xkbcomp errors on keymap returned from xkb_x11_keymap_new_from_device()

[rtcm]

I've recently added support in mutter to get the host X server's keymap with xkb_x11_keymap_new_from_device() when running nested:

https://git.gnome.org/browse/mutter/tree/src/backends/x11/meta-backend-x11.c#n567

mutter then sends this keymap to its wayland clients, including an xwayland instance. When xwayland tries to compile this keymap it errors out.

[rtcm]

The host keymap: https://rtcm.fedorapeople.org/host.xkb
The keymap that we're sending to xwayland: https://rtcm.fedorapeople.org/xwayland.xkb
xkbcomp error messages: https://rtcm.fedorapeople.org/xkbcomp.errors

[bluetech]

Ouch. I was able to reproduce with host.xkb (but strangely, with none of the keymaps I ever tried previously), and 01753c0 should fix it. It'd be nice if you can confirm.

If it works, I'll do a release soon (with some added tests).

[bluetech]

Btw, the reason there are no real xkbcommon-x11 tests running with make check is that such tests would naturally need to communicate with the X server, and use its keymap. So for any sort of useful test, I need a running X server, and more so, to change its keymap. I don't know how to do this in a deterministic way and without possibly trashing the tester's keymap. I guess that's the same reason why libxcb doesn't have any useful tests to speak of...

Any suggestions here (from anyone watching) are welcome.

[rtcm]

This seems to have fixed the issue for mutter. Thanks!

As for tests, I think Xvfb should work.

[fooishbar]

Mmm, we used to have Xfake for this, but starting Xorg with the dummy/void
drivers is really the state of the art here. Requires shipping your own
xorg.conf though - ugh. Would be nice if Xephyr had a mode that didn't
involve it ever actually mapping a window.

On 8 August 2014 15:01, Ran Benita notifications@github.com wrote:

Btw, the reason there are no real xkbcommon-x11 tests running with make
check is that such tests would naturally need to communicate with the X
server, and use its keymap. So for any sort of useful test, I need a
running X server, and more so, to change its keymap. I don't know how to do
this in a deterministic way and without possibly trashing the tester's
keymap. I guess that's the same reason why libxcb doesn't have any useful
tests to speak of...

Any suggestions here (from anyone watching) are welcome.

—
Reply to this email directly or view it on GitHub
#9 (comment)
.

[magcius]

What about Xvfb?

[bluetech]

Thanks for the suggestions. I got Xvfb to work and added a test/stringcomp-like test (it found one additional issue - we didn't handle private actions). The test needs work but I'd like to do a release first.