halting problem :: History of GNOME / Episode 1.5: End of the road

:: ~12 min read

With version 1.2 released, and the Foundation, well, founded, the GNOME community was free to spend its time on planning the next milestones for the project, namely: the last release of the 1.x cycle, and the first of the 2.x one.

GNOME 1.4 was supposed to be a technology preview for some of the changes in the platform that were going to be the centerpiece of the GNOME 2.0 core platform; additionally, applications such as Nautilus, Evolution, and Gnumeric were going to be added to the release as official GNOME components. Maciej Stachowiak volunteered to work as the release manager, and herd all the module maintainers in order to put all the cats in a row for long enough to get distribution-ready tarballs out of them.

From the technological side, one of the long term integration jobs was finally coming to fruition: a specification for interoperation between GNOME and the various window managers available on X11. The work, which was started back in the pre-1.0 beta days by Carsten Haitzler with the goal to make Enlightenment work with the GNOME components, attempted to specify various properties and protocols to be implemented by window managers and toolkits, in order to allow those toolkits to negotiate capabilities and perform operations that involved the window manager, such as making windows full screen, for media players and presentation tools; listing the number of virtual workspaces available, and moving windows across them; defining “struts”, or areas of the screen occupied by special parts of the desktop, such as panels, and that would not be covered when resizing or maximising windows; specifying properties for icons, window titles, and other ancillary information to be displayed by session components such as windows and task lists; creating protocols for embedding windows from different processes, used to implement panel applets, and shared components.

The Extended Window Manager Hints specification, built on top of the original Inter-Client Communication Convention Manual, or ICCCM, was the first attempt at real cross-desktop collaboration, with developers for toolkits, window managers, and desktop environments working together in order to let applications behave somewhat consistently, regardless of the environment in which they were being used. Alongside the EWMH, additional specifications like the XEMBED protocol, the clipboard specification, and the XSETTINGS specification, were created as foundations to the concept of a “free desktop” stack, and we’ll see where that will lead in the GNOME 2 days.

The creation of the EWMH gave a further push towards the establishment of a default window manager for GNOME. Enlightenment had steadily fallen out of favour, given its increase in complexity and its own push towards becoming a full environment in its own right. Additionally, its release planning had become somewhat erratic, with Linux distributions often forced to package development snapshots of dubious quality. While Enlightenment was never considered the default window manager for the GNOME desktop environment, it was the most integrated for a long while, especially compared to the older, or more “free spirited” ones, meant mostly to be used by people building their own environments out of tinfoil, toothpicks, hopes, and prayers. Nevertheless, it became clear that having at least a sensible default, released alongside the rest of the project’s modules and maintained within the same community, and more importantly with the same goals, was necessary. The choice for a window manager fell on Sawmill, a window manager written in a Lisp-like language called “rep”, and originally released around the same time as GNOME 1.2. Sawmill, later renamed Sawfish to avoid a naming collision with a log analysis software for Linux, was, like many Lisp-based projects, programmable by the user; it was possible to write rules to match windows created by specific applications, and apply policies, such as automatic maximisation, or moving to specific workspaces at launch, or saving the geometry and state of the windows when closing them. The window manager decorations were also themable, because of course they were.

Another change introduced for GNOME 1.4 as a technology preview in preparation for GNOME 2.0 was GConf, a configuration storage and notification subsystem written by Havoc Pennington. For the 0.x snapshots and 1.0 releases, GNOME provided a simple key/value configuration system called gnome-config. Each application would use the gnome-config API to store and retrieve their settings, which were saved inside flat, INI-like files in an hidden location under the user’s home directory. The system was fairly simple, but not very reliable, as it could theoretically leave files in an inconsistent state; it wasn’t at all performant, as it needed to traverse hierarchies on the file system to get to a key/value pair; there was no database schema for the keys and values, and it made it impossible to validate the contents of the configuration. Additionally, storing complex values required a fair amount of manual serialisation and deserialisation, so application developers typically ended up writing their own code for storing settings in flat configuration files under their own directory. It was in theory possible to use gnome-config to store and read settings for the desktop itself, but without a schema to define the contents of the settings and without any type of access arbitration there was no guarantee that different system components would end up stomping on each other’s toe, and there was no way for interested components or applications to be notified of changes to the configuration. The limitations of the API led to limitations in the user interface: all settings and preferences were by necessity applied explicitely, to allow saving them to a file, and ensure a consistent state of the underlying storage.

The goal of GConf was to provide a consistent storage for preferences. Values would be addressed via a path, like a file system; read and write operations would go through a session daemon, which was the onl ccomponent responsible for reading and writing the data to disk, and would be able to subscribe listeners to changes to a specific path. This way, applications could listen to system configuration changes and update themselves in response to them; not only that, but utilities could read and write well-known configuration keys, decentralising the configuration of the environment and application from the system settings control panel. Notification and daemon-mediated writes also made it possible to write instant apply preferences dialogs, getting rid of the “Apply” button.

All seemed to be going well for the 1.4 relase, and for the 2.0 planning.

In January 2001, Helix Code renamed itself to Ximian, once it couldn’t secure a trademark on the former name; nothing outside of the name of the company really changed.

In March 2001 Eazel released, at long last, Nautilus 1.0, and simultaneously dropped a bomb on the community: the company was going to lay off most of its employees, after failing to raise enough capital to keep the operations going. Eazel finally closed down in June 2001, after only two years of operations. The Dot-com bubble that propelled the IT industry into unsustainable heights was finally bursting, and everything was crashing down, with the startups that did not manage to get acquired by bigger companies closing left and right, leaving engineers scrambling to get a new job in a bearish market, and leaving cheap hardware, as well as cheap Herman Miller chairs, in the rubbish pile behind their vacant offices.

Eazel published all their remaining code, and its employees promised to remain involved in the maintenance of their projects while the community picked up what they could; which, in fairness, they did, until most of the engineering team moved to Apple’s web browser team.

Nautilus did languish after GNOME 1.4, but as a centerpiece for GNOME 2.0, and with the port to GTK 2.0, it got progressively whittled down to remove the more startuppy bits, while remaining a file manager with first class remote volume browsing; the GNOME community was left with a lot of the technical debt that remained unpaid for the following 15 years, as a result of Eazel’s abrupt end.

The second edition of GUADEC, held in Copenhagen in April 2001, right in the middle of the Eazel drama, nevertheless featured more attendees and more presentations, including one that would prove fundamental in setting the future direction of the project. Sun’s usability engineer Calum Benson presented the results of the first round of user testing on GNOME 1.4, and while the results were encouraging in some areas, they laid bare the limits of the current design approach of a mish-mash of components. If GNOME wanted to be usable by professionals, not curating the offering of the desktop was not a sustainable option any more.

Sun had shipped GNOME 1.4 as a technology preview for Solaris 8, and conducted the first extensive set of user tests on it by sitting professional users in front of GNOME and asking them to perform certain tasks, while recording their actions. The consistency, or lack thereof, of the environment was one of the issues that the testing immediately identified: identical functionality was labelled differently depending on the component; settings like the fonts to be used by the desktop were split across various places; duplication of functionality, mostly for the sake of duplication, was rampant. Case in point: the GNOME panel shipped with not one, not two, but five different clock applets—including, of course, the binary clock, because nothing says “please quickly tell me if it’s time for a very important meeting with my boss” like having to read the answer in binary, from fake leds on panel 48 pixels tall. Additionally, all those clocks, like all the panel applets, could be added and removed by sheer accident, clicking around on the desktop. The panel itself could simply go away, and there was no way to revert to a working state without nuking the settings for the user. The speed of the few animated components, like automatically hiding the panel to increase the available on screen space, could be controlled down to the millisecond, just in case you cared, even if the refresh rate of your display could never have the same resolution. Nautilus had enough settings that they had to be split into a class-based system, with Novice, Intermediate, and Expert levels, as if managing your files required finishing RPG campaigns, and acquiring enough experience points from your dungeon master.

The most important consequence of the usability study, though, was that in order for GNOME to succeed as a desktop environment for everyone, it had to stop being made just for the people that wrote it, or that could spend hours of time tweaking the options, or learning how the environment worked first. Not everyone using GNOME would be a white male geek in their 20s, writing free software. Web designers, hardware engineers, financial analysists, musicians, artists, writers, scientists, school teachers; GNOME 2.0 would need a set of guidelines for applications and for the desktop, to ensure a consistent experience for every user, similar to how the platforms provided by Apple and Microsoft defined a common experience for the operating system.

In 2001, though, the Linux desktop encountered an unexpected problem, in the same way a sentence encounters a full stop. Apple, fresh out of both the return of its co-founder Steve Jobs, the subsequent successful product launches for new lines of personal computers, and after releasing what was nominally a beta in late 2000, unveiled their new operating system, OS X. Sure, it was slow, buggy, and definitely nothing more than an alpha release; but it was a shiny, nice looking desktop environment working on top of a Unix-like user space—precisely what the free software equivalents were trying to achieve. The free desktop community was so worried about Microsoft that it never really considered Apple to be a contender.

September 2001 saw a new point release of OS X, and in just six months from its initial unveiling, Apple had managed to fix most of the instability of the previous snapshot; on top of that, they announced that all new Apple computers starting from January 2002 would be sold with OS X as the default operating system. GNOME 2.0 was released in June 2002, and in August 2002 Apple released the second point release of OS X, which improved performance, hardware support, and added accelerated compositing to the stack. Apple rose back from the ashes of the ‘90s with a new, compelling software platform, and Linux found itself a new, major competitor outside of Windows. A competitor hungry for the same kind of professional and amateur markets that Linux was trying to target; and, unlike Linux, Apple had both a hardware platform and a single driving force, with enough resources to sustain this effort.

The fight for the survival of GNOME, and of the Linux desktop, was back on, and GNOME had to start from 2.0.


We have now reached the point in the story where the GNOME community of volunteers and the ecosystem of companies around them is working on a new major release of the core platform, the desktop, and the applications around it, so it’s a good place to take a break. Writing the History of GNOME is quite intensive: there’s a lot of material that needs to be researched in order to write the 2500 words of an episode, edit them, record them, and edit the recording.

For the next three weeks we’re going to have some short side episodes on technologies and applications relevant to GNOME; the first side episode is going to be about GTK; the second will be about how GNOME wrote language bindings in the early years; and the third episode will be about the first GNOME applications that entered the project.

After these side episodes, we’re going to pause for four weeks, while I gather my notes for the next chapter in our history of the GNOME project; we are going to come back on January 17th, with the story of how GNOME got its groove back with version 2.0, in the next chapter: “Perfection, improved”.

References

history of gnome gnome podcast