Friday, 2015-04-24

*** tpb has joined #timvideos00:00
*** CarlFK has joined #timvideos00:53
*** ChanServ sets mode: +v CarlFK00:53
*** Sewar has quit IRC02:30
*** Sewar has joined #timvideos02:32
*** Sewar has quit IRC03:06
*** Sewar has joined #timvideos03:11
*** Sewar has quit IRC03:16
*** Sewar has joined #timvideos03:25
*** Sewar has quit IRC03:32
*** Sewar has joined #timvideos03:38
*** Sewar has quit IRC03:50
*** Sewar has joined #timvideos03:55
*** Sewar has quit IRC04:12
*** Sewar has joined #timvideos04:19
*** Sewar has quit IRC04:23
*** Niharika has joined #timvideos04:28
*** Sewar has joined #timvideos04:44
*** Sewar has quit IRC05:04
*** Sewar has joined #timvideos05:09
*** sb0_ has joined #timvideos05:43
*** Sewar has quit IRC05:52
*** Sewar has joined #timvideos05:56
*** Sewar has quit IRC06:01
*** Sewar has joined #timvideos06:15
*** Sewar has quit IRC06:20
*** Sewar has joined #timvideos06:22
*** Niharika is now known as Niharika|away06:22
*** Sewar has quit IRC06:30
*** Sewar has joined #timvideos06:36
*** Niharika|away has quit IRC06:36
*** Sewar has quit IRC06:58
*** Sewar has joined #timvideos07:01
*** sb0_ has quit IRC09:00
*** hyades has joined #timvideos09:40
*** hyades has quit IRC11:59
mithroMaZderMind: is the latest version of your code at ?12:29
tpbTitle: [WIP] Rewrite of the Integration-Tests, Take 2 by MaZderMind · Pull Request #198 · timvideos/gst-switch · GitHub (at
MaZderMindmithro: kinda. I'm having issues comparing the Audio-Streams on Travis. My method rendering a spectrum and comparingthe image works locally but stalls on Travis. I'm experimenting with comparing the naked spectral values in
tpbTitle: Comparing · MaZderMind/gst-switch · GitHub (at
mithroMaZderMind: any idea why it stalls?12:33
mithroMaZderMind: do you have a link of the travis output that I can look at?12:34
MaZderMindI guess it has something to do with the numpy comparison of the frames12:34
MaZderMindfirst I had this, what I think was a memory leak in the frame-comparison-pipelin:
tpbTitle: Travis CI - Test and Deploy Your Code with Confidence (at
MaZderMindafter reducinc the memory footprint it started to stall:
tpbTitle: Travis CI - Test and Deploy Your Code with Confidence (at
MaZderMindI then turned up logging which gave me this:
tpbTitle: Travis CI - Test and Deploy Your Code with Confidence (at
mithroMaZderMind: I assume they run fine locally?12:38
MaZderMindthat tells me two things: a) on localhost the spectral images match after ~10 samples, they do not match after 100+ samples on travis, so there must be an issue with the audio coming through12:38
MaZderMindb) after 100+ frames the gst-pipeline stalls instead of throwing a RuntimeException when the 5s delay are over12:38
MaZderMindMy guess for a) is that gst-switch and the audio-source can't run in realtime on travis. while video-frames can be repearted, audio-dsamples usually get dropped, resulting in clicks in the audio which disturbes the spectral comparison12:39
MaZderMindi don't have a guess for b)12:40
MaZderMindmithro: they do run locally on multiple machines/servers12:40
mithroMaZderMind: have you tried saving the audio file and uploading it?12:40
MaZderMindmithro: not yet. it would have uploaded the spectral-images after a timeout exception12:41
MaZderMindbut as it stalls…12:41
mithrohow is the "timeout" implemented?12:41
mithroMaZderMind: btw - I pushed a couple of small patches to gst-switch to deal with a problem around IPv612:42
MaZderMindtoo simple, I guess:
tpbTitle: gst-switch/ at int-tests-rewrite2 · MaZderMind/gst-switch · GitHub (at
MaZderMindas sample = appsink.emit('pull_sample') deadlocks, it never reaches the comparison12:43
MaZderMindI guess i should better use GLib.timeout_add_seconds(timeout, self.quit_mainloop)12:43
MaZderMindor sth. in this direction12:43
MaZderMindfwiw: all tests are rewritten and pass locally, after I sorted this out somehow it's just a question of cleaning up doc-comments and renaming intergration2 to integration in all places.12:48
mithroThe only foolproof method would be to run it inside a separate process using multiprocessing and then actually use kill()12:48
MaZderMindmithro: I thought about launching gst-launch as a separate process and write images using multifilesink12:49
mithroMaZderMind: that process seem like too much work?12:50
MaZderMindhm, guess not. it's a question of getting it as solid as possible12:51
MaZderMindso the basecode must never be touched again ^^12:51
MaZderMindI'll enable travis for my own clone and do some experiments there, I'll also squash the last testing-this/trying-that commits in the PR12:52
MaZderMindto keep the PR clean while experimenting12:52
mithroMaZderMind: okay12:53
mithroMaZderMind: if you force push to the branch the PR comes from, then it'll drop all the old commits and just show the new ones12:53
MaZderMindthat's what I planned to do12:53
mithroMaZderMind: great, thanks for all your work on this12:56
MaZderMind*grml*, i'm really happy when that PR is merged ^^ I was not expecting that to take so long12:57
mithroTesting things is hard :(13:00
MaZderMindbut merging this will close ~10 issues at once13:06
mithroMaZderMind: I have a solution for the recording issue you pointed out14:49
MaZderMindmithro: which recording issue?14:52
tpbTitle: Composite-Output is 30fps, independent of configured format · Issue #200 · timvideos/gst-switch · GitHub (at
MaZderMindahh ok. du you have an idea where this comes from?14:53
CarlFK"foolproof method would be to run it inside a separate process" is that for production or because tests are failing ?14:53
MaZderMindCarlFK: it's because a gstreamer-pipeline in the tests is stalling and thus the test is stalling until killed by travis14:55
CarlFKthat sounds like "because tests are failing"14:56
CarlFKif it isn't for production, I think it is a little too much work.   tests are nice, but so is using it in production14:57
MaZderMindand we're talking about running parts of the test in a separate process14:57
CarlFKand I am trying to discourage that :p14:57
CarlFKI keep trying to use it "for real" and run into blockers14:58
MaZderMindto use 'what' for real?14:58
CarlFK"quotes" cuz it would be like 1 talk user group meetings where complete fail is acceptable14:58
CarlFKmixy (my short name for gst-switch)14:59
CarlFKMaZderMind: have you used
tpbTitle: CarlFK/dvsmon · GitHub (at
MaZderMindI'm not sure if I get your point correct: you'd rather not test the audio pipeline? or test it in a less exact way then to have too complex tests?15:03
MaZderMindCarlFK: no have not seen this tool. Our sources and sinks usually run in runsv and are configured via ansible15:04
CarlFKabove point:  currently mixy is too broken to use in production.  (unless things have changed and no one told me)15:06
CarlFK I would rather not have a bunch of time spent making a test run when that time could be spent getting it working better15:06
CarlFKdvsmon is handy for dev/testing.  it's really just a little terminal manager.15:08
mithroCarlFK: patches welcome15:10
CarlFKim patching behavior :p15:10
mithroCarlFK: Best way to do that is lead by example.15:20
CarlFKnotice I am not working on tests.15:20
CarlFKI am working on packaging ... hopefully that will get some of the debian-video group spending time on it15:21
mithroCarlFK: I don't see any open issues from you about gst-switch being unusable in production?15:27
CarlFKmithro: would you try to use it in production?15:30
CarlFKMaZderMind: if you needed to record something next week, would you use dvswitch?15:41
*** Niharika has joined #timvideos18:05
*** CarlFK has quit IRC18:34
*** Niharika has quit IRC19:32

Generated by 2.13.1 by Marius Gedminas - find it at!