VCS package guidelines
Version control systems can be used for retrieval of source code for both usual statically versioned packages and latest (trunk) version of packages. This article covers both cases.
The ABS package provides prototypes for cvs, svn, git, mercurial, and darcs PKGBUILDs. When is installed, you can find them in
/usr/share/pacman. Latest versions can be found in the prototypes directory in the ABS Git repository.
- Properly suffix
-gitetc. If the package tracks a moving development trunk it should be given a suffix. If the package fetches a release from a VCS tag then it should not be given a suffix. Use this rule of thumb: if the output of the package depends on the time at which it was compiled, append a suffix; otherwise do not.
- A VCS package may be updated as and when needed to adopt changes to the build system, including ammendments to dependencies, URL, sources, etc. If the revision number remains the same after such an update, but produces a resulting binary which is different, increasing the
pkgrelis mandatory. If both the revision number and the resulting binary remain the same,
pkgrelshould be kept intact. There is no need to update the VCS package just to accommodate a revision bump, but one may choose to do so.
- When makepkg is run, by default it will check for newer revisions and then update the
pkgverin the PKGBUILD. Look at
--holdverin man makepkg if you want otherwise.
--holdveronly works for cvs and svn, which allow checkout of older revisions.
- Check for package conflicts. For example fluxbox-svn will conflict with fluxbox. In this case, you need to use
- Use the
providesfield so that packages that require the non-VCS package can be installed (
- You should AVOID using
replaces=...as it generally causes unnecessary problems.
- When using/defining the cvsroot, use
anonymous@to avoid a password prompt and having to enter a blank password OR use
anonymous:password@if a password is required.
- Don't forget to include the appropriate VCS tool (cvs, subversion, git, ...) in
- To preserve the integrity of the checked-out code consider copying the original build directory if you have to make edits. For example, having checked out source code to
$startdiryou can use:
mkdir src/$_cvsmod-build cd src/$_cvsmod-build ../$_cvsmod/configure
cp -r src/$_cvsmod src/$_cvsmod-build cd src/$_cvsmod-build
- With the introduction of the AUR, it is most important to avoid using backtick execution to create package variables. makepkg will automatically bump the
pkgveranyway when building the package (unless
- You should make sure that there are no VCS directories and files left over in your package. If there are, you may want to remove them, by adding a command similar to this one at the end of the the package() script:
rm -rf $(find "$pkgdir" -type d -name ".svn")
- When using Git, one can speed up the cloning operation using the
--depth=1parameter. This creates a shallow clone, and has only the last change history - since histories are unimportant for builds most of the time.
git clone git://hostname.dom/project.git --depth=1
- It's possible to create the package also from a branch other than the master. To do so add
--branch branch_nameafter the first
git clone, in this way:
git clone "$_gitroot" "$_gitname" --branch branch_name
Remember to save package with a different name, for example
pkgname-branchname-git, in order to avoid confusion with the package from the branch master.
- Copy paste script when building from repo
If you are lazy here is a sample script when making git-based PKGBUILDs.
cd "$srcdir" msg "Connecting to GIT server..." if [ -d $_gitname ] ; then cd $_gitname && git pull origin msg "The local files are updated." else git clone --depth=1 $_gitroot $_gitname fi msg "GIT checkout done or server timeout"
- Temporary build directories: When using Git, and where you need to create a separate build directory (e.g., for building/compiling), you should avoid copying over the
.gitdirectory located in the parent folder because it contains history information that Git uses internally. With repos with thousands of commits, this .git directory will contain upwards of hundreds of MiB of useless commit history that has *nothing* to do with the current working tree. The only time you'd need to copy over the .git directory itself is when you need to build from a specific, older commit (which is generally never the case as the point of a VCS PKGBUILD is to pull from the latest bleeding edge commit). Thus, instead of
rm -rf "$srcdir/$_gitname-build" cp -R "$srcdir/$_gitname" "$srcdir/$_gitname-build" # copy everything, including the useless .git folder cd "$srcdir/$_gitname-build" make # build/compile from source
you should do
rm -rf "$srcdir/$_gitname-build" cd "$srcdir/$_gitname" && ls -A | grep -v .git | xargs -d '\n' cp -r -t ../$_gitname-build # do not copy over the .git folder cd "$srcdir/$_gitname-build" make # build/compile from source
to cut down on build time and disk usage.