SDDM v0.18.1

We are happy to release a new SDDM version.

SDDM is a Qt based graphical login manager developed in a collaborative fashion by people from Liri, KDE and LXQt.

This time around we made a point release with the following hot fix:

Use C++ scoping for handling buffer deletion - link
SDDM is developed by volunteers, please consider donating money that can be used to support the features that you most desire.

We now have a Bountysource page and I hope that we can get more developers to help us, as we are sometimes struggling to keep a consistent pace with the development.  Having a full time developer, would even be better, but we can start with bounties and see what happens.

Download the source code here.

SDDM v0.17.0

We are happy to release a new SDDM with a few improvements.

SDDM is a Qt based graphical login manager developed in a collaborative fashion by people from Liri, KDE and LXQt.

This new version include the following changes:

* Added possibility to change border width of ComboBox widget.
* Added missing utmp/wtmp/btmp handling.
* Make greeter see icons set by AccountsServices.
* Fix sddm.conf man page syntax error and update.
* Fix ComboBox widget.
* Fix connection of PropertiesChanged event for LogindSeat.
* Avoid race conditions with Xsetup.
* Update de translation.
* Update lt translation.
* Update zh_TW translation.
* Adjust order of components in the default PATH.
* Set default input method to qtvirtualkeyboard.

Download it here.

Qt Virtual Keyboard support on Hawaii

Ever since Qt Virtual Keyboard was introduced a couple of years ago I desired to integrate it with Hawaii, especially considering the effort to make the whole workspace adapt to different devices started a while ago.

However Qt Virtual Keyboard was a value-add component for Qt Enterprise customers, until things changed recently.

As part of the recent licensing change announcement, Qt Virtual Keyboard is now available also with GPLv3 license for Open Source users in addition to commercial Qt licensees.

This move allowed me to finally integrate Qt Virtual Keyboard with Hawaii, which soon will also be available on SDDM.

Qt Virtual Keyboard will be included with Qt 5.7 but it's possible to build the sources from the git repository now, against Qt 5.6.

The virtual keyboard is an input method so activating it is as easy as setting an environment variable, as outlined in the deployment guide:

export QT_IM_MODULE=qtvirtualkeyboard

To make it easier and less error prone a setting for Hawaii has been added that you can enable from the keyboard settings.

On desktop systems the keyboard is resized so that it won't cover search and password text fields.

Obligatory screenshots below:

SDDM 0.12.0 released

We are proud to announce the immediate availability of SDDM 0.12.0.

This release features working Wayland sessions for the first time, but the greeter is still running on Xorg.

This means that you will be able to login into a Wayland desktop environment without problems however the greeter screen still depends on X11.

Work on the Wayland support for the greeter has already been started and will likely be available with the next release.

Highlights of this release:

• BACKWARDS-INCOMPATIBLE: SDDM now depends on Qt >= 5.3
• BACKWARDS-INCOMPATIBLE: Display setup script is now killed if it hasn't finished within 30 seconds
• LICENSING: The Maldives non-free background has been replaced by a CC-BY-SA alternative
• WAYLAND: Wayland sessions support
• CONFIG: New sddm.conf option `XephyrPath` (defaults to `/usr/bin/Xephyr`)
• CONFIG: new sddm.conf option `ServerArguments` (defaults to `-nolisten tcp`)
• CONFIG: new sddm.config option `DisplayStopCommand` (executes Xstop)
• THEMES: Assorted improvements to the default theme
• BUGFIX: Honor TryExec in Xsession desktop files
• BUGFIX: Fix session startup on zsh
• BUILD: Allow setting a custom `DBUS_CONFIG_FILENAME` at compile time
• BUILD: Allow `SYSTEMD_SYSTEM_UNIT_DIR` to be overridden at compile time
• BUILD: Allow `QT_IMPORTS_DIR` to be overridden at compile time
• BUILD: No longer hardcode min/max UIDs, allow setting `UID_MIN` and `UID_MAX` at compile time
• BUILD: Assorted FreeBSD fixes
• BUILD: Fix build warnings with Qt 5.5
• BUILD: Enable `QT_NO_CAST_FROM_ASCII`
• I18N: Added Arabic, contributed by Safa Alfulaij
• I18N: Added Hungarian, contributed by @mortein79
• I18N: Various i18n updates and improvements
• I18N: Full right-to-left languages support

More information can be found here.
Sources and SHA1 hashes follow:

• sddm-0.12.0.tar.gz (SHA1: 9d2b1f7ac3bbbf7c60555b3809179f5c95449e09)
• sddm-0.12.0.tar.xz (SHA1: 985aee2bbafd42969353786cc3dd63b2695cb848)

How to make your Qt application icon appear on Wayland task managers

Several months ago I was redesigning the Hawaii launcher.

Now Green Island offers a class called ApplicationManager that emits signals when an application is registered or unregister (that is when a window of a certain class first appears or when the last one is closed).  This class is also exposed by a Wayland protocol for shells that are running in another process.

Soon I noticed that for some applications the launcher panel didn't show an icon and I realized there was something wrong with the app_id (also known as the class in wl_shell_surface).

The class name identifies the general class of applications and can be used by a task manager to know which desktop entry contains application information.

Unfortunately this is not as easy as it sounds because there is no way to tell that automatically.
So back in the day I had to set the class name to the executable name as you can see here.
This approach worked almost always because desktop entries were usually named after the executable.

However things changed recently, as a matter of fact desktop file names now follow the reverse DNS convention while executables keep being called like before.

This is now achieved by QtWayland with this commit reversing the QCoreApplication::organizationDomain property and appending the executable name. If the property is not set, QtWayland falls back to the old behavior.

The patch is available on Qt 5.4.2 and 5.5.0.

TL;DR: Rename your desktop files using the reverse DNS convention and set the organizationDomain properly.

What's coming to Green Island: part 2

In the previous installment I forgot to post a screenshot of multiple output windows.

Here it is:

The first big chunk of code for multiple output support was recently merged into the qtwayland dev branch, but there is one more patch that is under review now.

Now I'm gonna talk about the recent development on Green Island.

Refactoring

Green Island now has:

• a library that people can use to make their compositors reusing Green Island bits;
• an executable named "greenisland" that links to the library and loads a separate QML based home screen;
• an executable named "greenisland-launcher" that helps you run the compositor in several different environments;
• a QML plugin that provides additional components such as the SurfaceRenderer for QML based home screens;

Launcher

Green Island can run on a number of different setups: X11, Wayland, Raspberry Pi, KMS, you name it.

There is now a launcher that detects where it is running, sets environment variables and runs the actual compositor accordingly.

Plugins

It's now possible to write custom home screens with QML.

The format is very simple but it might change slightly until the next iteration is released:

• the home screen is a directory containing QML files and it's placed under $XDG_DATA_DIRS/greenisland (for example /usr/share/greenisland/org.vendor.coolhomescreen/)
• the directory has a file called Compositor.qml and this is where you connect to the compositor and get events such as when a window is mapped or unmapped and create a QML window representation

xdg-shell

Green Island supports the old wl-shell protocol as well as the new xdg-shell protocol that is now required for weston and Gtk+ clients.

Qt clients can also use xdg-shell but it's not the default, users need to set an environment variable to enable it.

For both protocols Green Island has its implementation which is independent from the one that comes from QtCompositor.

This allows me more freedom and frees me from the qtwayland release cycle.

What's coming to Green Island: part 1

I want to share with you some of the progress I recently made.
This is not all, more will come in another post.

Multi output support: part 1

Green Island now supports multiple outputs.

QtCompositor only support one output so in order to do this I did a little abstraction to enumarate outputs and played with QML to show them in the compositor window.

How does it work?

It's pretty simple: it has a ScreenModel that provides data such as name, primary boolean flag and geometry. The main QML file has a repeater using this model to create an OutputView for each output.

Every OutputView has layers to stack different kind of windows (panels, lock screen, workspaces), one of the layers is HotCorners which contains 8 hot corners.

During test and development one often need to fake outputs to test several different configurations without having real monitors, but this requires a dedicated backend and a configuration format.

So ScreenModel has its own backend code, one of the backends is based on QScreen and the other is for fake outputs.

Studying the problem and doing some prototype made me realize that QScreen has a very limited API (for example it doesn't detect when the primary output changes) and that this matter was already implemented with libkscreen.

It happens to be a framework that doesn't pull in undesired dependencies, so now Green Island is using it and I have to say it saved me a lot of work.

In the video below you can see a Green Island window with two 1024x768 outputs side by side, at some point the first one (which is also primary) is removed and all the windows are migrated to the other output that is now the new primary output.

Multi output support: part 2

A single big fat window for all the outputs is not such a great idea, it was good enough to continue the development keeping multiple outputs in mind but it's not the solution in the long run.

Such a solution may hit hardware limits pretty, plus outputs can have a different refresh rate so they really should not be done this way.

QtCompositor handles only one window so I patched it to be aware of multiple outputs with one window for each of them.
The patch targets the dev branch and at the time of this writing is in the review queue.

All the QML code was reused except for the Repeater and the logic to move windows when an output goes away was moved to C++.
This means that almost none of the code previously wrote was removed.

The hard part came when I needed to figure out how to show the same surface on multiple output windows.

Considering that QQuickItems cannot be shared between multiple windows I had to create a view for each output.

When a shell surface is created for a surface, the compositor creates a view that inherits from QQuickItem, the output is assigned based on the mouse pointer coordinates. No other view is created at this time because the position is calculated to be within output bounds considering the surface size.

More views are created on demand when the surface is mapped.

As windows are moved all views are moved accordingly, global coordinates are mapped to the output coordinates so that windows are shown only where they are meant to be.

Unresponsive applications

Wayland offers a ping/pong API that compositors use to know whether a surface is responsive or not, even with CSD (in the past there was some concern about this).

When a window is clicked, Green Island ping the surface and if it doesn't reply with a pong within 200ms it marks it as unresponsive and apply a colorize effect, draw some explanatory text, and intercepts mouse input. It also adds a couple of push buttons, one to wait for the application to become available again and the other to brutally murder the application.

Hawaii memory usage

Since Hawaii 0.2.0 is closer I wanted to collect some stats about its memory usage.

Measuring resources usage reliably is a bit hard, but it seems that KSysGuard provides detailed results and keep shared libraries into account.

While I'm here I will compare Hawaii with XFCE, although there's no much value in it due to the difference between the two stacks except for being aware of what another low footprint desktop is doing in its current form.

So, let's see how much memory does it take Hawaii, measuring a session with some applications open (System Preferences, Terminal, Eyesight, Swordfish and QupZilla and Communi) after using it for a while.

Currently Hawaii is using Weston although in the future it will use Green Island, a QtQuick compositor using the QtCompositor API.

As you can see from the screenshot below, there are two Weston processes running.

Weston processes

Details below:

weston process memory usage

weston-keyboard process memory usage

Total memory usage: 23.4 MB

Now let's see the desktop environment...

Hawaii Shell memory usage

dconf-service memory usage

dconf-service is included to match XFCE which has xfconfd.

Total memory usage: 64.47 MB

Compare now with a low footprint X11 desktop environment: XFCE.

In the X11 world we need two programs to cover the Weston features: the Xorg server and a window manager/compositor, XFCE uses xfwm4.

Total memory usage: 122.9 MB

From the features point of view we should compare Xorg + xfwm4 with weston (122.9 MB vs 9 MB) but I think this is due to the different architecture.

To be fair I'm measuring only XFCE components that has a Hawaii counterpart, so xfce4-power-manager is not included and judging by the package description xfconfd can be compared to dconf.

The panel-6-systray process is not included either, if it does what the name suggests then Hawaii can't match it because it doesn't support the System Tray Protocol.

xfsettingsd might be left out (although I'm not sure) because Hawaii doesn't have a daemon to propagate settings.  However Hawaii doesn't need it for settings such as icon theme, widgets style, fonts etc... they are managed by a platform theme plugin but other settings such as mouse cursor theme and size would need something to propagate those settings to Qt applications as well as non-Qt applications.

XFCE doesn't have a PolicyKit agent like Hawaii, or at least I couldn't find any.

xfce4-panel memory usage 

xfdesktop memory usage

xfce4-session

panel-2-actions memory usage 

xfconfd memory usage

xfsettingsd

xfce4-notifyd memory usage

Total memory usage: 31.22 MB

Desktop	Usage
Weston	9 MB
Hawaii (weston-keyboard + hawaii-desktop-shell + dconf)	14.4 + 63.7 + 0.773 = 78.83 MB
Total: 9 + 78.83 = 87.87 MB
Xorg and xfwm4	115.7 + 7.2 = 122.9 MB
XFCE (xfce4-panel + xfdesktop + xfce4-session + panel-2-actions + xfconfd + xfsettings + xfce4-notifyd)	9.8 + 6.5 + 4.2 + 2.8 + 0.721 + 2.9 + 4.3 = 31.22 MB
Total: 122.9 + 31.22 = 154.12 MB

EDIT: As requested in the comments I capture the output of free and memuse on my Archlinux system before running Hawaii from KMS (fullscreen 2560x1440) and from the Hawaii session.

Output of free after the system is started: https://gist.github.com/plfiorini/6326618
Output of memuse after the system is started: https://gist.github.com/plfiorini/6326621

Output of free running from Hawaii on KMS: https://gist.github.com/plfiorini/6326633
Output of memuse running from Hawaii on KMS: https://gist.github.com/plfiorini/6326628

Maui

Some years ago I started several projects all based on the Linux kernel and GNU userland with the intent to create a better desktop user experience.
I was not satisfied by GNU/Linux distributions in the day by day use, my ideal system was BeOS which by that time was dead with Be, Inc the company behind it. An Open Source effort to recreate it, known as OpenBeOS (now Haiku), was started in 2001, but it certanly wasn't a viable option because it was in an embryonic state and what was missing from the original BeOS API was scheduled after R1 which as of today it's not reached yet!

BeOS shipped with a state of the art C++ API and a very elegant desktop with many features that later became standards like metadata and queries; it simply was everything Linux wasn't.

Basically I though GNU/Linux on desktop had the following major problems:

• No coherent API and design.
• Low propensity of developers of all desktop environments and toolkits towards a functional, uncomplicated and pleasant, out of the box experience.
• Several software to do the same thing, none of them complete.
• Inability to move to modern graphics architecture because of the dogma of X11 compatibility.

Haiku was progressing, and I kept thinking that it wouldn't go anywhere without a Linux kernel and its better hardware compatibility.  Although Haiku was still years away from completion, in 2003 I started porting its API to Linux.

I made two prototypes, the first was running on X11 with Cairo (which I believe was called Xr/Xc at that time) and the other more successful was based on DirectFB and had not display server (all client side). The Haiku API would have been a good start to implement what was missing from BeOS because of its age (layouts, notifications, standardized actions a-la QAction and many more) without waiting a lot of years.

At some point the enourmous amount of work needed to complete the project and the lack of interest from the potential users threw me into despair. We have to be honest here, BeOS zealots were not willing to use a Linux-based BeOS bastard (they did prefer to wait more that ten years for an early 2000 era operating system almost exactly the same as BeOS), at the same time Linux users was skeptical about it and I certanly did not help missing the second part of the "release early, release often" mantra.

With this project left behind be I decided to take another path: between 2004 and 2005 Qt 4 Technology Preview came out and it was amazing. Clean, complete, modular, supported the MVC pattern, multi-threading and OpenGL. In 2005, the Mockup project was started.
I didn't have much spare time to dedicate on the project, so in the end it was abandoned.

The years passed...

...and suddenly last year, browsing the web, I discovered a nice project: Wayland. Finally there's a viable option that can finally let X11 retire. I started writing some code for a file manager and a panel without the goal to restart the Mockup project.

I was just looking for a challenging pet project and I didn't want to publicly announce something.
Some more month passed, it's 2011 now and I feel it's time for me to write this blog entry because I'm proud of the code I wrote so far - in the last month I wrote a lot of code, my effort and committment is way more than my previous attempts.

So stay tuned to learn more about the Maui operating system.