Re: support multiply webcam in chromium-based browsers

On 2025/01/02 19:44, Kirill A. Korinsky wrote:
> On Thu, 02 Jan 2025 19:11:05 +0100,
> Klemens Nanni <kn@openbsd.org> wrote:
> >
> > 31.12.2024 23:11, Kirill A. Korinsky пишет:
> > > Robert,
> > >
> > > I'd like to adjust one patch on all chrome-based browsers in ports tree.
> > >
> > > A few days ago I had commited a fix [1] in uvideo.c which allows user to
> > > select in Firefox between connected webcams.
> > >
> > > Thus, the same can be done in Chrome-based browsers, but it needs small
> > > adjustment in the patch.
> > >
> > > I had used the same limit of number of possible devices with libwebrtc [2],
> > > and updated unveil files to allow /dev/video0 and /dev/video1 which is
> > > created by default.ta
> >
> > Makes sense to unveil all default devices, but trying to open 64 of them like
> > libWebRTC? I guess we'd be fine with probing just the first two, matching unveil.
> >
> > No objection, I can see how matching similar code might make sense, but that
> > limit sounds absurdly high.
> >
>
> On my usual setup I have three devices and use /dev/video0 and /dev/video2,
> and the only way to increase such hard limit, is recompile the chrome which
> requires a lot of space (almost 40Gb) and a lot of times (depends of CPU,
> migth be a days or two).
>
> I may put here 8 or 16 or anything, but I prefer to match some behaviour and
> webrtc looks as good as it can be.

Common situation: laptop has a visible light camera, an infrared camera,
and an camera plugged into a USB port.

Whatever the maximum in the loop, I suggest unveiling at least 4, to
allow the above and one extra, without other tweaks.

> Keep in mind that this code is running only when user visits a page which
> looking for devices. It's very rare, and shouldn't create any issue.

I don't think it's _too_ unreasonable for the code to check higher than
the default unveil to allow unusual cases without a recompile, although
this is probably going to trigger warnings in /etc/daily if accounting
is enabled (the "lastcomm | grep") so perhaps it does make sense to use
a smaller number (8 seems alright to me) and align the loop and unveil.