Difference between revisions of "Build Script Anatomy"

From Second Life Wiki
Jump to navigation Jump to search
Line 70: Line 70:
         "darwin")
         "darwin")
</pre>
</pre>
Since Darwin is a UNIX variant, we can use the common configure and make commands.  Note how the ''--prefix'' option is set to the build directory so the files get installed mostly where we want.  We still need to move the header files into ''include/zlib''.
Since Darwin is a UNIX variant, we can use the common configure and make commands.  Note how the ''--prefix'' option is set to the build directory so the files get installed mostly where we want.  
<pre>
<pre>
             ./configure --prefix="$stage"
             ./configure --prefix="$stage"
             make
             make
             make install
             make install
</pre>
We still need to move the header files into ''include/zlib''.
<pre>
                         mkdir -p "$stage/include/zlib"
                         mkdir -p "$stage/include/zlib"
                         mv "$stage/include/"*.h "$stage/include/zlib/"
                         mv "$stage/include/"*.h "$stage/include/zlib/"

Revision as of 09:31, 16 February 2011

Because autobuild tool itself does not provide direct mechanisms to configure and build applications and libraries (that's just too big a problem for one tool to solve), we need to provide wrappers that autobuild can use to preform these tasks for each platform. For Linden projects we have chosen to use shell for scripting builds as we can run these on both UNIX like platforms and Windows (using CYGWIN). On this page we show an annotated typical example of a build script that we use along with autobuild to build a package. While some details obviously will change from library to library, many of the elements seen here are common to the majority of third party packages we build. You may see the original source for this page at https://bitbucket.org/lindenlab/3p-zlib/src/826e9d636a6d/build-cmd.sh. Note that build scripts are conventionally named build-cmd.sh.

Example build-cmd.sh

#!/bin/bash

Build scripts by default are run from the build directory. This command will change directories to the location of the build script which is useful for this library.

cd "$(dirname "$0")"
# turn on verbose debugging output for parabuild logs.
set -x
# make errors fatal
set -e

ZLIB_VERSION="1.2.5"
ZLIB_SOURCE_DIR="zlib-$ZLIB_VERSION"

The autobuild command should be passed as an environment variable, but we sanity check that.

if [ -z "$AUTOBUILD" ] ; then 
    fail
fi

On windows systems, the path contained in the AUTOBUILD environment variable is stored in DOS format. CYGWIN isn't smart enough to understand DOS paths so we need to convert the path to a UNIX format with the cygpath command.

if [ "$OSTYPE" = "cygwin" ] ; then
    export AUTOBUILD="$(cygpath -u $AUTOBUILD)"
fi

The autobuild source_environment command returns a string with shell code. This code contains some useful functions that we will use later in the build process.

# load autbuild provided shell functions and variables
set +x
eval "$("$AUTOBUILD" source_environment)"
set -x

Now we get to the platform specific build commands.

stage="$(pwd)/stage"
pushd "$ZLIB_SOURCE_DIR"
    case "$AUTOBUILD_PLATFORM" in
        "windows")
            load_vsvars

The packagers of zlib included this batch file to build some assembly code.

            
            pushd contrib/masmx86
                ./bld_ml32.bat
            popd

The packagers of zlib added a VisualStudio project to build this package on windows. We take advantage of this by using the biulld_sln function defined in the source environment to build the desired projects from the command line.

            
            build_sln "contrib/vstudio/vc10/zlibvc.sln" "Debug|Win32" "zlibstat"
            build_sln "contrib/vstudio/vc10/zlibvc.sln" "Release|Win32" "zlibstat"

The build product is output into the source directory, so we need to manually copy the libraries and headers into the autobuild build directory. Note how the debug and release versions of the library are copied into their respective lib/debug and lib/release directories. The headers are copied into include/zlib

            mkdir -p "$stage/lib/debug"
            mkdir -p "$stage/lib/release"
            cp "contrib/vstudio/vc10/x86/ZlibStatDebug/zlibstat.lib" \
                "$stage/lib/debug/zlibd.lib"
            cp "contrib/vstudio/vc10/x86/ZlibStatRelease/zlibstat.lib" \
                "$stage/lib/release/zlib.lib"
            mkdir -p "$stage/include/zlib"
            cp {zlib.h,zconf.h} "$stage/include/zlib"
        ;;
        "darwin")

Since Darwin is a UNIX variant, we can use the common configure and make commands. Note how the --prefix option is set to the build directory so the files get installed mostly where we want.

            ./configure --prefix="$stage"
            make
            make install

We still need to move the header files into include/zlib.

                        mkdir -p "$stage/include/zlib"
                        mv "$stage/include/"*.h "$stage/include/zlib/"
        ;;
        "linux")

Linux looks pretty much like Darwin.

            CFLAGS="-m32" CXXFLAGS="-m32" ./configure --prefix="$stage"
            make
            make install
                        mkdir -p "$stage/include/zlib"
                        mv "$stage/include/"*.h "$stage/include/zlib/"
        ;;
    esac

Finally we copy the license file into the LICENSES directory regardless of platform.

    mkdir -p "$stage/LICENSES"
    tail -n 31 README > "$stage/LICENSES/zlib.txt"
popd

pass