Liri session does not start

[schnitzeltony]

After going through cmake/linker issues, starting Liri I get a more or less clean log wayland-session.log. That looks way better than few days ago but there is just a 'little' problem: Liri session does not start...

My environment:

• Machine: RaspberryPi3 / graphics open source mesa / VC4. That works perfectly fine for all Desktops I have - particularly Wayland based ones as Weston / Plasma Wayland
• Openembedded build / Image contains the Liri components as of [1]. Don't have https://github.com/lirios/wayland in my images: Opened an issue there and response sounded as if that is going to be deprecated.

I think the point of interest is third line in log file:

No usable crtc/encoder pair for connector "Composite1"

This is created by liri-eglfs / QKmsDevice::createScreenForConnector in case QKmsDevice::crtcForConnector returns -1. And now my problem starts:

• Have not yet found a solution to remote debug this to see why QKmsDevice::crtcForConnector returns -1.
• Looking at QKmsDevice::crtcForConnector is not exactly helpful to me: Seems there are some black magic binary flags coded/checked - have no idea how/where isPossible / isAvailable bits are set

So I would be thankful for any advice how to get more information out of current environment to see why it fails.

[1] https://github.com/schnitzeltony/meta-qt5-extra/blob/master/recipes-liri/packagegroups/liri-world.bb

[schnitzeltony]

Did build qtbase for gles instead of gl (desktop). My idea was egl and GLES are friends so try that but it did not work: liri-eglfs does not build for qtbase build for GLES.

[schnitzeltony]

Party time: After ages Liri is running again on Raspi!!!

What I did:

• Modified session script so that it starts gdbserver for liri-shell
• In the first debug sessions I had strange debug session abortions. gdb complained with: "gdbserver unexpected r_debug version 0". Some investigation lead to LIRI_SHELL_DEVELOPMENT_BUILD.
• Next debug session with LIRI_SHELL_DEVELOPMENT_BUILD [1] enabled segfaulted in QEglFSKmsGbmScreen::recordFrame: Variable funcs was null and that is no suprise: Raspi's VC4 does support 2.0 GL(ES) only - 3.0 is too much for VC4 :) But what made me wonder was that compositor made it to it's first page flip and all variables seemed to contain reasonable data.
• Created a patch to disable recording by default [2].

With this liri starts up.

I think it needs both modifications - have the feeling that shell's disablePtrace is not working for me.

Aah and next question: Have no icons - so what is the suggested icon-theme for liri?

[1] schnitzeltony/meta-qt5-extra@32f6933
[2] schnitzeltony/meta-qt5-extra@3bde996

[plfiorini]

Can I see the modified liri-session script?

disablePtrace() disable ptrace (unless LIRI_SHELL_DEVELOPMENT_BUILD is set to ON), but it still allows you to run through gdb, using gdbserver is not allowed.

But as said enabling LIRI_SHELL_DEVELOPMENT_BUILD should bypass disablePtrace() and you'll be fine.

[schnitzeltony]

Was something like:

[Desktop Entry]
Comment=QtQuick and Wayland based desktop environment
Exec=gdbserver :5000 /usr/bin/liri-shell -platform liri
Icon=start-here
Name=Liri
TryExec=/usr/bin/liri-session
Type=Application