Build the Viewer on Linux
The following are instructions for building the Second Life viewer on linux. This process has been used on debian and debian based systems like ubuntu, and also on Fedora. For other platforms, see Get source and compile.
Required tools
There are a number of tools that need to be installed first.
- cmake [package: cmake]
- Currently (snowglobe 1.3.2 and higher) cmake 2.6.2 is the minimum required version.
- yacc or compatible tool [suggested package: bison]
- lex or compatible tool [suggested package: flex]
- python [package: python]
- python 2.4 was suggested as the best choice
- python 2.6 seems to be fine as well
- python 2.3 or before doesn't work in SL development
- python 2.5 or later works, but it is (was?) not recommended, since it generates a lot of warning messages when running develop.py (still true as of Snowglobe2?)
- g++ [package: g++]
- Note (not relevant for standalone): gcc 4.4 (which is in recent Ubuntu and debian) won't work in versions prior to 2.0 (or Snowglobe 1.3) if you are not building standalone (see below), because it chokes on some parts of boost prior to 1.37 (http://svn.boost.org/trac/boost/ticket/2069). The solution is to install GCC 4.3 and to run 'export CXX=/usr/bin/g++-4.3' or whatever your binary is before trying to compile. Another workaround is to use update-alternatives; for example, if you already have 4.4 installed, try this:
sudo apt-get install g++-4.3 sudo update-alternatives --remove-all gcc sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.3 43 --slave /usr/bin/g++ g++ /usr/bin/g++-4.3 --slave /usr/bin/gcov gcov /usr/bin/gcov-4.3 sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.4 44 --slave /usr/bin/g++ g++ /usr/bin/g++-4.4 --slave /usr/bin/gcov gcov /usr/bin/gcov-4.4
then choose 4.3:
sudo update-alternatives --config gcc
- libboost-program-options-dev was needed on Ubuntu 8.04 to use cmake, but it's not needed anymore to build a 1.23-render-pipeline on Ubuntu 9.04-beta (nor Snowglobe 2 on Ubuntu 9.10)
The build process may use the following optional tool:
- distcc distributed compiler (useful if you have multiple PCs.)
What does 'Standalone' mean?
A standalone build of the viewer refers to building a viewer against the shared libraries that are installed on your system instead of using precompiled libraries provided by Linden Lab.
In order to build standalone, you will have to configure using --standalone
, see configuration below.
The advantage of building standalone is that you might use less RAM: you'll be using the same shared libraries that other applications use. However this is a rather minor advantage. The disadvantages of building standalone is a long list of potential problems: you will have to manually install all the development packages of many many libraries. In some cases there isn't a package for it from your distribution, so you have to especially get those sources separately and compile and install them in a way that the viewer can find them. Building standalone is not supported officially by Linden Lab and therefore not tested. Hence, it might not work at all (although the open source snowglobe developers usually make sure that it works for snowglobe). Nevertheless, every now and then it happens that the viewer won't build with some new released library (ie, boost), in which case you most likely will have to use the latest source code retrieved with subversion.
Nevertheless, there might be reasons that you want to do a standalone build anyway. The most important one being that Linden Lab also doesn't support 64-bit: they do not provided shared libraries for 64-bit; so, if you want to build native 64 bit you must use standalone. Another reason might be that Linden Lab is using a few rather old library versions that are incompatible with newer versions. If you have any reason to use a newer version for some library, then chances are that things will break unless you build standalone. One reason would be if you want to debug a library and fix it because it is buggy.
Conclusion: if you are building the viewer for the first time, and you are building for 32-bit, then you should not use standalone.
Getting the source
Releases and release candidates are available as source archives (zip or tar) from Source downloads, or you can get the latest (experimental) sources with Subversion.
Using a source archive is safer and strongly recommended for first-time builders. However, if you plan to do active development and want to make your own modifications, while merging Linden Lab's changes, or if you want to live on the bleeding edge, then Subversion should be your choice.
Using the source archives
Get the source and artwork from Source downloads. You probably won't need to get the libraries because they are downloaded automatically with the newer releases (both 2.x and SG 1.x). However, if you are building version 1.23 or earlier you will also need to get the libraries archive (assuming you are not going to build standalone).
The sources and art work are extracted into a directory called linden.
For example,
% tar xzf snowglobe-src-viewer-<version>.tar.gz
% unzip slviewer-artwork-oss-viewer-<version>.zip
Using subversion
Select a source branch to check out.
Check it out with:
% svn checkout http://svn.secondlife.com/svn/linden/<branch> <directory>
If you don't specify the directory, it'll use the last part of the branch path by default.
Now you will have a directory with the bare source code.
You will have to get the art work from the Source downloads and unpack them into the tree (note that Viewer 2.x and Snowglobe automatically download the libraries; you will only need to get the artwork).
If the directory SVN created is called 'linden' then you can directly unzip the archive into it.
If it's called something else, create a symlink and then unpack the art work:
% ln -sf <directory> linden
% unzip slviewer-artwork-oss-viewer-<version>.zip
Installing the required libraries (that Linden Lab can not or does not provide)
Even non-standalone still uses a few shared libraries from your system. This paragraph deals with those libraries. For standalone you will need a lot more (see later on).
Libraries and header files that usually come with a Linux distribution
Make sure the libraries and header files for the following packages are installed on your system:
Library | Debian/Ubuntu | Fedora/Red Hat |
---|---|---|
libGL.so | nvidia-glx-dev or mesa-common-dev | mesa-libGL-devel |
libGLU.so | libglu1-mesa-dev | mesa-libGLU-devel |
libc.so | libc6-dev | glibc-devel |
libstdc++.so.6 | libstdc++6 | libstdc++-devel |
libX11.so | libx11-dev | libX11-devel |
libz.so | zlib1g-dev | zlib-devel |
libssl.so | libssl-dev | openssl-devel |
libXrender.so (for viewer 2.0) | libxrender-dev | ? |
shortcut commands for the above
Debian/Ubuntu:
sudo apt-get install mesa-common-dev libglu1-mesa-dev libc6-dev libstdc++6 libx11-dev zlib1g-dev libssl-dev libxrender-dev
Fedora:
sudo yum install mesa-libGL-devel mesa-libGLU-devel glibc-devel libstdc++-devel libX11-devel zlib-devel openssl-devel
Necessary libraries when building standalone
If you want to build 'standalone', then you will need the follow packages:
Installing the required dependencies (prepackaged by Linden Lab)
This step has become obsolete for viewer 2.x and snowglobe 1.x and higher.
Obviously, you can also skip it if you are building standalone.
The Second Life Viewer has a number of compile/link dependencies on external libraries which are needed - to help you, the source download page contains a link to a slviewer-linux-libs package which you unpack over the source tree to fill most of the dependencies (and thus avoid most of the fiddly work described on this page).
If you download the libs to the top folder, where the linden folder is after getting and extracting the viewer source code tarball, the following command should unpack everything to the right spot.
tar xvfz slviewer-linux-libs-<version>.tar.gz
Note: as of 1.21, very little of what once was included in the libraries tarball are still there, so this is a very small tarball. The other libraries will be downloaded as part of the build process.
If you don't wish to use the precompiled libraries, see Building the viewer libraries (Linux).
Compiling with CMake
This page describes how to build the Second Life viewer with CMake. CMake is a system for generating per-platform build files. On Linux, it will generate your choice of Makefiles or KDevelop project files.
NOTE: These instructions are for the viewers using cmake (versions 1.21 and beyond). For older viewers (1.20 and earlier) see Compiling the viewer with SCons (Linux)
Configuring your tree
Before you first run a build, you'll need to configure things. There's a develop.py
script that will create a reasonably sane default configuration for you.
From the command line, cd into the indra
subdirectory,
% cd indra
Next you can configure the project by running,
% ./develop.py [-m64|-m32] [--standalone] [--type=Debug] configure [-D...]
Note that -m32
is the default even on 64bit. In order to build for 64-bit you must specify -m64 --standalone
. Any cmake defines (starting with -D, see the notes directly below) must be specified after (to the right of) the 'configure' keyword.
Note: To build for 'standalone' (see `What does 'standalone' mean?` above) add --standalone
to the configure commandline.
Note: If you want to use KDevelop add -G KDevelop3
to the configure commandline.
Note: Configuring a "non-standalone" version of the source code tree will cause the required (32-bit) third party library packages (as built by Linden Lab) to be downloaded during the CMake build process.
Note: The environment variables CC and CXX are picked up automatically as usual. However, this is not the case of CXXFLAGS and LDFLAGS. If you are compiling Snowglobe then you can pass the following to the configure commandline: -DCMAKE_CXX_FLAGS:STRING="$CXXFLAGS" -DCMAKE_EXE_LINKER_FLAGS:STRING="$LDFLAGS"
to enable the use of these environment variables.
Note: You can make the build process more verbose (make it print the compiler commands it executes) by passing -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
as well.
Note: For debugging purposes you can build without optimization and with debugging symbols by specifying --type=Debug
.
Note: By default the build process runs a testsuite at the end. You can turn this off by passing -DLL_TESTS:BOOL=FALSE
.
If you want to redo this step, erase the CMakeCache.txt
file in the build directory and rerun develop.py.
Starting the build
To start a build, do one of the following:
- Use the same command that you used to configure (but minus the -D... options), but replace 'configure' with 'build'.
For example: ./develop.py -m64 --standalone --type=Debug build
-- or --
- Find your build directory and change to it
- In the CMake world, we keep source and object files separate. The
develop.py
script will create and populate a build directory for you. On Linux, this will be namedviewer-linux-ARCH-BUILD
(where "ARCH" is "i686" or "x86_64", and "BUILD" varies on debug level)
- In the CMake world, we keep source and object files separate. The
- Build with your preferred build tool
- Type
make
- -- or---
- Load it into KDevelop
- Type
Where's the built viewer?
On Linux, your build will be here:
viewer-linux-ARCH-BUILD/newview/packaged
...where "ARCH" is something like "i686" or "x86_64" (depending on your platform), and "BUILD" depends on which debug level you chose.
Using ccache
It is very highly recommended that you use and install ccache even before the first compile; it will speed up subsequent compiles with a factor of 10 in case you need to redo the compile!
The easiest way to start using ccache is to make symbolic links to ccache
from /usr/local/bin
. Make sure that /usr/local/bin
comes first in your PATH
,
before the real g++. Thus, as root and assuming you are using gcc/g++,
% apt-get install ccache
% cd /usr/local/bin
% ln -s ../../bin/ccache gcc
% ln -s ../../bin/ccache g++
And then to test, as normal user (make sure you are not in /usr/local/bin anymore),
% which g++
This should print /usr/local/bin/g++
, if not fix you PATH
.
Note that by default ccache puts the cache in $HOME/.ccache
and will grow till about 1 GB, so make sure you have that diskspace there. Alternatively you can set the environment variable CCACHE_DIR
to change the location of the cache.
Using distcc
If you have multiple PCs, you can speed up the build process by using distcc.
To use distcc, you need to pass an environment variable CXX
containing the distcc command prefix to develop.py
when configuring, e.g.,
CXX="distcc g++" ./develop.py configure
Or, if you are using a heterogeneous rows of PCs, you need to add a config prefix to g++
, e.g. (in my case, as a Fedora user),
CXX="distcc i386-redhat-linux-g++" ./develop.py configure
Once you did so, the generated makefiles contain instructions to use distcc. You don't need to specify anything special when invoking develop.py
to build. The number of concurrent jobs (i.e., -j
option to make
command) to run is automatically determined by develop.py
.
If you invoke make
command by yourself, don't forget to add the -j
option with an appropriate number.
Prebuilt libraries vs. standalone builds
Everything below is probably outdated. It should be carefully sorted, cleaned up and updated (added 7/7/2010)
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.
- cmake-SL: a script for standalone, pre-built and mixed libraries viewer builds under Linux.
- Problems when compiling a standalone Linux viewer
Testing and packaging the client
Testing the result from inside the tree
You may find it simpler and less error-prone to follow the instructions in the Packaging the client section below to run the client under the same conditions as an end-user would.
- 2008-05-29 (Ochi Wolfe): Compiling the 1.20.7 r88152 viewer, it seems like even when compiling as "release" the viewer is built ready-to-go inside the newview/packaged/ directory including the message_template.msg and message.xml in the right place. Try to cd to the newview/packaged/ directory and run SL from there with the ./secondlife command as you would normally do.
Otherwise:
- Preparing to run 'in-tree'
- ensure that you have indra/newview/app_settings/static_*.db2 - if not, you'll find it in the 'slviewer-artwork' download (a zip file).
- now, from the indra directory:
$ cp ../scripts/messages/message_template.msg newview/app_settings/
$ cp ../etc/message.xml newview/app_settings/
Important: Starting from version 1.18.0, copying message.xml is also required. Missing it will cause group IMs to fail to work, although the viewer will run fine otherwise.
- Running it: The LD_LIBRARY_PATH stuff ensures that the binary looks for its libraries in the right places. From the indra directory:
$ ( cd newview && LD_LIBRARY_PATH="`pwd`"/../../libraries/i686-linux/lib_release_client:"`pwd`"/app_settings/mozilla-runtime-linux-i686:${LD_LIBRARY_PATH}:/usr/local/lib ./secondlife-i686-bin )
The client seems kinda slow.
By default, the open-source Second Life Viewer uses the open-source OpenJPEG library to decode the (many) JPEG-2000 texture images it receives from the servers. This isn't quite of comparable speed to the proprietary third-party library which the Linden Lab viewer builds have traditionally used, for which we are not permitted to redistribute the source.
However, the slviewer-linux-libs package includes two pre-built libraries which facilitate the use of this slightly faster image decoding method: libkdu_v42R.so and libllkdu.so. These are provided for your testing; again, we are not permitted to grant you the right to re-distribute these libraries to downstream users, but the viewer will still work (albeit slightly slower) without them.
To use these faster image-decoding libraries, they simply need to be put into the right places relative to the viewer runtime directory - nothing needs to be reconfigured or recompiled. If you're running the client from the source tree, the following will make the KDU libraries available:
cp "$SLSRC/libraries/i686-linux/lib_release_client/libllkdu.so" "$SLSRC/indra/newview/libllkdu.so"
mkdir "$SLSRC/indra/lib"
cp "$SLSRC/libraries/i686-linux/lib_release_client/libkdu_v42R.so" "$SLSRC/indra/lib/libkdu_v42R.so"
The file indra/newview/viewer_manifest.py contains some commented-out entries describing where these libraries belong; if you uncomment the two lines corresponding to libllkdu and libkdu then they will be automatically copied into the right place in the runtime directory when you follow the 'Packaging the client' instructions below.
File Dialogs Don't Work on 64 bit system
If you run a 64 bit system, and your file dialogs don't work, or they worked before and stopped after you installed an update, it may be due to a mismatch between the headers used to compile the viewer and the library it's using. The log will contain something like this:
2007-06-21T01:28:35Z INFO: ll_try_gtk_init: Starting GTK Initialization.
2007-06-21T01:28:36Z INFO: ll_try_gtk_init: GTK Initialized.
2007-06-21T01:28:36Z INFO: ll_try_gtk_init: - Compiled against GTK version 2.10.11
2007-06-21T01:28:36Z INFO: ll_try_gtk_init: - Running against GTK version 2.10.6
2007-06-21T01:28:36Z WARNING: ll_try_gtk_init: - GTK COMPATIBILITY WARNING: Gtk+ version too old (micro mismatch)
What happens here is that your distribution includes 32 bit GTK libraries, but the package only includes the libraries themselves and not the headers. When building, the SL client will build against the headers included with the main 64 bit GTK package. This will work if the 64 bit version of the library is the same or older than the 32 bit one. However, if your 32 bit library is older, then the viewer will detect the mismatch (built with headers for a newer version of GTK than it's using) and turn GTK off.
Possible solutions:
- Download the source for the version of the 32 bit GTK libraries your distribution comes with, and build your viewer against those headers.
- Upgrade your 32 bit GTK package so that it's the same or newer as the 64 bit one.
- Downgrade your 64 bit package (may not be a good idea).
Packaging the client
If you substitute 'BUILD=release' with 'BUILD=releasefordownload' in the 'Compiling' section above, then packaging the resulting code, libraries, data and documentation into a tarball for the end-user will be done automatically as the final stage of the build process; the pristine end-user client distribution has been assembled into the directory indra/newview/SecondLife_i686_1_X_Y_Z/ and has also been tarred into indra/newview/SecondLife_i686_1_X_Y_Z.tar.bz2
The file which controls what (and where) files go into the end-user runtime viewer directory is indra/newview/viewer_manifest.py
Resident contributed instructions
Automated libraries and headers adjustments, compilation and packaging
Here are two scripts (one for v1.20 and older, and one for v1.21 and newer viewers) that basically do all what is described above, and more, and entitle you to compile a SL client very easily:
- User:Henri Beauchamp/Automated Linux Build Script (1.20 and earlier)
- User:Henri Beauchamp/Building the viewer with CMake/cmake-SL script
FreeBSD
A list of patches is given for Compiling the viewer (FreeBSD).
What to do if it doesn't work for you
- Ask for help on IRC (irc.freenode.net #opensl)
- Find someone on the opensource-dev mailing list
- Fix it: Modifying CMake Files (and please, submit a patch!)
Please also see the (user contributed) instructions at User:Michelle2_Zenovka/cmake
Submitting Patches
This is probably far down the road, but if you make changes to the source and want to submit them, see the page about submitting patches.