When creating a new autobuild package of a library, it can be unclear where to start. Autobuild How To can be a little intimidating. We're packaging a simple third party library just like all the rest of ours, so we don't need to use autobuild in it's full generality. We can skip some steps and share script fragments with other library packages.
Autobuild is largely self documenting. You can access help via the commands
autobuild --help or
autobuild <subcommand> --help, for example
autobuild edit --help. If the help text is unclear, please file a bug.
What will we be doing
Your packaging efforts will mostly consist of creating or modifying these two files: :
- Autobuild package definition metadata. This specifies the actual shell commands that autobuild should execute for each step. Modify this file using the
autobuild editcommand. Editing it by hand can be fragile.
- Shell script you are going to specify as the build command in autobuild.xml. This script encapsulates a lot of the complexity for third party libraries, so you can just clone it from an existing library and customize it for your package.
What do we do
pip install autobuild
Copy the skeleton from an existing package.
hg init hello cd hello curl -OL http://bitbucket.org/lindenlab/3p-zlib/raw/tip/BuildParams curl -OL http://bitbucket.org/lindenlab/3p-zlib/raw/tip/build-cmd.sh chmod +x build-cmd.sh curl -OL http://bitbucket.org/lindenlab/3p-zlib/raw/tip/.hgignore hg add BuildParams build-cmd.sh .hgignore
Now you'll need to edit build-cmd.sh and customize it for your package in our case, GNU Hello. Mostly this is just a search and replace (replace occurrences of zlib and ZLIB with the name or your package). For example, the sed search & replace commands would be
s/ZLIB/HELLO/g in this case. You'll likely also need to update the version number variables to match your package's version.
Additionally you'll need to update the 3 sections under the case statement with any platform specific quirks for how to build your package natively. Darwin and Linux can be pretty standardized, but under windows there are a lot of different ways to build your package. Most of these have been worked through already for one package or another. You can use these Autobuild Examples as a starting point.
You'll need some information about your package from the upstream package distributor for filling in some of the prompts. For example you should copy the copyright string and license info verbatim from the upstream distributor. Additionally we should always write our build script to copy the text of the license into
LICENSES/<package name>.txt and specify that location as the "Path to license file"
Next you'll want to configure autobuild to use your build-cmd.sh.
autobuild edit package # Name of package> hello # Package description> The GNU Hello program produces a familiar, friendly greeting. \ Yes, this is another implementation of the classic program that prints “Hello, world!” when you run it. # Copyright string (as appropriate for your package)> Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/> # Type of license (as appropriate for your package)> gpl3 # Path to license file relative to package root, if known> LICENSES/hello.txt # Version> 2.6 hg add autobuild.xml autobuild edit build # Name of config> default # Platform of config> common # Command to execute> bash # Options for command> -c,../build-cmd.sh # Arguments for command> # Filter command output> # Run by default> True autobuild edit platform name=linux build_directory=stage autobuild edit platform name=darwin build_directory=stage autobuild edit platform name=windows build_directory=stage autobuild edit build name=default platform=linux default=true autobuild edit build name=default platform=darwin default=true autobuild edit build name=default platform=windows default=true
Now if you run
autobuild print its output should look like this autobuild print example
Next you'll want to configure the manifest for packaging.
autobuild manifest --platform common add "include/hello/*.h" autobuild manifest --platform common add "LICENSES/hello.txt" autobuild manifest --platform linux add "lib/*.a" autobuild manifest --platform darwin add "lib/*.a" autobuild manifest --platform windows add "lib/debug/*.pdb" autobuild manifest --platform windows add "lib/debug/*.lib" autobuild manifest --platform windows add "lib/release/*.pdb" autobuild manifest --platform windows add "lib/release/*.lib" autobuild manifest --platform linux add "bin/*" autobuild manifest --platform darwin add "bin/*" autobuild manifest --platform windows add "bin/*.exe"
Now if you run
autobuild print its output should look like this autobuild manifest example
Now things should be complete (for this simplified example), and you should be able to run
autobuild build && autobuild package to get a package generated. Hooray!
What did we do
- We defined a build script that drives the native build of the package in a way that is compatible with our viewer builds and sandboxed from the rest of the system.
- We configured autobuild to use that build script on all 3 platforms.
- We packaged the result in a tarball consistent with Autobuild/Package_Layout
What haven't we done
- We haven't addressed how to handle dependencies. For example libcurl depends upon zlib and openssl already having their autobuild packages available for use. If you're packaging a new library that has dependencies you should begin at the leaves of the dependency tree.
- Autobuild_How_To#Adding_dependencies covers this excellently.
- We skipped features of autobuild that aren't useful for the way we're currently packaging third party library packages.
- like multiple configurations (Debug/RelWithDebInfo/Release)
- like options and arguments to the build command.
- like distinct configure and build stages.
- We haven't covered the specifics of how autobuild integrates with the windows Visual Studio build system.
- We haven't covered specifically what the ABI requirements are for packages intended to be linked into the viewer.
It is also useful to look at the tiniest possible example of autobuild without the added complexity of building a large project like the Second Life Viewer, which has additional interacting build scripts, CMake and other build time elements that can obscure the picture of what autobuild is doing. To that end, there is a toy autobuild project here: Hello Autobuild which can be checked out and played with to get a feel for autobuild. Please do not check back changes to this repo so that everyone gets the same baseline to work from.