The strength of RPM lies in its ability to define dependencies and identify conflicts accurately. Red Hat Network relies on this aspect of RPM. Red Hat Network offers an automated environment, which means that no manual intervention can take place during the installation of a package. Therefore, when building RPMs for distribution through Red Hat Network, it is imperative to follow these rules:
Learn RPM. It is crucial to have a fundamental understanding of the important features of RPM to build packages properly. For more information about RPM, start with the following resources:
When building an RPM for a child channel, build the package on a fresh install of Red Hat Enterprise Linux of the same version as the child's base channel. Be sure to apply all updates from Red Hat Network first.
The RPM package must install without using the --force or --nodeps options. If you cannot install an RPM cleanly on your build system, Red Hat Network cannot install it automatically on a system.
The RPM package filename must be in the NVR (name, version, release) format and must contain the architecture for the package. The proper format is name-version-release.arch.rpmpkgname-0.84-1.i386.rpm, where name is pkgname, version is 0.84, release is 1, and arch is i386.
The RPM package should be signed by the maintainer of the package. Unsigned packages may be distributed through Red Hat Network, but the Red Hat Update Agent (up2date) must be forced to accept them. Signing packages is highly recommended and is covered in Section 3.2, “Digital Signatures for RHN Packages”.
If the package is changed in any way, including changing the signature or recompiling, the version or release must be increased incrementally. In other words, the NVRA (including architecture) for each RPM distributed through RHN must correspond to a unique build to avoid ambiguities.
No RPM package may obsolete itself.
If a package is split into separate packages, be extremely careful with the dependencies. Do not split an existing package unless there is a compelling reason to do so.
No package may rely upon interactive pre-install, post-install, pre-uninstall, or post-uninstall scripts. If the package requires direct user intervention during installation, it cannot work with Red Hat Network.
Any pre-install, post-install, pre-uninstall, and post-uninstall scripts should never write anything to stderr or stdout. Redirect the messages to /dev/null if they are not necessary. Otherwise, write them to a file.
When creating the spec file, use the group definitions from /usr/share/doc/rpm-. If there is not an exact match, select the next best match.
<version>/GROUPS
Use the RPM dependency feature to make sure the program runs after it is installed.
Do not create an RPM by archiving files and then unarchiving them in the post-install script. This defeats the purpose of RPM.
If the files in the archive are not included in the file list, they cannot be verified or examined for conflicts. In the vast majority of cases, RPM itself can pack and unpack archives most effectively anyway. For instance, do n't create files in a %post that you do not clean up in a %postun section.