What is a Platform?

Often when looking for apps on Linux, one might search for something “cross-platform”. What does that mean? Typically it refers to running on more than one operating system, e.g. Windows, macOS, and GNU/Linux. But, what are developers really targeting when they target GNU/Linux, since there’s such diverse ecosystem of environments with their own behaviors? Is there really a “Linux Desktop” platform at all?

The Prerequisites

When developing an app for one platform, there are certain elements you can assume are there and able to be relied on. This can be low-level things like the standard library, or user-facing things like the system tray. On Windows you can expect the Windows API or Win32, and on macOS you can expect Cocoa. With GNU/Linux, the only constants are the GNU userspace and the Linux kernel. You can’t assume systemd, GLib, Qt, or any of the other common elements will be there for every system.

What about freedesktop? Even then, not every desktop follows all of the specifications within freedesktop, such as the Secrets API or the system tray. So making assumptions based on targeting freedesktop as a platform will not work out.

To be a true platform, the ability to rely on elements being stable for all users is a must. By this definition, the “Linux Desktop” itself is not a platform, as it does not meet the criteria.

Making Platforms Out of “The Linux Desktop”

It is possible to build fully realized platforms on top of GNU/Linux. The best example of this is elementary OS. Developers targeting elementary OS know that different elements like Granite will be present for all users of elementary OS. They also know elements that won’t be there, such as custom themes or a system tray. Thus, they can make decisions and integrate things with platform specifics in mind. This ability leads to polished, well-integrated apps on the AppCenter and developers need not fear a distro breaking their app.

To get a healthy app development community for GNOME, we need to be able to have the same guarantees. Unfortunately, we don’t have that. Because GNOME is not shipped by upstream, downstreams take the base of GNOME we target and remove or change core elements. This can be the system stylesheet or something even more functional, like Tracker (our file indexer). By doing this, the versions of GNOME that reach users break the functionality or UX in our apps. Nobody can target GNOME if every instance of it can be massively different from another. Just as no one can truly target the “Linux Desktop” due to the differences in each environment.

How do we solve this, then? To start, the community idea of the “Linux Desktop” as a platform needs to be moved past. Once it’s understood that each desktop is target that developers aim for, it will be easier for users to find what apps work best for their environment. That said, we need to have apps for them to find. Improving the development experience for various platforms will help developers in making well-integrated apps. Making sure they can safely make assumptions is fundamental, and I hope that we get there.

Developing GNOME: The Basics

I’ve been working in the GNOME community for a little under a year and a half now. During my time contributing to the project, I’ve seen a lot of misunderstandings from the community about how we work. So, I’ve decided to write a multi-part series on how development on GNOME works.

This first post will cover the basics. In future I’ll explain our tooling, how apps are created, and how changes happen across the ecosystem.

The Release Cycle

At the center of GNOME development is the release cycle. Every 6 months we put out a major release, with patch releases in-between. Major releases typically happen in September and March, and are named after the city of the most recent conference. GNOME 3.30 was named after the city we held GUADEC in, and 3.32 is named after where GNOME.Asia took place.

At different intervals in the cycle we have freeze periods, after which no changes can be made to certain parts of the code. Core apps, such as Web, Files, and Settings all strictly follow the release cycle and freezes. Apps outside of the core set like Polari, Builder, and Fractal can follow their own schedules or versioning schemes, but tend to put out releases alongside the GNOME core.

The freeze deadlines often determine what contributions make it into a release. For example, if a UI change is submitted after the UI freeze, maintainers need to seek approval from the Design and Release teams before the change can be merged. Freezes are staggered, and come in the following order:

  • UI, Feature, and API/ABI Freezes
  • String Freeze
  • Hard Code Freeze

The hard code freeze ends once the major release for the cycle is out. If you want to apply a change that violates the other freezes, you need to create a branch for the latest stable release. So, if I need to merge a UI change after the 3.32 release is out, I need to first create a gnome-3-32 branch before I accept the change onto master. This branch will then be used to cherry-pick changes for the 3.32.X releases.

How Apps Are Maintained

Each project within GNOME has its own developers. The Music developers aren’t necessarily the same people working on the Shell, and the Shell developers generally aren’t the same people working on GTK. While many developers work across the ecosystem on different projects, there is no one team of developers. This is why “GNOME decided such and such” is often inaccurate.

The maintainers of a project have full say over what contributions are or are not accepted. While certain things can be proposed, maintainers have the right to reject proposals. This is, for example, is why Terminal did not have a HeaderBar until 3.32 and doesn’t enable it by default. Nobody is forced to do anything, but often developers and designers will agree on a direction for an app, or the ecosystem at large.

Contrary to popular belief, most maintainers are not paid by Red Hat although some core components like Files and GNOME Shell do have Red Hat employees employed to work on them. Other companies such as Canonical, Purism, and Endless employ developers to work on the parts of the stack that matter to them. That said, most contributors are not on company time even if they are employed by the likes of Red Hat. And of course those that are employed to work on GNOME still aren’t paid for all of their work on GNOME. Most of our work is done on our own time, as limited by the rest of our lives.

It’s also worth noting that GNOME is built with a wide range of technologies; while GTK is written exclusively in C, Music is a Python project and Calculator is implemented with Vala. The Python developers working on Music are great with Python and GStreamer, but they aren’t going to be much help fixing a rounding error in Calculator as a general rule, and as volunteers it wouldn’t be fair to expect them to be, either.

tl;dr: GNOME is a community of individuals each with their own motivations and vision for their own part of the project doing their best to build a great platform for users.