From Fedora Project Wiki

Creating Package HOWTO

This page describes the basic mechanics of how to create an RPM package specifically for Fedora (such as how to create a .spec file). It also gives some practical warnings about stuff that will or won't work, which may save you hours of time later. This is not the list of official package guidelines for Fedora (though it should be compatible with them). This doesn't go into depth on some issues, but it does point to other documents that do. Unlike many RPM Howto documents, this document explains the specifics for Fedora (with lots of links to Fedora-specific guidelines) and it tends to be more up-to-date (because it is maintained through the Fedora Wiki).

If you plan to create an RPM package for the Fedora repository, follow the process for How to join the Fedora Package Collection Maintainers, including following the various Fedora guidance.

Nearly all Linux distributions can install and uninstall programs as "packages". Fedora, and many other Linux distributions, use the "RPM" format for packages. There are tools that make it easy to create RPM packages. The main task is to write a ".spec" file that explains to RPM how to build and install the program. This page describes how to create RPM packages specifically for Fedora, concentrating on that very important ".spec" file.

Setting up your system and account

Before you create RPM packages on Fedora, you need to install some core development tools and set up the account(s) you will use. As root (don't type the "#"!):

 # yum groupinstall "Development Tools"
 # yum install rpmdevtools

You can create a new "dummy user" specifically for creating rpm packages. That way, if something goes terribly wrong, the program or build process can't trash your files, or send your private files/keys to the world. At the very least, you should never create your packages as user root.

You can create a new user named "makerpm" quickly by doing:

 # /usr/sbin/useradd makerpm

Then log in as that special dummy user (makerpm).

Once you're logged in as the user who is creating packages, create the directory structure in your home directory by executing (don't type the "$"):

 $ rpmdev-setuptree

The "rpmdev-setuptree" program will create an "rpmbuild" directory in your $HOME directory. Underneath "rpmbuild" are a set of subdirectories (such as SPECS and BUILD), which you will use for creating your packages. The "rpmdev-setuptree" also creates an "~/.rpmmacros" file which will cause rpm and rpmbuild to use them when appropriate.

Once you've set up the user account, you won't normally need to do these steps again.

Getting ready to package a particular program

If there are special programs that are required to build or run the program you are packaging, install those other programs and write down what they were (you'll need that information).

To package a program, you must package pristine (original) sources, along with the patches and build instructions. It's generally not okay to start with pre-compiled code. Install the file with the original source (usually a .tar.gz file) in the "~/rpmbuild/SOURCES" directory (of the rpm building user account).

Read through the manual installation instructions for your program; you're going to be automating this by editing a ".spec" file, so you have to understand what you're supposed to do first. It's probably best if you try a "dry run", going through its installation procedure without trying to do it via RPM first (that's especially true if you're not familiar with RPM).

Try to reuse what you can. Obviously, make sure you aren't packaging something that is already packaged; you can find a list of existing packages in Fedora Package Collection in the Fedora Package Database. Also check the In Progress Review Requests (for packages that are currently being reviewed) and the Retired Packages list. Failing that, see if someone has already started to package it for Fedora. Google for "PROGRAMNAME Fedora rpm" or similar... maybe you can pick up where they started. You can use http://cvs.fedoraproject.org/viewcvs/rpms/ directly view .spec files (and patches) of any similar packages already in Fedora. You can download the source RPMs and install them, too; go to a Fedora mirror's http or ftp page, select releases/9/Everything/source/SRPMS (replace "9" with the Fedora release you want), and download the source RPMs you want (they end in .src.rpm). If you followed the directions above, you can install the source RPM (which places its .spec file into ~/rpmbuild/SPECS and source files in ~/rpmbuild/SOURCES) by running:

$ rpm -ivh PROGRAMNAME-*.src.rpm

You can also unpack the .src.rpm in a directory using rpm2cpio:

$ mkdir PROGRAMNAME_src_rpm
$ cd PROGRAMNAME_src_rpm
$ rpm2cpio ../PROGRAMNAME-*.src.rpm | cpio -i

Sometimes it's easiest to start with an existing package, and then clean it up for Fedora. RPM Find may help you find rpm's for non-Fedora systems. (You can install source RPMs for other systems the same way as for Fedora). Failing that, you might look at the source package files (not the .deb binary package files) for Ubuntu or Debian (source package files are standard tarballs with a "debian/" subdirectory, possibly associated with patch files). If the FreeBSD ports collection has it, you could download the FreeBSD ports tarball and see if their packaging information helps as a starting point. However, this is sometimes not helpful at all. Different distributions have different rules, and what they do may be quite inappropriate for Fedora.

Creating a spec file

You now need to create a ".spec" file in the "~/rpmbuild/SPECS" directory. You should name it after the program name, e.g., "program.spec", use the archive name or the name advocated by the software author.

Creating a blank spec file

When you're creating a spec file for the first time, you can create its initial version using emacs or vim, they will automatically create a template for you. E.G.:

 $ cd ~/rpmbuild/SPECS
 $ vi program.spec

Here's an example of what that template will look like:

Name:		
Version:	
Release:	1%{?dist}
Summary:	
Group:		
License:	
URL:		
Source0:	
BuildRoot:	%{_tmppath}/%{name}-%{version}-%{release}-root-%(%{__id_u} -n)

BuildRequires:	
Requires:	

%description

%prep
%setup -q

%build
%configure
make %{?_smp_mflags}

%install
rm -rf %{buildroot}
make install DESTDIR=%{buildroot}

%clean
rm -rf %{buildroot}

%files
%defattr(-,root,root,-)
%doc

%changelog


You may have $RPM_BUILD_ROOT instead of %{buildroot}; just be consistent.

An example: eject

Here's a simple example, a Fedora 9 package for the "eject" program:

Summary: A program that ejects removable media using software control.
Name: eject
Version: 2.1.5
Release: 11%{dist}
License: GPL
Group: System Environment/Base
Source: http://metalab.unc.edu/pub/Linux/utils/disk-management/%{name}-%{version}.tar.gz
Source1: eject.pam
Patch1: eject-2.1.1-verbose.patch
Patch2: eject-timeout.patch
Patch3: eject-2.1.5-opendevice.patch
Patch4: eject-2.1.5-spaces.patch
Patch5: eject-2.1.5-lock.patch
Patch6: eject-2.1.5-umount.patch
BuildRoot: %{_tmppath}/%{name}-%{version}-%{release}-root
URL: http://www.pobox.com/~tranter
ExcludeArch: s390 s390x
BuildRequires: gettext
BuildRequires: automake
BuildRequires: autoconf
BuildRequires: libtool

%description
The eject program allows the user to eject removable media (typically
CD-ROMs, floppy disks or Iomega Jaz or Zip disks) using software
control. Eject can also control some multi-disk CD changers and even
some devices' auto-eject features.

Install eject if you'd like to eject removable media using software
control.

%prep
%setup -q -n %{name}
%patch1 -p1 -b .versbose
%patch2 -p1 -b .timeout
%patch3 -p0 -b .opendevice
%patch4 -p0 -b .spaces
%patch5 -p0 -b .lock
%patch6 -p1 -b .umount

%build
%configure
make

%install
rm -rf %{buildroot}

make DESTDIR=%{buildroot} install

# pam stuff
install -m 755 -d %{buildroot}/%{_sysconfdir}/pam.d
install -m 644 %{SOURCE1} %{buildroot}/%{_sysconfdir}/pam.d/%{name}
install -m 755 -d %{buildroot}/%{_sysconfdir}/security/console.apps/
echo "FALLBACK=true" > %{buildroot}/%{_sysconfdir}/security/console.apps/%{name}

install -m 755 -d %{buildroot}/%{_sbindir}
pushd %{buildroot}/%{_bindir}
mv eject ../sbin
ln -s consolehelper eject
popd

%find_lang %{name}

%clean
rm -rf %{buildroot}

%files -f %{name}.lang
%defattr(-,root,root)
%doc README TODO COPYING ChangeLog
%attr(644,root,root) %{_sysconfdir}/security/console.apps/*
%attr(644,root,root) %{_sysconfdir}/pam.d/*
%{_bindir}/*
%{_sbindir}/*
%{_mandir}/man1/*

%changelog
* Wed Apr 02 2008 Zdenek Prikryl <zprikryl at, redhat.com> 2.1.5-11
- Added check if device is hotpluggable
- Resolves #438610

You can also use the "BuildRoot:" entry from the template instead, although both are acceptable, the one from the template is the preferred one.

Spec file pieces explained

The RPM Guide, section on creating RPMs, describes the details of how to fill in a spec file. The developerWorks series "Packaging software with RPM" Part 1, Part 2, and Part 3 is also handy. Maximum RPM has the most complete information, but is dated.

You will need to follow the Fedora guidelines, such as the Package Naming Guidelines, Packaging Guidelines, and Package review guidelines.

You can insert comments with a leading "#" character, but don't insert potentially-multiline-macros (words beginning with "%") in a comment (macros are expanded first); if you're commenting out a line, double the percent signs ("%%"). Also, don't use inline comments ("#") on the same line after a script command.

Here are the major fields/areas that you will need to fill in:

  • Name: The (base) name of the package. It must follow the Package Naming Guidelines. In many cases, this will be in all lower case. Elsewhere in the spec file, you can refer to the name using the macro %{name} - that way, if the name changes, the new name will be used by those other locations. This name should match the spec file name.
  • Version: The upstream version number. See Packaging/Naming guidelines - package version for more information. If the version is non-numeric (contains tags that are not numbers or digits), you may need to include the additional non-numeric characters in the release field. If upstream uses full dates to distinguish versions, consider using version numbers of the form yy.mm[dd] (so a 2008-05-01 release becomes 8.05). Elsewhere in the spec file, refer to this value as %{version}.
  • Release: The initial value of the release should normally be "1%{?dist}". Then, increment the number every time you release a new package for the same version of software. If a new version of the software being packaged is released, the version number should be changed to reflect the new software version, and the release number should be reset to 1. See Name Guidelines - package release for more. Packaging/DistTag describes the "dist" tag, which isn't required but can be useful. Use %{release} to reuse this value.
  • Summary: A brief, one-line summary of the package. Use American English, and do not end in a period.
  • Group: This needs to be a pre-existing group, like "Applications/Engineering"; run "less /usr/share/doc/rpm-*/GROUPS" to see the complete list. If you create a sub-package "...-doc" with documentation, use the group "Documentation".
  • License: Its license; for software, this must be an open source software license. Use a standard abbreviation, e.g., "GPLv2+". Try to be specific, e.g., use "GPLv2+" (GPL version 2 or greater) instead of just "GPL" or "GPLv2" where it's true. See Licensing and the Licensing Guidelines for more information. Call this tag "License"; don't use the older, inaccurately named tag "Copyright".
  • URL: The URL for more information about the program, e.g., the project website. Note: This is NOT where the original source code came from, see "Source" (next!).
  • Source0: The URL for the compressed archive containing (original) pristine source code, as upstream released it. "Source" is synonymous with "Source0". If you give a full URL (and you should), its basename will be used when looking in the SOURCES directory. If possible, embed %{name} and %{version}, so that changes to either will go to the right place. Warning: Source0: and URL: are different - normally they are both URLs, but the "URL:" entry points to the project website, while the "Source0:" entry points to the actual file containing the source code (and is typically a .tar.gz file). As noted in the guidelines, "When downloading sources, patches etc, consider using a client that preserves the upstream timestamps. For example wget -N or curl -R. To make the change global for wget, add this to your ~/.wgetrc: timestamping = on, and for curl, add to your ~/.curlrc: -R." If there is more than one source, name them Source1, Source2, and so on. See Packaging/SourceURL for more information on special cases (using revision control, when upstream uses prohibited code, etc.).
  • Patch0: The name of the first patch that you will apply to the source code. If you need to patch the files after they've been uncompressed, you should edit the files, save their differences as a "patch" file in your ~/rpmbuild/SOURCES directory. Patches should make only one logical change, so it's quite possible to have multiple patch files.
  • BuildRoot: This is where files will be "installed" during the "%install" process (which happens after the %build compilation process). Normally you should just leave this line alone; under the usual Fedora setup, this will be a macro that will create a new special directory under /var/tmp.
  • BuildRequires: A comma-separated list of packages required for building (compiling) the program. These are not automatically determined, so you need to include everything needed to build the program. There are a few packages that are so common in builds that you don't need to mention them, such as "gcc"; see the Packaging Guidelines for the complete list of the packages you may omit. You can also specify minimum versions, if necessary, like this: "ocaml >= 3.08". You can have more than one line of BuildRequires (in which case they are all required for building). If you need file /EGGS, you can get its package by running "rpm -qf /EGGS"; if EGGS is a program, you determine its package quickly by running "rpm -qf which EGGS". Try to specify only the minimal set of packages necessary to properly build the package, since each one will slow down a "mock"-based build (e.g., try to use sed instead of perl if you don't really need perl's abilities). Watch out: Some applications permanently disable functions if their package isn't detected during the build; in those cases you may need to include those additional packages.
  • Requires: A comma-separate list of packages that are required when the program is installed. Note that the list of packages for Requires (what's required when running) and BuildRequires (what's required to build the binary RPM) are independent; a package may be in one list but not the other, or it could be in both. The "Requires" list is automatically determined by rpm, so normally you don't need to put anything here. But if you want to highlight some specific packages as being required, or require a package that rpm can't detect should be required, then add it here.
  • %description - A longer, multi-line description of the program. Use American English. All lines must be 80 characters or less. "Blank lines are assumed to separate paragraphs. Some graphical user interface installation programs will reformat paragraphs... (lines that) start with whitespace, such as a space or tab, will be treated as preformatted text and displayed as is, normally with a fixed-width font." (per the RPM Guide).
  • %prep - Script commands to "prepare" the program, that is, to uncompress it so that it will be ready for building (compiling). Typically this is just "%setup -q" or some variation of it; a common variation is "%setup -q -n NAME" if the source file unpacks into NAME. See the "%prep" section below for more.
  • %build - Script commands to "build" the program, that is, to compile it and get it ready for installing. The program should come with instructions on how to do this. See the "%build" section below for more.
  • %check - Script commands to self-test the program. This is run after %build and before %install, so you should place it there if you have this section. Often it simply contains "make test" or "make check". This is separated from %build so that people can skip the self-test if they desire. This isn't documented in many places.
  • %install - Script commands to "install" the program. The commands should copy the files from the "build directory" %{_builddir} (which would be under ~/rpmbuild/BUILD) into the buildroot directory, %{buildroot} (which would normally be under /var/tmp). See the "%install" section below for more.
  • %clean - instructions to clean out the build root. Typically:
rm -rf %{buildroot}
  • %files - the list of files that will be installed. See the "%files" section below for more.
  • %changelog - Changes in the package. Use the format example above.
  • ExcludeArch: If the package does not successfully compile, build or work on an architecture, then those architectures should be listed in the spec in an ExcludeArch tag.
  • You can add sections so that code will run when packages are installed or removed on the real system (as opposed to just running the %install script, which only does a pseudo-install to the build root). These are called "scriptlets", and they are usually used to update the running system with information from the package. See the "Scriptlets" section below for more.

Don't use the tags "Packager" or "Vendor". Don't use "Copyright" - use "License" instead. Don't create a "relocatable" package - they don't add value in Fedora yet they make things more complicated.

RPM supports subpackages, that is, a single spec file can generate many binary packages. For example, if the documentation is very large, you might generate a separate "-doc" subpackage. See below for more.

%prep section

The "%prep" section describes how to unpack the compressed packages so that they can be built. Typically, this is a set of "%setup" and/or %patch commands, which reference the Source0:, Source1:, etc. lines above. See the Maximum RPM section on %setup and %patch for more details.

Warning: In spec files, don't use in-line comments (a "#" comment on the same line after a command), and don't put macros (words beginning with "%") in a comment unless you quote the "%" as "%%". Macros can cause failures if they are in a comment, because they are always expanded (even when in a comment) and they can expand to multiple lines. This is true for %prep, %build, and so on.

%prep section: %setup command

The "%setup" command unpacks a source package, and takes several switches. Normally you should use "-q" (quiet) to prevent setup from babbling about every file it unpacks. Here are a few switches besides -q:

  • -n name: If the name of the rpm is something other than what the Source unpacks to, use this switch to state the name it unpacks to. E.G., if the tarball unpacks into a directory MYNAME, use %setup -q -n MYNAME
  • -c name: If the tarball doesn't unpack into a single directory, this creates a directory named name and then unpacks into it. Useful if you have one of those annoying tarballs that doesn't have a single common subdirectory embedded in it.

There are more %spec options if you are unpacking multiple files, which is primarily useful if you are creating subpackages (see below). The key ones are:

-a number Only unpack the source directive of the given number, such as –a 0 for source0:, after changing to the directory.
-b number Only unpack the source directive of the given number, such as –b 0 for source0:, before changing to the directory.
-D Do not delete the directory before unpacking.
-T Disable the automatic unpacking of the archives.

%prep section: %patch commands

The "%patch0" command applies patch 0 (similar for 1, 2, etc.). The normal "-pNUMBER" option applies, which simply passes that argument on to patch. Patch file names often look like "telnet-0.17-env.patch", that is, %{name}-%{version}-patch_purpose.patch (some people omit -%{version}). Patch files are typically the result of a "diff -u"; if you do this from the subdirectory of ~/rpmbuild/BUILD, you won't have to specify a -p level later. You can use all the normal ways of creating a patch file. If you're creating a patch file a single file FILENAME, a common way is to copy it to FILENAME.orig, modify it, and then save the results of "diff -u FILENAME.orig FILENAME":

cp X/Y.Z X/Y.Z.orig
vim X/Y.Z
diff -u X/Y.Z.orig X/Y.Z > ~/rpmbuild/SOURCES/PKGNAME.REASON.patch

If you're going to edit many files, one easy method is to copy the whole subdirectory underneath BUILD, and then do subdirectory diffs; once you're in BUILD/whatever, you can:

cp -pr . ../PACKAGENAME.orig
... many edits ...
diff -u . ../PACKAGENAME.orig > ~/rpmbuild/SOURCES/PKGNAME.REASON.patch

If you edit many files in one patch, you can also copy the original files using some consistent ending such as ".orig" before editing them. Then, you can use "gendiff" (in the rpm package) to create a patch with the differences. Do "man gendiff" for more information.

You should send your patch upstream, and document the upstream bug/email that sent it; see Packaging/PatchUpstreamStatus.

%prep section: Unmodified files

Sometimes, you'll package just a straight file that doesn't need to be uncompressed, e.g., a "Source1:" that is just a simple PDF file. You can "prep" those into the build directory by doing this (replace "1" with whatever number it is):

 cp -p %SOURCE1 .

%build section

The "%build" section is sometimes complicated; here you configure and compile/build the files to be installed.

Many programs follow the GNU configure approach (or some variation). By default, they will install to a prefix of "/usr/local" (/usr/local/bin, /usr/local/lib, etc.), which is a reasonable default for unpackaged files - but since you are packaging it, you will want to change the prefix to "/usr". Often some variation of this will work:

 %configure
 make %{?_smp_mflags}

Sometimes you'll want to override the variables of a makefile; you can easily do that by passing them as parameters to make, like this:

make %{?_smp_mflags} CFLAGS="%{optflags}" BINDIR=%{_bindir}

If you need to do something complicated with GNU-generated configure, take a look at "GNU autoconf, automake, and libtool". A good presentation on these as well as "make" is "Open Source Development Tools: An Introduction to Make, Configure, Automake, Autoconf" by Stefan Hundhammer.

If you include some self-tests (and that's a good idea), put them in a separate "%check" section that immediately follows the "%build" area, instead of including them in %build. That way, it will be easy for the system to skip unnecessary self-tests.

%check section

The "%check" section does testing, often it's "make test". This is not documented in many other sources of RPM info.

%install section

The "%install" section is a set of script commands to "install" the program. The commands in this section should copy the files from a directory inside the "build directory" %{_builddir} (normally ~/rpmbuild/BUILD/something) into the build root directory, %{buildroot} (normally /var/tmp/something), creating the directories inside %{buildroot} as necessary.

Watch out: Some of the terminology is very misleading:

  • The "build directory" (under which compilations occur during %build) and the "build root" (where files are copied into during the %install process) are different. The point of the %install process is to copy files, such as those under the build directory, to the right place in the build root. Perhaps "buildroot" should be called "installroot", but it's too late now, the terminology is entrenched.
  • The build directory is normally ~/rpmbuild/BUILD. The %prep stage will normally create a subdirectory underneath the build directory as part of %setup. During during %build, the current directory will actually start at %{buildsubdir}, that newly-created subdirectory under the build directory. Typically %{buildsubdir} is something like ~/rpmbuild/BUILD/%{name}-%{version}.
  • The "%install" script is not used when the binary rpm package is installed by the end-user!! The term "%install" is misleading, in fact, the script must not install the programs in the REAL final locations (e.g., in /usr/bin), but under the buildroot %{buildroot}.

As noted in the guidelines, "when adding file copying commands in the spec file, consider using a command that preserves the files' timestamps, eg. cp -p or install -p". So, if the makefile lets you override the install command (typically named INSTALL), you might want something like INSTALL="install -p" as a make parameter.

Normally, the install script would first erase the %{buildroot} directory, and then do some variation of "make install" (ideally using DESTDIR=%{buildroot}, if the program supports it). Here's an example of an %install section:

%install
rm -rf %{buildroot}
make DESTDIR=%{buildroot} INSTALL="install -p" install

Many older RPM documents suggest using "%makeinstall", which might work if "make install" doesn't support DESTDIR. However, as noted in the Fedora guidelines, "Fedora's RPM includes a %makeinstall macro but it must NOT be used when make install DESTDIR=%{buildroot} works. %makeinstall is a kludge that can work with Makefiles that don't make use of the DESTDIR variable..." - but sometimes it has subtle failures. See the Fedora guidelines if you want the details.

Ideally, every program would have a "make install" that supported DESTDIR; anything else is much harder. So, if there's no "make install", or "make install" doesn't support DESTDIR, you may have to do things by hand. So if there is no "make install", it would do some sort of sequence that would create directories that weren't already created by the "BuildRequires" packages (typically using install -d), followed by copying of files from the current directory (inside the build directory) into the buildroot directory. Running "make -n install" may make it easy to determine what it should be. For example:

%install
rm -rf %{buildroot}
cp -p mycommand %{buildroot}%{_bindir}/

A nice solution, of course, would be to patch the makefile so that there is a "make install" that supports DESTDIR.

%files section

The %files section identifies what files and directories were added by the package - and thus, which files and directories are owned by the package. Ownership is important - when you type "rpm -qif blah", you'll see who owns blah.

%files Basics

The %files section normally begins with a %defattr line which sets the default file permissions. The format of this is %defattr(<file permissions>, <user>, <group>, <directory permissions>), that is, one can specify the permissions to apply to files and directories in the %files section. The fourth parameter is often omitted. Usually one uses %defattr(-,root,root,-), where "-" means "use the default permissions".

This is followed by names or patterns of the directories or files to be installed and owned by this package. You should use macros for directory names, e.g., use %{_bindir}/myfile instead of /usr/bin/myfile, and %{_sbindir}/killaccount instead of /usr/sbin/killaccount. If a name or pattern begins with "/" when expanded, then it is presumed to have been copied into the %{buildroot} followed by that pattern; when installed on the final system, it will be copied into that name without the buildroot prefix. If you don't precede the pattern with "/", then it is presumed to be in the current directory (e.g., inside the build directory) - this is used for "documentation" files. So if your package just installs /usr/bin/mycommand, then your %files section could simply say:

%files
%defattr(-,root,root,-)
%{_sbindir}/mycommand

Any file or directory identified in the %files section is owned by the defining package. You should make sure that you declare ownership of every new file or directory the package creates. You can use wildcards (*) which match a set of files - this makes the package less sensitive to changes. For example, you can declare that all the files that were copied into %{buildroot}/usr/bin are owned by this package by declaring:

%{_bindir}/*

Note that "%{_bindir}/*" does not claim that this package owns the /usr/bin directory - it claims that all the files that were installed inside the build root 's /usr/bin are owned by the package. If you list a directory in the %files section, then you are claiming that this package owns that subdirectory and all files and directories in it, recursively (all the way down) if they are present in the build root. Do not list the "/usr/bin" or "%{_bindir}" directories directly in your %files list, because that would claim ownership of /usr/bin and everything inside it. Claiming ownership of "%{_bindir}/*" is fine, though; that just claims ownership of the subdirectories and files you placed under %{buildroot}/%{_bindir}. If you create a subdirectory such as %{_datadir}/%{name}, (/usr/share/NAME), you should include that directory in the %files list:

%{_datadir}/%{name}/

It's usually easier to use wildcards for filenames, and that's also better at coping with changes in upstream. Older RPM documentation typically shows long lists under %files with individual names, such as /usr/bin/program1 followed by /usr/bin/program2. Because of the way Fedora now uses buildroots, that is no longer necessary.

It's an error if no file matches the wildcard of a line, so only note the directories that actually matter. Also, you can't identify the same file or directory more than once. Finally, it's an error to have something in the buildroot and not listed under %files; the whole point of copying something into the buildroot is because you intend to have it installed in the final system. If you don't intend that, remove those files during the %install process.

It is also possible to exclude files from a previous match by using a %exclude glob. This can be useful for including "almost all" of the files that match a different glob. However, note that, like any other file glob, even a %exclude glob will fail if it matches nothing. (This might be considered counterintuitive, as the whole point is essentially to ensure that a certain file ISN'T there, so this rule is especially important to remember.)

%files prefixes

You may need to add one or more prefixes to a %files entry (if more than one, use a space to separate them).

Typically there is a "%doc" entry with a list of documentation files that didn't get copied into the buildroot; usually there is at least a README and LICENSE file. You must include the license file, if there is one. You may prefix some of these with %attr(mode, user, group) to set the file permission modem, user, or group. You don't need to claim ownership of the /usr/share/doc/%{name} directory, that's automatic. Any %doc entry must must not affect the runtime of the application (if it is in %doc, the program must run properly if it is not present).

If you save configuration files (under /etc - don't put them under /usr), you should normally prefix them with %config(noreplace) unless this program version uses a non-backwards-compatible configuration format (in which case, prefix them with %config).

Prefixing a %files entry with "%attr(mode, user, group)" lets you set the permissions for particular file(s), e.g., "%attr(0644, root, root)". A "-" means "use the default".

If a file is in particular natural language, use %lang to note that. E.G.:

%lang(de) %{_datadir}/locale/de/LC_MESSAGES/tcsh*

Some documentation claims that %license and %readme are valid prefixes; they are not valid in Fedora. Use %doc instead.

%files and Filesystem Hierarchy Standard (FHS)

You should follow the Filesystem Hierarchy Standard (FHS), i.e., ordinary application executables go into /usr/bin, global configuration files go into /etc, ordinary libraries go into /usr/lib, and so on, with one exception: executables that should not normally be executed directly by users or administrators should go into a subdirectory of /usr/libexec; usually you'd refer to the necessary directory as "%{_libexecdir}/%{name}".

You shouldn't be installing files under /usr/local; that is where unpackaged files go. Typically there will be a "prefix" attribute that lets you set the prefix to be "/usr" instead of "/usr/local".

Unfortunately, many programs' "normal" installation routines do not follow the FHS. In particular, many programs normally place architecture-independent libraries under /usr/lib, instead of under /usr/share as the FHS requires. The FHS /usr/lib section says that /usr/lib is for architecture-dependent data (e.g., ELF files like .so files), while /usr/share is for architecture-independent data. That way, systems with different CPUs can share /usr/share. There are many exceptions to this rule in Fedora (e.g., Python and Perl), but Fedora applies this rule more strictly than some distributions. Note, for example, that rpmlint will complain if you put just about anything other than ELF files into /usr/lib.

%files example

Here's a simple example of a %files section:

%files
%defattr(-,root,root,-)
%doc README LICENSE
%{_bindir}/*
%{_sbindir}/*
%{_datadir}/%{name}/

Scriptlets

You can add sections so that code will run when packages are installed or removed on the real system (as opposed to just running the %install script, which only does a pseudo-install to the build root). These are called "scriptlets", and they are usually used to update the running system with information from the package.

The scriptlets in %pre and %post are run before and after a package is installed (respectively). The scriptlets %preun and %postun are run before and after a package is uninstalled. The scriptlets %pretrans and %posttrans are run at start and end of a transaction. See Packaging/ScriptletSnippets for more examples and details. For example, every binary RPM package which stores shared library files (not just symlinks) in any of the dynamic linker's default paths, must call ldconfig in %post and %postun (post-install and post-uninstall). If the package has multiple subpackages with libraries, each subpackage should also have a %post/%postun section that calls /sbin/ldconfig. For example:

%post -p /sbin/ldconfig
%postun -p /sbin/ldconfig

Beware: The "-p" option specifies what command processor to use for the commands on the following lines. If there are no following lines, then using /sbin/ldconfig as the "command processor" is a minor efficiency improvement compared to putting "/sbin/ldconfig" on the next line, and letting the shell invoke it. That's because by using "-p", the shell isn't invoked simply to invoke a single program. But if you have multiple shell commands, don't use "-p" or /sbin/ldconfig after it! Instead, leave it blank, and include the shell commands under it.

If you are going to run programs in scriptlets, they must be installed before you run them. You have to use special variants of the "Requires:" tag, so that the program will be installed before you try to use it. These are of the form "Requires(CONTEXT):", e.g., "Requires(post)".

Most scriptlets (%pre, %post, %preun, and %postun) provide an argument you can use, accessed via $1, which is the number of packages of this name which will be left on the system when the action completes. Don't compare for equality with 2; check if they are greater than or equal than 2, since users can arrange to have multiple versions of a package installed simultaneously. For %pretrans and %posttrans, $1 is always 0.

For example, after adding an info manual to the system the dir file which indexes the info manuals should be updated. Basically, after you install the info manual, you need to run the program install-info. That's fine, except that install-info is part of package info, and there's no guarantee that info is installed unless we require it. Also, if "install-info" fails, we don't want to fail all processing. Here's one way to do that:

Requires(post): info
Requires(preun): info
...
%post
/sbin/install-info %{_infodir}/%{name}.info %{_infodir}/dir || :
%preun
if [ $1 = 0 ] ; then
/sbin/install-info --delete %{_infodir}/%{name}.info %{_infodir}/dir || :
fi

Another scriptlet-like abilility are triggers. You can define triggers for when other packages are installed or uninstalled. See Maximum RPM for more information about triggers.

Macros

Spec files may contain "macro" references (text beginning with "%"), which are replaced with other values. You can follow % by a word, e.g., "%name", but just like shell variables you must bracket the name with {...} if letters or digits immediately follow, e.g., "%{name}".

As noted in the Packaging Guidelines, There are two styles for referring some values such as the rpm Build Root and Optimization Flags:

  • "macro style": %{buildroot}, %{optflags}
  • "variable style": $RPM_BUILD_ROOT, $RPM_OPT_FLAGS

Pick a style and use it consistently throughout your packaging; this document uses "macro style".

Here are some typical macros:

Macro Typical Expansion
%{_bindir} /usr/bin
%{_builddir} ~/rpmbuild/BUILD (build directory; see %buildsubdir)
%{buildroot} /var/tmp/... - buildroot, where files are "installed" during %install
%{buildsubdir} %{_builddir}/%{name} - where files are compiled during %build. It's under %{_builddir}, set after %setup.
%{_datadir} /usr/share
%{_includedir} /usr/include
%{_infodir} /usr/share/info
%{_initrddir} /etc/rc.d/init.d
%{_libdir} /usr/lib
%{_libexecdir} /usr/libexec
%{_localstatedir} /var
%{_mandir} /usr/share/man
%{name} Name of package, set by Name: tag
%{_sbindir} /usr/sbin
%{_sharedstatedir} /usr/com
%{_sysconfdir} /etc
%{version} Version of package, set by Version: tag

To see more about macros you can look in /etc/rpm/* and the "macros" files under "/usr/lib/rpm/", especially /usr/lib/rpm/macros. You can also use "rpm --showrc" to show the values rpm will use for all of the options currently set in rpmrc and macro configuration files.

You can set your own macro values using %define; be sure to define them before you use them. Macro definitions can refer to other macros. For example:

%define myvalue 50

You can use rpmbuild to find the value of some macro, using its "-E" (--eval) option. For example, to find the current expansion of %{_bindir} in myfile.spec, you can run:

rpmbuild -E '%{_bindir}' myfile.spec

Packaging/RPMMacros has more information on macros, as does RPM Guide chapter 9.

Other tags

We noted the "Requires" and "BuildRequires" tags earlier. There are a few other tags for controlling dependencies: Provides, Obsoletes, Conflicts, and BuildConflicts.

  • "Provides:" lets you list virtual package names that this package provides. Sometimes there are several different packages that can provide a function, and using packages won't care which one. In that case, each of the packages that provide the function should "provide" a virtual package, and then using packages can list the virtual package name under "Requires:". For example, several different packages might provide "latex"; if you depend on the virtual package "tex(latex)", then users can choose which package to get "latex" from. If you provide virtual packages, you might also want to use the "alternatives" system. You can find out what a given package provides (both virtual and non-virtual names) by querying "rpm -q --provides PACKAGENAME". Some virtual packages in Fedora are:
    • MTA : Used for mail transport agents, such as sendmail.
    • tex(latex) : Used for latex
  • "Obsoletes:" lets you state that installing this package should (normally) cause the removal of the other named package(s). This is useful when a package's name changes, or when a package wholly replaces a different package.
  • "Conflicts:" lets you state what packages cannot be installed simultaneously this one. Obviously, try to avoid this if you can; see Packaging/Conflicts if you think you need to use it.
  • "BuildConflicts:" lets you state what packages cannot be installed when building this package. Obviously, try to avoid this if you can.

You can control which architectures a package builds (or doesn't build). For example, if your package can't compile on ppc, you can do this:

ExcludeArch: ppc

There's also an "ExclusiveArch" tag. The valid architectures one can specify in these tags are listed in the Architectures section.

Subpackages

A spec file can define more than one binary package, e.g., client and server, or runtime and developer packages. If there's a large amount of documentation, it may be split into a NAME-doc subpackage. You will always have one spec file and one source RPM (SRPM), even if there are multiple binary RPMs that they generate. A spec file that produces multiple binary packages still has only one creation process, so there is only one %prep, %build, %check, and %install section that creates all the files for all the packages.

In a spec file, use the %package directive to start defining a subpackage:

%package sub_package_name

By default, the subpackage name is PACKAGE_NAME, "-", SUBPACKAGE_NAME; you can use "-n" to override this and make a new name:

%package -n new_sub_package_name

After the %package directive, list the tags for the subpackage. This should include at least the "Summary:" and "Group:" tags and directives "%description SUBPACKAGE_NAME" and "%files SUBPACKAGE_NAME". Anything not specified by the subpackage will be inherited from its parent. For the directives, if you used "-n" with %package, you'll need it again for these directives. You need to specify the name for the other directives, e.g., %pre and %post, if you use them in the subpackage.

See the RPM Guide section on subpackages for more information.

Conditionals

You can insert conditional statements. E.G., you can test if you are creating a binary for a certain architecture with:

%ifarch ARCHITECTURE_NAME

the negated version with:

%ifnarch ARCHITECTURE_NAME

or the more general conditional:

%if TRUE_OR_FALSE

There is an optional "%else" section; all of these are closed with "%endif".

Miscellaneous hints

If you want to see lots of examples of scriptlets, you can show all the scriptlets on installed programs using:

 rpm -qa --queryformat "\n\nPACKAGE: %{name}\n" --scripts | less

Packaging/FrequentlyMadeMistakes has information on frequently-made mistakes.

Don't try to interact with the user; RPM is designed to support batch installs. If an application needs to show a EULA, that needs to be part of its initial execution, not its installation.

You might not want to start services, because in a big install that could slow things down. If you install an init script, consider using chkconfig to arrange for the service to be started and stopped on the next reboot. Before uninstalling you should normally try to stop its services if it's running.

Uninstall should reverse most changes made during installation, but don't remove any user-created files.

Normally, if there are binary executables, a separate "debug" package is created with the symbols, and the symbols are stripped from the normal binary packages. If this shouldn't happen, you can disable the package-creation and stripping with:

%define _enable_debug_package 0
%define debug_package %{nil}
%define __os_install_post /usr/lib/rpm/brp-compress %{nil}

There are also some recommendations and controversial tricks on PackageMaintainers/PackagingTricks.

GUI programs must have a desktop entry (so that people can invoke it from a graphical menu). The desktop entry spec (for .desktop files) and icon theme spec defines how to do that (e.g., how to use /usr/share/icon).

Some older documents about RPM have the most information, but some older documents make claims that are no longer true:

  • rpm files are no longer placed in a shared /usr/src/redhat directory. This is an obsolete way of using rpm and not recommended; modern systems set a %{_topdir} instead like ~/rpmbuild.
  • the %install process does not install files in their final location. Instead, it "installs" files to the buildroot.
  • The "rpm" command no longer creates packages (e.g., "rpm -ba" was once legal). Use the separate "rpmbuild" program instead.

Quick test with rpmlint

Before trying to build anything from it, you might want to run rpmlint on the spec file:

rpmlint program.spec

This will catch many errors early. If the reported error doesn't make sense, run it again with the "-i" option (this gives longer messages).

Generally, you should not have errors from rpmlint, but sometimes rpmlint is excessively noisy. The Fedora packaging guidelines explain which ones to ignore, e.g., ignore "no-packager-tag" and "no-signature" errors.

Creating RPMs from the spec file

Once you've create a spec file, say "program.spec", you can create source and binary RPMs by simply running this:

 $ rpmbuild --clean -ba program.spec

This will attempt to perform the following stages:

  • %prep (preparation) stage, which uncompresses and installs the sources and patches into %_builddir (a subdirectory of ~/rpmbuild/BUILD)
  • %build stage, which builds (e.g., compiles) the files to be installed in %_builddir. Usually this is some equivalent of "make".
  • %install stage, which copies the files from the build directory %_builddir (which would be under ~/rpmbuild/BUILD) into the buildroot directory, %{buildroot}. The buildroot directory is set by the earlier "BuildRoot:"; if you leave it to its normal value beginning %{_tmppath}/%{name}..., then the buildroot will be inside /var/tmp.
  • Create the binary and source RPM packages (.rpm and .src.rpm files). The binary RPM files are created using the information from the %files list.

Watch out: the "build directory" (where compilations occur during %build) and the "build root" (where files are installed during the %install) are different.

When things go wrong, you can "cd" into the appropriate directory and see what's left over. If you want to skip earlier stages, use the "--short-circuit" option; this is handy if you had a successful build, but have an error in the %install section. For example, to restart at the %install stage (skipping earlier stages), do this:

 $ rpmbuild -bi --short-circuit program.spec

If it is successful, you'll find your binary RPM(s) in the "~/rpmbuild/RPMS/" subdirectory, and the source RPM in "~/rpmbuild/SRPMS".

If you just want to create a source RPM (.src.rpm), do this in the SPECS directory:

rpmbuild -bs program.spec

This will create the source RPM in ~/rpmbuild/SRPMS. Creating only a source rpm (.src.rpm) is quite quick, because rpm simply needs to copy the .spec file and associated SOURCES files into a .src.rpm file. Creating a binary rpm typically takes much longer, because this requires running the %prep, %build, and %install scripts.

Testing RPMs you've built

If you "cd" to the "~/rpmbuild/RPMS" directory, and then cd to the architecture subdirectory, you'll find some binary rpms. You can quickly see their files and their permissions by using rpmls (check to see that they are what you expect):

$ rpmls *.rpm

Run rpmlint on the binary RPM (rpmlint works on both .spec and RPM files, and finds different things):

$ rpmlint *.rpm

If those look okay, you can become root and try to install them:

# rpm -ivp XYZ1.rpm XYZ2.rpm XYZ3.rpm ...

Then, you can test them out. If it's a command-line tool, can you invoke it without prepending /usr/bin? If it's a GUI tool, does it show up in the menu (if it doesn't, something is wrong with your .desktop entry).

You can uninstall them later using:

# rpm -e XYZ1 XYZ2 XYZ3

If that all works, you can use Mock to do a rigorous test that you have accurate build requirements. See PackageMaintainers/MockTricks for more information about how to use Mock; once your account is a member of the "mock" group, you can run commands like this to do local testing:

$ mock -r fedora-9-i386 rebuild path_to_source_RPM

Once Mock works on your system, you can use Koji (which uses Mock) to do builds on many different systems, some of which you may not have. PackageMaintainers/Join and PackageMaintainers/UsingKoji have more information about Koji. Once it's set up, you can test your source RPM on a variety of platforms by running commands like:

$ koji build --scratch dist-f9 path_to_source_RPM

Helpful tools

The "rpmdevtools" package has a number of helpful tools; "rpm -qil rpmdevtools" will show you what it installs. One particularly useful tool is rpmdev-bumpspec, which has this form:

rpmdev-bumpspec --comment=COMMENT --userstring=NAME+EMAIL_STRING SPECFILES

rpmdev-bumpspec will bump the release tag in the spec file(s), and add a changelog comment with the right datetime and version format. COMMENT should typically start with "- ".

You might find RUST useful (GPL). It is "a drag & drop RPM creation GUI and a 'sandboxing' toolkit that allows you to do software installations within a chrooted environment and automatically generate RPMs from arbitrary source code, without ever seeing a spec file." If you're creating spec files, it can help you determine the %files. (Note: it is no longer at "rusthq.com".)

Alien converts between package formats. It won't produce clean source RPMs, but converting an existing package might provide helpful information.

Guidelines and rules

When you create your packages, you'll need to follow the following rules and guidelines:

There are many official guidelines that will help you with specific circumstances (Java programs, OCaml programs, GNOME programs, etc.); the Packaging Guidelines include cross-references to those guidelines. You can also learn more from the SIGs and Package Maintainers sections. You can also see the list of all Wiki pages about Packaging to see if any apply.

Failing that, you might find some useful recommendations in the unofficial Packaging Drafts and Packaging Drafts To Do. These are unofficial, obviously. You might find ideas from SuSE, Debian, but distributions differ in their rules, so do not presume they can be used directly.

The .spec files that you create must be open source software, as noted in the CLA.

For more information

The Package Maintainers page links to many other useful pages, and the Updating Package HOWTO describes how to update an existing package you already maintain in Fedora.

For more information, outside of the Fedora Wiki, see:

Note: The rpm5.org site has some documentation, but do not depend on it; that is the home of a fork of RPM maintained by Jeff Johnson. The RPM used by Fedora (and Novell/SuSE) is instead based at rpm.org. lwn.net has a brief article about this.