[Ffmpeg-devel-irc] ffmpeg-devel.log.20130916

[burek]

[00:04] <durandal11707> 21:24 <+wm4> I heard claims that it was some of the more useful filters // well it still works better than fieldmatch with telecine filter, i dunno what arguments fieldmatch needs
[00:09] <Compn> wm4 : it was thought that just including the whole file (english generated one) was easier than string replacing and such
[00:10] <Compn> well , you'd have to ask michael :)
[00:16] <Daemon404> -#define bswap_16(v) av_bswap16(v)
[00:16] <Daemon404> how nice
[00:16] <Daemon404> purposely fucking up namespacing
[00:16] <Daemon404> those files were lovely.
[01:08] <j-b> Daemon404: libavfi files?
[01:09] <Daemon404> j-b, from the old mplayer files that paul just removed
[02:45] <Compn> what do you mean purposefully ?
[02:46] <Compn> the bswap was in mplayer, when it was imported into ffmpeg its easier to do a define than a search and replace, no ?
[02:46] <Compn> you know what, nevermind
[02:46] <Compn> arguing about how it was done wont help anyone anyways
[02:46] <Compn> back to reviewing patches
[02:47] <Compn> guess i'll add durandal's stuff to changelog. i dont mind. something easy i can do :)
[03:12] <cone-270> ffmpeg.git 03Michael Niedermayer 07master:1e4e0bd907eb: avcodec/mjpegdec: change error return value to be outside the 16bit range
[03:12] <cone-270> ffmpeg.git 03Michael Niedermayer 07master:7fb2622456b9: avcodec/mjpegdec: rgba64 ljpeg support
[03:55] <cone-270> ffmpeg.git 03Compn 07master:9ea359808572: riff: g723_1 decodes vivo g723.1 in asf works on bellair_digital_movie.asf
[03:58] <Compn> one twocc down, like 400 to go.
[11:25] <durandal_1707> how to troll on HA: http://www.hydrogenaudio.org/forums/index.php?showtopic=102672&hl=
[12:06] <cone-282> ffmpeg.git 03Hendrik Leppkes 07master:a06a5b78e2c3: mathops/x86: work around inline asm miscompilation with GCC 4.8.1
[12:06] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:97f8d9feb75d: Merge commit 'a06a5b78e2c3bd2e12aff0627a3b65c6fee555a5'
[12:26] <cone-282> ffmpeg.git 03Martin Storsjö 07master:342c43d154e5: omadec: Properly check lengths before incrementing the position
[12:26] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:a444ddff0386: Merge commit '342c43d154e586bc022c86b168fe8d36f69da9d3'
[12:38] <cone-282> ffmpeg.git 03Martin Storsjö 07master:0ef1660a6365: sierravmd: Do sanity checking of frame sizes
[12:38] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:cfa909ef5b7e: Merge commit '0ef1660a6365ce60ead8858936b6f3f8ea862826'
[12:38] <durandal_1707> michaelni: could you do those merges little faster?
[12:44] <cone-282> ffmpeg.git 03Martin Storsjö 07master:5bbfe193a0a4: vocdec: Don't update codec parameters mid-stream
[12:44] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:ff0a382758b8: Merge commit '5bbfe193a0a41bd2adb648c8c3f6901a575734a2'
[13:00] <saste> durandal_1707, not working on scripting at all
[13:02] <durandal_1707> did you ever started?
[13:03] <saste> durandal_1707, i did some research, I wrote a small lua filter, but then gave up for the much more ambitious SWIG scripting, which is more a libav* wrapper rather than being a lavfi specific thing
[13:03] <saste> but then I didn't proceed any further
[13:04] <saste> i have a fancy to propose a crwodfunding project based on it, now I don't know if i'll ever had the time for it
[13:04] <saste> *I had a fancy
[13:05] <durandal_1707> how useful was that lua filter?
[13:08] <durandal_1707> this delogo filter is nonsense
[13:11] <cone-282> ffmpeg.git 03Martin Storsjö 07master:e90a6846c2c0: cavsdec: Make sure a sequence header has been decoded before decoding pictures
[13:11] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:18e0dd05df13: Merge commit 'e90a6846c2c006fbebd00e1f2789f4a86fafacef'
[13:13] <saste> durandal_1707, useful? almost nothing, it just implemented a callback to print something when the frame was filtered
[13:14] <saste> the problem with lua scripting is that in order to do something useful you also need to bind a significant part of the libav* API
[13:14] <saste> i can post my small lua filter POC
[13:14] <durandal_1707> well for start i need simple one that can call different filter for each frame
[13:15] <durandal_1707> i didn't studied how filtergraph thing works
[13:15] <durandal_1707> is it anywhere explained?
[13:15] <saste> different filters?? like as in interleave?
[13:16] <saste> what's your use case?
[13:16] <durandal_1707> also feature to make possible for filter to reconfigure itself (zoom in/out)
[13:16] <durandal_1707> if x negate else vflip
[13:17] <saste> durandal_1707, dynamic filtergraph reconfiguration
[13:18] <durandal_1707> yes?
[13:18] <saste> there is an RFC and a proposal from nicolas
[13:18] <durandal_1707> from almost year ago :((
[13:19] <durandal_1707> does vf_scale calls configure on other filters?
[13:19] <durandal_1707> it does, in filter_frame
[13:20] <durandal_1707> hmm, i could use that ...
[13:20] <saste> durandal: lua filter patch: http://paste.org/67357
[13:20] <durandal_1707> if its just printf ....
[13:20] <saste> durandal_1707, yes, but it is broken
[13:22] <durandal_1707> so is what scale filter does good ?
[13:22] <durandal_1707> it just means we need update other filters to support reconfiguration without init/uninit
[13:22] <durandal_1707> or add flag for such filters that do not need reinit
[13:22] <saste> yes
[13:26] <durandal_1707> another idea is to make it possible to call another (not same) filter inside filter
[13:29] <saste> durandal_1707, why do you need that? it's easier to output to several filters, a filter which uses other filters internally seems a mess
[13:31] <durandal_1707> well if there is scripting than there is no need for such feature
[13:32] <saste> durandal_1707, what do you think it is higher priority?
[13:32] <saste> probably we should implement dynamic reconfiguration (and hope that the fork won't reimplement it)
[13:33] <saste> libav* scripting is a larger task, and would be useful outside lavfi
[13:33] <durandal_1707> cropdetect + crop would be perfect dynamic configuration testcase
[13:34] <saste> once you have the binding, then it should be relatively easier to write a custom scripting filter
[13:34] <durandal_1707> i already wrote one that uses metadata
[13:34] <durandal_1707> i just did not know how to change video size
[13:34] <saste> durandal_1707, you can't right now
[13:35] <saste> the idea is that when a filter detects link changes it will reconfigure the filtergraph
[13:35] <saste> some filters may need to insert scale/format
[13:36] <saste> it is efficient only if we suppose that link properties don't change oft (which is perfectly acceptable)
[13:36] <saste> we could even change the number of input/output dynamically
[13:37] <durandal_1707> crop just needs to reconfigure itself when size changes
[13:54] <cone-282> ffmpeg.git 03Martin Storsjö 07master:ac9d159015a8: proresdec: Properly make sure an index doesn't run past the limit
[13:54] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:e0fcd0294e2f: Merge commit 'ac9d159015a88aa2721b271875d18482f713f354'
[14:20] <cone-282> ffmpeg.git 03Martin Storsjö 07master:6fc8226e2905: vp3: Check the framerate for validity
[14:20] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:b3970b39491f: Merge commit '6fc8226e29055858f28973bb3d27b63b3b65e616'
[14:29] <cone-282> ffmpeg.git 03Martin Storsjö 07master:85e8192b8593: avpacket: Fix error checking in packet_alloc
[14:30] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:7283d2bb90ab: Merge remote-tracking branch 'qatar/master'
[14:40] <cone-282> ffmpeg.git 03Paul B Mahol 07master:fcea6f71ad9d: avfilter/vf_stereo3d: avoid pointless copy
[14:40] <cone-282> ffmpeg.git 03Paul B Mahol 07master:c74d5929df99: avfilter/vf_stereo3d: BGR24 support for anaglyphs
[14:40] <cone-282> ffmpeg.git 03Paul B Mahol 07master:563057731a5a: avfilter/vf_stereod3d: add const qualifiers to ana_convert()
[15:04] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:02f56bfcc3f5: avformat/sierravmd: use av_freep() for freeing context fields
[16:17] <cone-282> ffmpeg.git 03Paul B Mahol 07master:ca5456db7fa6: avformat/segafilm: use av_freep()
[16:34] Action: durandal_1707 too many hacks from old days
[16:40] <cone-282> ffmpeg.git 03Paul B Mahol 07master:42b8f5fba1f0: avformat/segafilm: remove deplanarization hack
[16:51] <durandal_1707> saste: /win4
[16:52] Action: durandal_1707 ups
[17:13] <saste> durandal_1707, ?
[17:15] <durandal_1707> saste: mistake
[17:18] <durandal_1707> can i get lgtm for adelay?
[17:20] <saste> durandal_1707, lgtm (that is, no more comments from me and I trust you tested it)
[17:42] <cone-282> ffmpeg.git 03Paul B Mahol 07master:9d05de225876: avfilter: add adelay filter
[17:55] <saste> so, we decided to stick with av* rather than with lav* in commit messages?
[17:55] <saste> i tend to prefer commit messages starting with lavfi: over avfilter:
[17:56] <durandal_1707> i just did it because michaelni started 
[17:58] <cone-282> ffmpeg.git 03Timothy Gu 07master:b7dd4598630f: doc/formats: Add documentation for 3 parameters that have been missing
[18:00] <durandal_1707> i think pullup can be changed to do less copy with refcounted frames, thus making it even faster
[18:10] <saste> durandal_1707, epic pullup patch
[18:11] <saste> what's your plan next? still working on eq filters?
[18:11] <durandal_1707> go to holiday after killing libmpcodecs
[18:11] <saste> :)
[18:12] <saste> so libavfilter will be pure again
[18:53] <ubitux> 17:55:41 <@saste> i tend to prefer commit messages starting with lavfi: over avfilter:
[18:53] <ubitux> same
[18:53] <ubitux> 17:56:19 <@durandal_1707> i just did it because michaelni started
[18:53] <ubitux> same as well.
[18:54] <ubitux> michael might have copied from the fork
[20:30] <durandal_1707> bunch of so called security checks are flawed
[20:38] <michaelni> durandal_1707, which ?
[20:39] <michaelni> ubitux, "avformat" and such prefixes have the advantage that one can just copy and paste part of the directorx name and filename to the commit message
[20:40] <michaelni> also lavfi and these are typo sesitive, juts one letter flipped and its another lib
[20:41] <michaelni> not sure how much that matters as i seem to manage to make biger typos sometimes ;)
[20:41] <ubitux> that's the curse of typing with fist on a keyboard
[20:46] <Daemon404> tone down the cola then
[20:47] <durandal_1707> its redbull
[20:50] <ubitux> Q: libavcodec/x86/vp8dsp.asm:L1129: "cglobal vp8_idct_add, 3, 3, 0, dst, block, stride"  num_xmm_regs = 0, and then i see some xmm0 usage
[20:50] <ubitux> what am i missing?
[20:52] <nevcairiel> sounds like a bug
[20:52] <ubitux> also, num_regs = 3, how is that defined?
[20:52] <ubitux> i see at least m[012367]; is that unrelated?
[20:53] <nevcairiel> registers is only for general-purpose, while m is mmx/sse registers
[20:54] <ubitux> ok
[20:55] <ubitux> maybe i'm missing something, but i don't see usage of general purpose registrers
[20:55] <ubitux> maybe in some macro, like ret? (ebp/esp/eax ?)
[20:55] <nevcairiel> the three parameters are your registers
[20:55] <nevcairiel> dst, block, stride
[20:56] <ubitux> ah
[20:56] <nevcairiel> they get variable-y looking names :)
[20:56] <nevcairiel> i guess it should set something in the xmm register value
[20:57] <ubitux> in what situation do you have a mismatching number of reg and number of parameter?
[20:57] <nevcairiel> not sure if its not a "bad" bug to skip it
[20:57] <ubitux> (above a certain number of arg i suppose?)
[20:57] <nevcairiel> ubitux: loop variables and stuff like that
[20:57] <nevcairiel> loop counters, i mean
[20:57] <nevcairiel> adds another general purpose register, for example
[20:57] <ubitux> ok i get the point
[20:57] <ubitux> thank you
[20:59] <nevcairiel> BBB can probably tell you if its "ok" to have 0 for the num of xmm regs
[20:59] <ubitux> i'll mail him
[21:01] <nevcairiel> it might only be important for win64 with its callee-saved xmm registers, and only above a certain number of used regs then, but i dunno
[21:23] <durandal_1707> what when avformat fails it does not call close like lavfi does
[21:43] <durandal11707> extradata must be always padded?
[21:43] <cone-282> ffmpeg.git 03Clément BSsch 07release/1.2:a0779a2ee504: avformat/srtdec: skip initial random line breaks.
[21:43] <cone-282> ffmpeg.git 03Clément BSsch 07release/1.2:c09acf988280: avformat/subtitles: add a next line jumper and use it.
[21:43] <cone-282> ffmpeg.git 03Clément BSsch 07release/1.2:0a55c882cb6b: avcodec/srtdec: fix potential overread.
[21:45] <michaelni> if extradata is passed to a bitstream reader than it must have been padded (or would need to be copied but thats inconvenient) so n practice its probably easiest to always pad
[21:46] <durandal11707> padding must always be initialized?
[21:48] <cone-282> ffmpeg.git 03Vignesh Venkatasubramanian 07master:23637f98f4a9: lavf/ogg: Support for end trimming Opus
[21:51] <durandal11707> it may make sense to add function which would allocate extradata + padding and zero padding
[21:51] <ubitux> "lavf" \o/
[21:52] <durandal11707> at even minute use lavf and odd use avformat?
[21:52] <nevcairiel> author decides? :p
[21:55] <durandal11707> why was _O_ff capitalized? someone manually signs off?
[22:02] <cone-282> ffmpeg.git 03Paul B Mahol 07master:c051d668b5db: avformat/avidec: use av_calloc() and check for allocation error
[23:03] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:9078b40d354b: swresample: replace 2 av_free() by av_freep()
[23:03] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:46a723ec7254: swscale/yuv2rgb: make sure yuvTable is set to NULL after deallocation
[23:03] <cone-282> ffmpeg.git 03Michael Niedermayer 07master:b9e7bde6bd84: lavf/avienc: use av_freep() avoid stale pointers
[23:48] <saste> uhm just noticed we lack many flags in automated output since we didn't set the flags
[23:59] <saste> ffmpeg is full of mysteries
[23:59] <durandal_1707> automated output of what?
[00:00] <Compn> which flags ?
[00:00] --- Tue Sep 17 2013