CMake

What's this about?

We're experimenting with the possibility of switching to CMake for building the Second Life viewer.

Why change what we're doing?

Our current build system is unsatisfactory in several respects.

Within Linden Lab, we use different tools on each platform: scons on Linux, Visual Studio 2003 on Windows, and XCode on OS X.

• Any time we add or rename a source file, updating the various build instructions is painful.
• We can't easily stay up to date with newer tools, such as Visual Studio 2005 or 2008.
• Merging changes to the project files used by XCode and Visual Studio is a nightmare.

What does CMake buy us?

CMake has the advantage of generating per-platform build files for us. On Linux, it will generate Makefiles and KDevelop project files. On OS X, it will generate Makefiles and XCode project files. On Windows, it will generate Makefiles (for nmake) and Visual Studio project files.

All of the "smarts" stay in the CMake files, so there's just one authoritative source of build knowledge. This means that people can use the development environment they prefer without having to worry so much about breaking other people's builds. Because CMake files are plain text, merging is easy, as is maintaining experimental patches.

CMake tells your build system how to rebuild its input files when it detects changes to CMake's configuration files. This means that you only need to run cmake once. After that, make or your IDE should keep the CMake files and its own project files in sync for you.

What have we tested?

We've performed test build-and-run cycles on the following platforms.

Linux	Debian sarge	i386	gcc 3.4	prebuilt libraries
Linux	Fedora 8	x86_64	gcc 4.1.2	standalone
Mac OS X	10.5 (Leopard)	i386	Xcode 3.0	prebuilt libraries
Mac OS X	10.5	PowerPC	Xcode 3.0	prebuilt libraries
Windows	XP	i386	VS 2005	prebuilt libraries

Not every platform has been equally tested, and not every feature is fully functional. Please help us to track problems by reporting any trouble you run into.

What are some downsides to CMake?

Please note: none of the following caveats is really serious.

The CMake configuration language is weird and ugly.

The tool has some peculiar inconsistencies: some variables and actions are inherited across files, and others are not. There's no obvious regularity to these, so you have to bump into them one by one.

The documentation is weak. It's mostly organised as reference documentation that assumes that you already know how to use CMake. This is a much bigger problem for new projects who are trying to figure out how to establish themselves than for ones that have already bit the bullet. We've already worked through the early "WTF?" stages, and it was a lot of work.

The good news is that if you have no prior experience, it's much easier to modify existing CMake files than it is to create new ones from scratch.

Performing a build with CMake

First of all, you'll need to download CMake, and install it. On Linux distros, it's usually available as a native package. On Windows and OS X, just use the prebuilt binaries.

To get started, check out a copy of the cmake branch from here.

svn co http://svn.secondlife.com/svn/linden/branches/cmake linden

You'll also need to download the artwork package, and if you're linking against our prebuilt libraries, the archive of those files. You can find links to the appropriate archives in in SVN.

As with the regular viewer build, unpack the artwork and (if you're using it) the prebuilt libraries on top of your SVN checkout.

Configuring your tree

Before you first run a build, you'll need to configure things. There's an indra/cmake.py script that will do this for you. Simply run it from the command line and it will create a reasonably sane default configuration.

In the CMake world, we're keeping source and object files separate. The cmake.py script will create and populate a build directory for you. On Linux, this will be named viewer-linux-i686. On OS X, it will be build-darwin-i686. I don't remember the directory name on Windows; sorry.

What to expect

Running cmake.py does not actually start a build. It just generates the makefiles, project files, or whatever that you'll need. After you're done, you'll have a top-level makefile or project file in your build directory. Run make or load it into your IDE, and away you go!

In principle, your build should run to completion. If you run into any problems, please report them. Better yet, if you can fix them and supply patches, we'd be thrilled! Please follow the usual contribution agreement guidelines.

Where's the built viewer?

The location of the newly built viewer depends on your platform. On Linux, it'll be here:

build-linux-i686/newview/packaged

On OS X, it will be here by default:

build-darwin-i686/newview/RelWithDebInfo/Second Life.app

If you change the kind of build you use, the intermediate directory will also change, e.g. from RelWithDebInfo to Release.

On Windows, the built viewer ought to be able to run from VS2005. Our Windows CMake expert is on vacation, so I can't give you more explicit details.

Prebuilt libraries vs. standalone builds

While many users will want to use the prebuilt libraries that we provide, we're also interested in making life as easy as possible for packagers who want to use their platform's native libraries.

If you run ccmake, you should see a STANDALONE option that determines whether the build will use your system's libraries or our prepackaged ones. Flipping this to ON should be all you need to do to perform a packager-friendly build.

For standalone builds, we'd really like to beef up the checks for system libraries so that for example cmake will fail if a required library (such as OpenJPEG) isn't installed. We welcome all patches that help out with this.

Patching guidelines

We welcome your patches! We can't test on every permutation of platform, compiler, IDE, and libraries, so if you have problems that you can fix, please contribute your fixes and we'll do our best to ensure that you only have to fix problems once.

If you're sending patches in, please follow a few simple guidelines:

• Use regular context diffs.
• Follow the existing coding style in the CMake files. I don't like code shouting at me, so prefer lowercase letters.
• One logical change per patch.
• Use spaces for indentation, not tabs.

I'd like to try an experiment with the development process. Instead of creating JIRA issues, please send your patches to bos at lindenlab dot com, and CC sldev. Inline patches are preferred over attachments, unless your email client trashes white space (as many do).

JIRA issues

• VWR-2871 Implement CMake for building the viewer