There is no /dev/ttyv10 on FreeBSD

[lantw44]

In file backend/session/direct-freebsd.c, function setup_tty, it constructs the path to vty with:

snprintf(tty_path, sizeof(tty_path), "/dev/ttyv%d", tty - 1);

It fails when tty - 1 is 10 or greater because there is no /dev/ttyv10 on the system. FreeBSD uses /dev/ttyva and /dev/ttyvb instead.

$ ls -1 /dev/ttyv*
/dev/ttyv0
/dev/ttyv1
/dev/ttyv2
/dev/ttyv3
/dev/ttyv4
/dev/ttyv5
/dev/ttyv6
/dev/ttyv7
/dev/ttyv8
/dev/ttyv9
/dev/ttyva
/dev/ttyvb

This causes sway to fail with 'Cannot open tty: No such file or directory.'

[ascent12]

Patches are welcome. We don't really have a prominent FreeBSD user/maintainer to test things like this.

[valpackett]

Did you launch a bunch of direct sessions simultaneously? :) I've never really used more than one display server, and I use consolekit anyway so my wayfire is on /dev/ttyv0 all the time.

This seems trivial, try changing that line to use %x instead of %d.

[lantw44]

Did you launch a bunch of direct sessions simultaneously? :)

Yes!

/dev/ttyv8 ran GDM (installed from FreeBSD ports) on Xorg.
/dev/ttyv9 ran GNOME (installed from upstream git) on Xorg.

The next vty available for use was /dev/ttyva, but wlroots chose /dev/ttyv10 and failed.

I've never really used more than one display server, and I use consolekit anyway so my wayfire is on /dev/ttyv0 all the time.

I know there is an option called CONSOLEKIT in x11-toolkits/wlroots port, but I don't know how to use it. I was just finding an easy way to run Wayland natively on FreeBSD. I used to run mutter in nested mode (mutter --wayland --nested --no-x11), but I guess applications and toolkits may behave differently if I run them natively.

This bug was found during testing a GTK4 bug: https://gitlab.gnome.org/GNOME/gtk/issues/2201.

This seems trivial, try changing that line to use %x instead of %d.

I already did it on my local installation, but I haven't checked how the kernel will pick the name if the number of consoles is greater than 16.

[lantw44]

This seems trivial, try changing that line to use %x instead of %d.

I already did it on my local installation, but I haven't checked how the kernel will pick the name if the number of consoles is greater than 16.

I increased the number of consoles to 1600 in a VM in order to understand how the kernel chooses names. It seems that it is a number of base 32 with digits 0 ~ 9 and a ~ v. When the number goes up to /dev/ttyvv, the next device is /dev/ttyv10. The same thing happens when the number goes up to /dev/ttyvvv. The next device is /dev/ttyv100.

[lantw44]

I opened #1945 to fix this issue.

[shamazmazum]

Please forgive me for a little offtopic.

@lantw44 With consolekit2 you do not have to mess with permissions of /dev/input/event* devices (like granting RW access for all users) or even set setuid flag on wayland compositor to get input devices working.

More info here, you may find this interesting: https://lists.freebsd.org/pipermail/freebsd-x11/2019-September/024038.html

[lantw44]

@shamazmazum Thanks for the link. It seems that I can't use CONSOLEKIT option without applying patches to ports.