#v4l 2019-09-10,Tue

↑back Search ←Prev date Next date→ Show only urls(Click on time to select a line by its url)

WhoWhatWhen
flingHow to workarond this? -> https://askubuntu.com/questions/1171192/multiple-dev-video-devices-for-1-camera-causing-issue-with-udev-rules [07:21]
hverkuilfling: the only way I can think of is to check the v4l2-ctl -D output and skip the video device if it is for metadata. [07:26]
flingthe problem is when I pass the "proper" devices to obs-studio it does not recognize any [07:27]
hverkuilI think we made a mistake when we decided (after long discussions) to keep /dev/videoX for metadata. [07:27]
flingbut works with single camerwa
single device, 2 nodes, obs works
two devices, only keep two nodes of four obs borks
two devices, keep all the nodes, obs borks
[07:27]
hverkuildon't know obs. No idea what it is doing. [07:28]
flingffmpeg and gstreamer are not working for me because software timestamps are broken for few initial frames
so it breaks live output when muxing with pulse
hverkuil: can't I tell it to use the previous behaviour via a module parameter?
I don't want to go v4l2loopback path :D
[07:28]
hverkuilcan't tell what to use which previous behavior?
(have to go, back in 30 min)
[07:34]
flinghttps://github.com/torvalds/linux/commit/088ead25524583e2200aa99111bea2f66a86545a ?
cya
[07:35]
ok
obs-studio does not like the metadata devices
obs-studo only sees one camera probably because it wants the device node names to be sequential
ok, works in lxd where I only share the proper device nodes thanks
only works in the selection screen. Can only capture from one at a time
I remember seeing the same thing in ffmpeg last time I tried
[07:40]
I could run second obs in qemu probably
but hardware is not fast enough for reincoding…
eee
[07:51]
hverkuil"obs-studo only sees one camera probably because it wants the device node names to be sequential": that's a bug in obs: there is no guarantee whatsoever that video device names are sequential.
"obs-studio does not like the metadata devices": obs should check using VIDIOC_QUERYCAP that this device is a video capture device: besides metadata capture devices there are also m2m devices, video output devices and multiplanar variants.
No, there is no way to go back to the old behavior.
But all this sounds like bugs in obs.
[08:03]
........ (idle for 35mn)
flingok
anyway for some reason can only capture from one device at a time
I need to debug this
Which distro has the best v4l support with all the things patched to the working state with h264 devices and everything?
[08:41]
........................ (idle for 1h59mn)
hverkuilpH5: those two patches look good. Nice clarifications. [10:41]
............. (idle for 1h4mn)
***al has quit IRC (Ping timeout: 264 seconds) [11:45]
............................... (idle for 2h30mn)
bparrothverkuil, ping [14:15]
hverkuilbparrot: pong [14:15]
bparrothverkuil, hey i ran v4l2-compliance on my patch set, and i am getting an error but i don not understand what it means, see https://pastebin.ubuntu.com/p/PPHnYHGKZT/ [14:18]
hverkuilIt looks like your sensor reports various discrete framesizes, and you try to support the selection API. That's not supported by V4L2. [14:21]
bparrotso you mean that we can't support "cropping" wiith sensor which support multiple discrete resolution? [14:23]
hverkuilThe problem is that S_FMT is used to select which discrete framesize should be used. But if you combine that with cropping then that falls apart.
It's an API limitation.
[14:25]
bparrothow? i mean you use s_fmt to choose a specific resolution and use selection to grab a portion of that, it seems a valid setup to me. And i am pretty sure it used to work
Not sure how to handle this, at this point. Does that mean i have to disable "selection" support?
was this always an API limitations?
[14:27]
hverkuilActually, calling S_FMT is ambiguous: it can pick a specific resolution and reset the selection rectangle, or it can keep the current resolution and change the crop rectangle to make the new format. The V4L2 API never defined what should happen.
IMHO we are missing an ioctl to explicitly select which discrete framesize should be used.
I think I tried to get it in, but never succeeded.
[14:33]
bparrothmm i thought these were ordered call
i mean you had to use s_fmt and then call s_selection after. At least that's how i have using this
[14:34]
hverkuilThe basic rule is that the driver should attempt to fulfill the called ioctl as best as it can, and this might mean changing selection rectangles or formats.
I suspect most drivers in this situation ignore the compliance error and just do what you want to do.
It's tested with a sensor?
Or is this a composite video input?
[14:36]
bparrothmm, so s_selection api can only be use if you only have a single discrete resolution or support continuous?
i have always tested with a sensor in this case an ov2659
alright so i won't attempt to fix this compliance issue and i'll just post that same result as part of the patch series. Is that ok?
[14:38]
hverkuilYes. Or it isn't a sensor but a SDTV or HDTV video input instead. [14:40]
bparrotin theory we could have an encoder like one of the TVP514x device hooked up to this also, but i don't not such hw setup so i cannot test it [14:42]
hverkuilYeah, just post the same results. [14:43]
bparrotalright will do. thanks [14:44]
hverkuilI've always been unhappy about this corner case. [14:45]
bparrotbtw, i will have other patch set to other driver coming in the next few weeks.
Is ok to sent them in one series per drivers like i did for this one, Or do you prefer i split them up further
[14:46]
hverkuilNo, this is fine.
got to go, back later.
[14:48]
bparrotok bye [14:48]
............. (idle for 1h2mn)
***benjiG has left [15:50]

↑back Search ←Prev date Next date→ Show only urls(Click on time to select a line by its url)