[14:48:51] <kshishkov> BTW, your nickname is a bit funny after learning PowerPC assembly
[14:50:07] <thresh> Kaspersky, probably?
[14:50:39] <kshishkov> could be
[15:13:23] <kshishkov> BBB: would you believe that I'd been offered to smoke something suspicious by complete stanger. Around noon and on main street. This place is going downhill.
[15:13:40] <BBB> what did he offer?
[15:13:51] <BBB> if it's a cigarette, I'd worry...
[15:13:57] <BBB> if it's anything good, you'd be an idiot not to
[15:13:58] <kshishkov> is was
[15:14:06] <BBB> just a smoke cigarette?
[15:14:10] <BBB> that's it?
[15:14:15] <BBB> yeah, place is going downhill
[15:14:21] <kshishkov> half-smoked
[15:14:21] <DonDiego> haha "just a cigarette" :)
[15:14:29] <DonDiego> dutch people .. ;)
[15:14:30] * kshishkov does not smoke in any case
[15:14:52] <BBB> nothing wrong with a good joint
[15:14:55] <kshishkov> blabbering something about "new pleasure" too
[15:15:24] * BBB wishes there was an easy way to split patches in quilt
[15:15:30] <BBB> my audio clipping patch got so huge
[15:15:39] <BBB> I accidently wrote a qcelp postfilter in that patch
[15:15:50] * kshishkov usually does that with text editor and diff
[15:15:53] <BBB> which was dropdead easy by the way
[15:16:10] <BBB> qcelp is almost a litteral copy of amrnb
[15:16:16] <BBB> just slightly different values
[15:47:52] <wbs> BBB: splitting patches is relatively easy with git ;P
[15:57:37] <BBB> git is one of those things I just don't understand yet :)
[15:57:42] <BBB> will get there, eventually
[15:57:45] * BBB splits patches in 6 or so
[15:58:30] <wbs> hope you don't get scared away by the learning curve, once you get it, it's better than sliced bread :-)
[16:14:51] <pentanol> hi guys, I testing ffmpeg for reading rtsp stream and got weird behaviour. actually It stop streaming after reaching frame= 1774 on around it 1784 1772... why it can be happening?
[16:15:41] <wbs> pentanol: I've seen something similar with some rtsp servers, when running over tcp - not sure why it happens
[16:17:54] <ramiro> BBB: "Apply an order 2 rational transfer function in-place." with in and out it's no longer in-place, right?
[16:20:18] <BBB> oh I didn't update comments, but yes you're right
[16:20:52] <BBB> pentanol, wbs: sounds like server closes connection because we're not sending keepalive requests
[16:21:14] <BBB> pentanol, wbs: try sending dummy OPTIONS requests, rtsp.c has some code for that but it's only enabled for Windows/Real servers right now
[16:21:20] <BBB> feel free to make that more general
[16:51:29] <mru> saintdev, Dark_Shikari, peloverde: I did some careful listening to those files
[16:51:46] <mru> it's _really_ hard to hear any difference between two of them
[16:56:57] <mru> but that sample is very hard to evaluate
[16:57:10] <mru> dozens of voices
[16:57:25] <kshishkov> go train at NSA
[16:57:37] <kierank> I think there's a very slight difference between the two
[16:57:45] <mru> there's a difference for sure
[16:57:52] <mru> especially in the quieter voices
[16:58:07] <mru> but they're something like 10 dB below the loud ones
[16:58:11] <mru> makes it very hard to hear
[16:59:02] <mru> certain types of classical music are much harder to encode transparently
[16:59:30] <ramiro> and that's why kostya re'd ape =)
[16:59:30] <kshishkov> I've heard that Chinese instrumental music is especially hard to code
[16:59:48] <kshishkov> ramiro: I have done nothing like that
[17:00:07] <ramiro> kshishkov: hmm, I might be mistaken. I remember you talking about some lossless codec to listen to classical music
[17:00:50] <kshishkov> APE yes but source was available
[17:01:05] <kshishkov> and some Rockbox guy made it easy to use
[17:01:14] <kshishkov> and some XBMC guy adapted it for lavc
[17:01:19] <ramiro> ah, ok. this is where: http://codecs.multimedia.cx/?p=50
[17:01:23] <kshishkov> I just picked it up and polished a bit
[17:02:38] <jai> hmm, i always thought it was because more classical music rips were in ape ;)
[17:02:58] <mru> ape is pointless
[17:03:10] <mru> it compresses only marginally better than flac or alac
[17:03:19] <mru> at 1000x or more cpu load
[17:03:24] <mru> and an insane design
[17:03:25] <jai> exactly
[17:03:34] <kshishkov> but people use it
[17:03:38] <ramiro> kshishkov: btw you mentioned a philharmonic here http://codecs.multimedia.cx/?p=71 . do you also play something?
[17:03:53] <kshishkov> ramiro: no, I'm a mere listener
[17:06:01] <kshishkov> this world is already suffers from too many people thinking they have a talent
[17:24:28] <siretart> DonDiego: aloha
[17:37:29] <Kovensky> does avcodec_encode_audio always consume the whole input?
[17:37:42] <Kovensky> nvm
[17:58:11] <Dark_Shikari> hmm, question related to a/v sync
[17:58:28] <Dark_Shikari> I have a customer who is streaming using ffmpeg and the a/v sync gets worse over time--and it gets worse every time there's corruption in the input
[17:58:34] <Dark_Shikari> (it's satellite feed or whatever and periodically has some loss)
[17:58:39] <Dark_Shikari> any idea what would cause this?
[18:12:11] <DonDiego> anybody seen yuvi around lately?
[18:17:10] <kshishkov> DonDiego: this channel is publicly logged and you know where logs are
[18:36:41] <Compn> Dark_Shikari : what filetype?
[18:37:09] <Compn> Dark_Shikari : could be that hes playing 29.97 (or vfps) at 30.00 constant framerate :P
[18:39:33] <Dark_Shikari> it's an MPEG-2 transport stream
[18:39:38] <Dark_Shikari> standard TV stuff
[18:44:44] <mru> packet loss causing desync usually indicates someone is ignoring pts
[18:44:52] <mru> and merely counting frames
[18:47:40] <Dark_Shikari> well, this is ffmpeg.
[18:47:59] <mru> no, it's not
[18:48:03] <mru> ffmpeg doesn't do display
[18:48:04] <kshishkov> go humour Baptiste
[18:48:15] <mru> something else has to be involved
[18:48:24] <mru> and it's well-known that ffmpeg.c butchers pts
[18:48:45] <Dark_Shikari> oh I think I know the issue maybe
[18:48:49] <Dark_Shikari> ffmpeg is outputting to FLV
[18:48:57] <Dark_Shikari> the ffmpeg FLV muxer is COMPLETELY BROKEN and forces CFR
[18:49:05] <mru> does flv have timestamps?
[18:49:06] <Dark_Shikari> it does not pass through pts despite being a pure-VFR format
[18:49:07] <Dark_Shikari> Yes
[18:49:10] <Dark_Shikari> FLV has no fps field
[18:49:12] <Dark_Shikari> it only has timestamps
[18:49:58] <Dark_Shikari> But last time I encoded a VFR file with ffmpeg, it changed the pts
[18:50:00] <Dark_Shikari> But only with flv
[18:50:05] <Dark_Shikari> with mp4 it passed them through
[18:53:01] <mru> sound like you found the culprit
[18:53:14] <mru> as I said, something is screwing with timestamps
[18:54:58] <Dark_Shikari> But the writer seems to be writing pts
[18:55:08] <Dark_Shikari> *properly
[18:55:09] <Dark_Shikari> it even has AVFMT_VARIABLE_FPS
[19:10:40] <CIA-24> ffmpeg: reimar * r22702 /trunk/libavformat/eacdata.c: eacdata: fix a memleak, return partial packets and use proper return values.
[19:11:16] <CIA-24> ffmpeg: reimar * r22703 /trunk/libavformat/txd.c: Use more appropriate return values in txd demuxer.
[21:15:09] <CIA-24> ffmpeg: stefano * r22704 /trunk/libavformat/avformat.h:
[21:15:09] <CIA-24> ffmpeg: Document the behavior of av_metadata_get() if the prev parameter is
[21:15:09] <CIA-24> ffmpeg: NULL.
[21:47:39] <hyc> I guess I should use more context lines when posting svn diffs
[21:48:23] <hyc> but also you guys ought to read the actual code in question before questioning the point of a patch
[22:10:03] <Compn> hyc : this is just the way of the open source project with many heads... we berate patch authors to make sure they will stil around thru our rings of fire and maintain such code :P
[22:10:29] <Compn> well , something about that sentence got broken
[22:11:31] <hyc> heh
[22:12:07] <hyc> once again, the focus should be on the actual code, not who the submitter is
[22:12:20] <hyc> and anyone who looks at the actual code will see that there's an inconsistency
[22:13:04] <hyc> and I certainly can't explain why the original code is the way it is
[22:13:48] <hyc> I didn't write it. if they want to put someone thru a ring of fire, it should be the original author who wrote such an inconsistency...
[23:26:06] <ramiro> mru: I'm confused as to what target-path must be. I mounted an nfs share at target-machine to where ffmpeg is being built on host, and point to there. but then the tests take long and the decoding tests fail. I assume the file is not being entirely sent back before the 2nd test starts and therefore it fails. I thought all the generated files would reside in target, is that not so?
[23:51:17] <Kovensky> http://cekirdek.pardus.org.tr/~ismail/ffmpeg-docs/avcodec_8h.html#c41ab8ea9bc0fd3bcf8dc9d3b446f5d3
[23:51:33] <Kovensky> [in] 	samples 	the input buffer containing the samples The number of samples read from this buffer is frame_size*channels <-- the wording is a bit confusing
[23:51:54] <Kovensky> I keep mixing up what is counted in samples and what is counted in bytes :S


More information about the FFmpeg-devel-irc mailing list