stereochrome

Building an RPM package: a guide for the unwilling

Some day, if you’re a software developer and develop software than runs on Linux, you will end up, whether you like it or not, having to build an RPM file.

I’ve just spent an absolutely inordinate amount of time working out how to do this. It wasn’t fun, and I don’t think I’d like to ever have to do it again, so to save myself or some other poor sod similar pain, I thought I’d write up what I did here.

First, a few assumptions. Unlike most explanations of how to do this, I’m not going to be building from an tarball, so I’m not going to be covering %prep and all that just yet. Also, I’m building the RPM from a tag in a subversion repository.

Before we do anything else, we need to set up our system so we can build the RPM without being root. To do that, open the file ~/.rpmmacros in your editor of choice and enter the following:

%_topdir %(echo ~/rpmbuild)

This tells RPM to use ~/rpmbuild as its working directory. RPM requires its working directory to have a certain format, so type the following to build it:

$ cd rpmbuild; mkdir -p BUILD RPMS/i386 SOURCES SPECS SRPMS

BUILD is where the work is done, RPMS is where the finished RPM is dumped, SOURCES is where you drop tarballs if you’re building from one, SPECS is where you put your spec files, and SRPMS is where finished source RPMs are dumped.

Let’s get to writing the spec file, which I’m going to call libexample.spec. The first section of the RPM contains metadata, so let’s write up some:

Summary: My example library.
Name: libexample
Version: 0.1
Release: 1
Group: Development/Libraries
Packager: J.R. Hacker <me@example.com>
License: BSD
BuildRoot: %{_tmppath}/%{name}-root

Most of these should be pretty self-explanatory: Summary is a short human-readable summary of what the package is for; Name is its actual name; Version is the version of the library you’re building; Release is the release number of that version; Group specifies the kind of package it is, Packager is your name and contact details, and License specifies the licensing details for the package.

The last one, BuildRoot needs a little more explanation. When you’re building an RPM, you’re building a directory tree that will be extracted to the filesystem root of your machine when the package is installed. You don’t want to pollute the directory of your machine with crap, so you need somewhere you can cleanly build it. In the case above, let’s say that the value of the %\{_tmppath} macro is /tmp. This would mean that BuildRoot would have the value /tmp/libexample-root. Similarly to how the Name header is referenced here as %{name}, we’ll be referencing BuildRoot as %{buildroot}.

Next comes the %description section, which contains a longer description of the package than given in the summary. Following this is normally the %prep section, which we’re going to ignore here and not use because we’re not fiddling with tarballs. Then we have the %build section. This is where the first bit of real work happens, but first a quick note on the directory structure we’re building.

We’re exporting a single directory from SVN called src. This contains a makefile that build a library using GNU libtool into a folder within it called src/.libs. The project doesn’t use the autotools, so the spec file does all the installation work after it’s built. I’m using this structure because I want to demonstrate explicitly how the contents of %{buildroot} is built.

Here’s what our %build section looks like:

%build
rm -rf src %{buildroot}
svn export https://svn.example.com/projects/libexample/tags/0.1/src/
(cd src; make)

When this section is being executed, RPM automatically changed to the %{_topdir}/BUILD directory. It’s here that we do our work. The second line cleans up our environment so that we can build it. The third exports the project’s src directory to a directory of the same name. The third then switches to this within a subshell and executes the makefile within it. Now it’s time to for the %install section, which is where we build the directory tree that’s going to end up in the RPM.

%install
# Create the directory hierarchy for building our RPM.
mkdir -p %{buildroot}%{_libdir} %{buildroot}%{_includedir}
# Copy whichever files that RPM will contain into it.
install --strip --mode=0755 src/.libs/%{name}.so.0 src/.libs/%{name}.a %{buildroot}%{_libdir}
install --mode=0755 src/%{name}.la %{buildroot}%{_libdir}
sed -i -e s/^installed=no$/installed=yes/ %{buildroot}%{_libdir}/*.la
install --mode=0644 src/*.h %{buildroot}%{_includedir}
(cd %{buildroot}%{_libdir}; ln -sf %{name}.so.0 %{name}.so)
# Build a manifest of the RPM's directory hierarchy.
echo "%%defattr(-, root, root)" >MANIFEST
(cd %{buildroot}; find . -type f -or -type l | sed -e s/^.// -e /^$/d) >>MANIFEST

The %{_libdir} and %{_includedir} macros specify the directories where your system installs libraries and header files. On a typical Red Hat derived system, these are going to have the values /usr/lib and /usr/include, though this isn’t necessarily what they’re going to be.

Within %{buildroot} we need to create the directory tree that’s going to end up in the RPM. We then need to copy anything that will end up in %{_libdir} after installation into the correct place under %{buildroot}, and to the same with any headers too. We use install to make sure that permissions and ownership for all the files are correct. The fourth last line makes a symbolic link pointing whatever linkers end up linking to our library to the version of it we’re installing. The second last and last lines build a manifest of the files within %{buildroot}. We’ll be using that later to save ourselves some work. Now that our root directory hierarchy’s been built, we can get on with the odds and ends.

Your RPM might need to have commands executed before or after installation and deinstallation. There are hooks for these, but I’m only going to cover two of them: %post, which is ran post-installation, and %postun, which is ran post-deinstallation. We’re installing libraries, which means that after they’re added and after they’re removed, we need to call ldconfig so that the dynamic linker knows about them.

%post
/sbin/ldconfig

%postun
/sbin/ldconfig

After the RPM is built, we need to clean up our environment again. That’s what the %clean section is for:

%clean
rm -rf src %{buildroot} MANIFEST

The last section is the file manifest, which is contained within the %files section. Here you specify what files the RPM will contain and where they reside in the filesystem. Normally, it’d look something like this:

%files
%defattr(-, root, root)
%{_libdir}/%{name}.so*
%{_libdir}/%{name}.la
%{_libdir}/%{name}.a
%{_includedir}/example.h

The %defattr macro specifies the permissions and ownership of the files that follow it and takes three arguments: the permissions, the owner, and the group. If any of these don’t matter, use a hyphen. This way, the attributes already specified on the files will be used instead.

However, we were clever earlier on and built a manifest of the contents of %{buildroot}, so we can just tell RPM to use that, which means the %files section is reduced to this:

%files -f MANIFEST

And that’s it! You should now have an RPM. Here’s our finished spec file:

Summary: My example library.
Name: libexample
Version: 0.1
Release: 1
Group: Development/Libraries
Packager: J.R. Hacker <me@example.com>
License: BSD
BuildRoot: %{_tmppath}/%{name}-root

%description
A library that acts as an example of how to build an RPM.

%build
rm -rf src %{buildroot}
svn export https://svn.example.com/projects/libexample/tags/0.1/src/
(cd src; make)

%install
# Create the directory hierarchy for building our RPM.
mkdir -p %{buildroot}%{_libdir} %{buildroot}%{_includedir}
# Copy whichever files that RPM will contain into it.
install --strip --mode=0755 src/.libs/%{name}.so.0 src/.libs/%{name}.a %{buildroot}%{_libdir}
install --mode=0755 src/%{name}.la %{buildroot}%{_libdir}
sed -i -e s/^installed=no$/installed=yes/ %{buildroot}%{_libdir}/*.la
install --mode=0644 src/*.h %{buildroot}%{_includedir}
(cd %{buildroot}%{_libdir}; ln -sf %{name}.so.0 %{name}.so)
# Build a manifest of the RPM's directory hierarchy.
echo "%%defattr(-, root, root)" >MANIFEST
(cd %{buildroot}; find . -type f -or -type l | sed -e s/^.// -e /^$/d) >>MANIFEST

%post
/sbin/ldconfig

%postun
/sbin/ldconfig

%clean
rm -rf src %{buildroot} MANIFEST

%files -f MANIFEST

There’s much more to creating RPMs to what I’ve detailed here, but this should at least give you a good start.

Nota bene: This was all written pretty much trail-of-though, so there may be errors and typos. If you find any before I do, post up a comment.

References

I didn’t really find anything particularly clear and informative online about building RPMs, except the Fedora Project’s RPM Guide, the HOW-TO linked to above, and this oldish tutorial. I also found this one after the fact, and it looks pretty decent, if terse.

Created at 22:31 on January 16th, 2008 and last modified at 18:02 on January 21st, 2010