[Ffmpeg-devel-irc] ffmpeg-devel.log.20190413

[burek]

[00:05:00 CEST] <mkver> Hello all. I have a question regarding inactive maintainers: What happens with patches to components whose maintainers are inactive?
[00:08:37 CEST] <BBB> someone else will eventually review them
[00:09:43 CEST] <mkver> Ok, then I'll simply ping the patchset. Thanks.
[08:10:25 CEST] <mkver> If an error happened during writing a packet during muxing, is the user then actually allowed to simply try to mux another packet or is he supposed to end muxing altogether and write the trailer? I presume the former, but I am not sure.
[08:14:10 CEST] <mkver> I'm asking because of the webm_chunk muxer. This muxer uses a Matroska/Webm muxer internally and if an error happens during writing the packet, the whole Matroska/Webm mux context gets freed. In this case a use-after-free (and also double free) also happens during writing the trailer. And depending upon the answer to my question, a use-after-free would also happen when one writes another packet after an error occurred during a 
[08:14:10 CEST] <mkver> packet.
[08:15:05 CEST] <mkver> during writing a packet.
[10:47:23 CEST] <nevcairiel> even if a call is technically wrong, it should not result in a use-after-free, so checks are needed regardless
[12:57:44 CEST] <atomnuker> "The string codecs="av01.0.01M.08" in this case would represent AV1 Main Profile, level 2.1, Main tier, 8-bit content with 4:2:0 chroma subsampling, ITU-R BT.709 color primaries, transfer characteristics, matrix coefficients, and studio swing representation."
[12:58:14 CEST] <atomnuker> why the cryptic codec name? that data is available elsewhere
[12:59:08 CEST] <atomnuker> not like hardware decoders care about colorspace or range
[13:02:10 CEST] <nevcairiel> the last part isnt even signaled in t he string, those are the defaults
[13:02:20 CEST] <nevcairiel> but it can be
[13:02:35 CEST] <nevcairiel> it may want to know if something is hdr vs not or such
[13:09:08 CEST] <atomnuker> av01.0.05M.08, so youtube signal it
[13:45:04 CEST] <nevcairiel> a full string would be something like "av01.0.04M.10.0.112.09.16.09.0" .. yours all end after the bitdepth, so no color info
[13:51:06 CEST] <atomnuker> oic, still very verbose
[15:03:53 CEST] <cehoyos> jkqxz: There is a GSoC proposal for an OpenCL Deshake Filte, you are listed as a possible mentor on trac - are you (still) interested?
[15:05:38 CEST] <BtbN> Anyone happen to know if my Visual Studio 2017 Enterprise License will activate 2019?
[15:06:03 CEST] <BtbN> I highly suspect it won't, but the Installer is offering me the update.
[15:10:04 CEST] <BtbN> btw. regarding opencl filters. I guess we are no longer integrating them with the existing native filters?
[15:10:18 CEST] <BtbN> Cause that OpenCL version of the colorkey filter had me wondering.
[15:10:42 CEST] <JEEB> BtbN: my first guess would be not to activate although I don't think I've noticed details being written about that anywhere
[15:10:54 CEST] Action: JEEB has mostly moved onto the free SDKs after leaving university
[15:11:21 CEST] <BtbN> I'm still at a university, and will be for quite a while, since I work there.
[15:11:44 CEST] <BtbN> But Microsoft rebranded MSDNAA/Dreamspark/Imagine/... once again, and has so far not supplied anyone with info on how to get access.
[15:12:10 CEST] <JEEB> oh, they changed that again
[15:13:04 CEST] <BtbN> "Azure Dev Tools for Education" is what it's called now
[15:13:47 CEST] <BtbN> "Microsoft has so far failed to provide us with information on how to gain access."
[15:13:55 CEST] <BtbN> That's my Unis statement about it
[15:14:21 CEST] <BtbN> They're a bit pissed, because they did not fail to send the next bill
[16:39:56 CEST] <durandal_1707> https://github.com/losnoco/vgmstream/issues/395
[16:55:27 CEST] <HickHoward> "i'm da badass here"
[16:57:36 CEST] <durandal_1707> look at https://github.com/losnoco/vgmstream/issues/394 for explanations
[17:01:57 CEST] <HickHoward> huh...
[17:02:22 CEST] <HickHoward> guy's a possible terrorist by this point
[17:47:05 CEST] <durandal_1707> why hevc-two-first-slice is still failing?
[17:52:54 CEST] <jamrial> because whoever runs those fate clients hasn't synced the sample
[17:53:30 CEST] <durandal_1707> who is responsible for it?
[17:54:19 CEST] <jamrial> "mik"
[17:54:46 CEST] <jamrial> i worry more about how no slot seems to have run in two days
[17:55:22 CEST] <jamrial> michaelni: is there something wrong with fate? it's been two days since the last submitted result
[17:56:42 CEST] <durandal_1707> ffmpeg got to big for fate
[17:56:49 CEST] <jamrial> oh, nevermind, it's just that no commits were made since then
[17:57:17 CEST] <jamrial> fate just discards new results if it's for the same commit as the last one
[17:58:07 CEST] <durandal_1707> even if it becomes green?
[17:58:37 CEST] <jamrial> no idea
[18:02:14 CEST] <cehoyos> Did anybody ever look at "warning: implicit conversion from 'int' to 'int16_t' (aka 'short') changes value from 44130 to -21406" when compiling fspp with clang?
[18:02:49 CEST] <durandal_1707> i ignore darkmagic filters
[18:03:19 CEST] <cehoyos> I thought you wrote it;-)
[18:04:15 CEST] <JEEB> could be valid depending on the values passed
[18:04:51 CEST] <JEEB> or technically: is valid, but whether it affects things depends on the values passed
[18:09:09 CEST] <durandal_1707> trac is unresponsive again
[18:15:28 CEST] <cone-188> ffmpeg 03Paul B Mahol 07master:16f7c1f2911c: avcodec: add LSCR decoder
[18:18:41 CEST] <JEEB> meanwhile brr, I really need to create that sample for that one sub2video FATE test that I wanted to add :V
[18:18:52 CEST] <JEEB> that one sub2video bugfix has been sitting on the ML for a while
[18:19:02 CEST] <JEEB> (use case that michael noticed)
[18:19:31 CEST] <michaelni> jamrial, one of my fate client VMs died, one has a clock thats a day off, iam checking the others dunno if thats the issue
[18:20:47 CEST] <jamrial> michaelni: no, it isn't. i noticed after asking you that there have been no new commits since then
[18:23:42 CEST] <jkqxz> cehoyos:  Yes, I've been talking to that student.
[18:24:03 CEST] <cehoyos> michaelni: Does this clang warning have any real-life effect: "warning: implicit conversion from 'int' to 'int16_t' (aka 'short') changes value from 44130 to -21406"
[18:24:08 CEST] <cehoyos> This is when compiling fspp
[18:24:23 CEST] <cehoyos> jkqxz: Please accept the mentor invitation by Google.
[18:25:40 CEST] <jkqxz> Hmm.  I'm trying to log in with my account from last year, but it says I don't have one.  Do I need to sign up again?
[18:25:59 CEST] <cehoyos> Did you not receive an invitation?
[18:26:31 CEST] <jkqxz> Where?
[18:26:45 CEST] <cehoyos> sw at jkqxz.net
[18:27:49 CEST] <cehoyos> Or to say it differently: Where do you want it?
[18:28:05 CEST] <jkqxz> Ah, I do have an email.
[18:28:11 CEST] <cehoyos> ;-)
[18:28:32 CEST] <cehoyos> Do you know how to setup Linux so that you don't have to login locally to enable remote vaapi?
[18:28:43 CEST] <cehoyos> remote: via ssh
[18:29:35 CEST] <jkqxz> Use a render node?
[18:30:17 CEST] <cehoyos> Sorry?
[18:30:17 CEST] <jkqxz> Does it fail because you have X forwarding enabled and it tried to connect via X?
[18:30:46 CEST] <cehoyos> It doesn't fail after I logged in on the system where I want to test vaapi
[18:30:50 CEST] <cehoyos> Even via ssh
[18:31:11 CEST] <jkqxz> Huh.  What is the error?
[18:31:42 CEST] <cehoyos> The error was: No VA display found for device: /dev/dri/renderD128.
[18:33:01 CEST] <jkqxz> Can you strace it to find out what fails?  That's a pretty generic top-level error because you don't get much feedback from libva.
[18:37:09 CEST] <cone-188> ffmpeg 03Paul B Mahol 07master:a208c277f811: avcodec/imm4: flush prev_frame after seeking
[18:40:34 CEST] <cehoyos> openat(AT_FDCWD, "/dev/dri/renderD128", O_RDWR) = -1 EACCES (Permission denied)
[18:41:46 CEST] <nevcairiel> there may be a group your user needs to be in to access that
[18:42:15 CEST] <jkqxz> Most distributions make that accessible to a group called "video".  Is the user you are running as in that group when you run it?
[18:43:18 CEST] <cehoyos> I see the same error after adding my user to that group
[18:43:50 CEST] <cehoyos> Works after logoff-login
[18:44:09 CEST] <cehoyos> I wonder why it worked locally and via ssh if I was logged in locally
[18:44:21 CEST] <JEEB> something could have made the file more accessible
[18:44:29 CEST] <JEEB> I just checked the rights on my fedora while logged in
[18:44:32 CEST] <JEEB> http://up-cat.net/p/d5e059e7
[18:44:48 CEST] <JEEB> which seems to be rather free for the render end point?
[18:44:52 CEST] <cehoyos> Anyway jamrial: Neither vaapi nor nv decode the sample(s) bit-exact, only old FFmpeg did
[18:45:25 CEST] <cehoyos> The third gop of the second sample is still different with your patch.
[18:47:26 CEST] <jkqxz> It's possible to have different sets of groups when logged in by different methods.  When running as the user, try "id".
[18:48:32 CEST] <cehoyos> Thank you!
[18:49:21 CEST] <cehoyos> Feel free to click "Want to mentor" for the deshake project.
[18:50:31 CEST] <jkqxz> Looking at it now.
[19:04:31 CEST] <michaelni> cehoyos, the cast looks correct
[19:05:39 CEST] <JEEB> that warning says implicit, so there is no explicit cast there
[19:05:48 CEST] <JEEB> unless clang is missing it
[19:06:09 CEST] <michaelni> theres a explicit cast in the line and a implicit one
[19:06:18 CEST] <JEEB> ah
[00:00:00 CEST] --- Sun Apr 14 2019