To build RPM you will need several other packages: -------------------------------------------------- The cmake build system, version 3.18 or later. It is available from https://cmake.org/ The popt library for option parsing, must be version 1.13 or later. It is available from http://ftp.rpm.org/popt/ The debugedit >= 0.3 tools for producing debuginfo sub-packages. It is available from https://sourceware.org/debugedit/ Lua >= 5.2 library + development environment. Note that only the library is needed at runtime, RPM never calls external Lua interpreter for anything. Lua is available from http://www.lua.org The zlib library for compression support. You might also need/want the unzip executable for java jar dependency analysis. All available from http://www.gzip.org/zlib/ The libmagic (aka file) library for file type detection (used by rpmbuild). The source for the file utility + library is available from ftp://ftp.astron.com/pub/file/ You will need a cryptographic library to support digests and signatures. This depends on the OpenPGP parser used: the default is rpm-sequoia library (>= 1.3.0 required), which is available from https://github.com/rpm-software-management/rpm-sequoia Use of rpm-sequoia is strongly recommended. Most importantly, the internal parser is considered insecure. It simply ignores various critical aspects of OpenPGP (such as sub-packet binding signatures) that are properly implemented in Sequoia. Some other Sequoia advantages include being implemented in a memory-safe language, configurable policy and user-relevant error messages. For more information, see https://sequoia-pgp.org/ If using the deprecated internal parser (-DWITH_INTERNAL_OPENPGP=ON), the default is libgcrypt, but alternatively OpenSSL can be used by additionally specifying -DWITH_OPENSSL=ON. libgcrypt library is available from https://www.gnupg.org/software/libgcrypt/ If using the OpenSSL library for encryption, it must be version 3.0.0 or later. Note: when compiling against OpenSSL, there is a possible license incompatibility. For more details on this, see https://people.gnome.org/~markmc/openssl-and-the-gpl.html Some Linux distributions have different legal interpretations of this possible incompatibility. It is recommended to consult with a lawyer before building RPM against OpenSSL. Fedora: https://fedoraproject.org/wiki/Licensing:FAQ#What.27s_the_deal_with_the_OpenSSL_license.3F Debian: https://lists.debian.org/debian-legal/2002/10/msg00113.html The OpenSSL crypto library is available from https://www.openssl.org/ RPM needs a database engine for normal operation. The main options are "ndb" and "sqlite", both enabled by default but can be disabled with -DENABLE_NDB=OFF and -DENABLE_SQLITE=OFF respetively. Additionally standalone support for read-only BDB databases is available as "bdb_ro" (-DENABLE_BDB_RO=ON) to aid with migration from BDB. The ndb and bdb_ro backends have no external dependencies. SQLite >= 3.22.0 is required for the sqlite database backend. SQLite is available from https://www.sqlite.org/ SELinux support is enabled by default but can be disabled with -DWITH_SELINUX=OFF. libselinux is is available from http://www.nsa.gov/selinux/ It may be desired to install bzip2, gzip, and xz/lzma so that RPM can use these formats. Gzip is necessary to build packages that contain compressed tar balls, these are quite common on the Internet. These are available from http://www.gzip.org http://www.bzip.org http://tukaani.org/xz/ Python bindings to RPM library are built by default, but it can be disabled with -DENABLE_PYTHON=OFF. You'll need to have Python >= 3.7 runtime and C API development environment installed. Python is available from: http://www.python.org/ POSIX.1e draft 15 file capabilities support is enabled by default, to disable use -DWITH_CAP=OFF. You'll also need recent libcap, available from: http://ftp.kernel.org/pub/linux/libs/security/linux-privs/libcap2/ POSIX 1003.1e draft 17 ACL verification support is enabled by default, to disable use -DWITH_ACL=OFF. You'll also need the ACL library, available from: ftp://oss.sgi.com/projects/xfs/cmd_tars/ For best results you should compile with GCC and GNU Make. Users have reported difficulty with other build tools (any patches to lift these dependencies are welcome). Both GCC and GNU Make available from http://www.gnu.org/ If National Language Support (NLS) is desired you will need gnu gettext (currently this is required to build rpm but we hope to lift this requirement soon), available from http://www.gnu.org/ By default, Rpm uses C.UTF-8 locale as it's default locale. If your environment does not support this, you can make rpm use the traditional C locale with -DENABLE_CUTF8=OFF. If you are going to hack the sources (or compile from source repository) you will need most of the GNU development tools including: autoconf, automake, gettext, libtool, makeinfo, perl, GNU m4, GNU tar available from http://www.gnu.org/ If you plan on using cryptographic signatures you will need a version of GPG, available from http://www.gnupg.org/ OpenMP multithreading support is automatically enabled if your C compiler has support for OpenMP version 4.5 or higher (to disable, use -DENABLE_OPENMP=OFF option). For GCC, OpenMP 4.5 is fully supported since GCC 6.1, which is available from http://www.gnu.org/ If glibc is used, it needs to be of version 2.27 or newer. Older glibc versions have a longstanding bug where glob() does not return dangling symlinks as matches, which broadly affects rpmbuild. Rpm requires a POSIX.1-2008 level operating system. To compile RPM: -------------- mkdir _build cd _build cmake .. You can view the various cmake compile options with: cmake -L .. and fine tune the built components + features, eg: cmake -DENABLE_PYTHON=OFF .. If you have tools outside the regular FHS paths that you need rpm to find, you can augment the search path via MYPATH environment variable to cmake. Now build the system with: make and then install with: make install By default, rpm installs a series of default platforms based on the CPU architecture names in subdirectories called /usr/lib/platform/- This is enough for many distributions. However, some distributions may use more specific platform names that refer to particular computer systems, like SBCs or specific CPU tuning when compiling. Examples of such platform names are: "genericx86_64", "intel_skylake_64", "raspberrypi_armv7", "raspberrypi_armv8", etc. If the platform name is put into /etc/rpm/platform, then rpmbuild uses it and the only macros file rpmbuild looks for is /usr/lib/platform/`cat /etc/rpm/platform`-/macros If this file does not exist, many rpm macros will not have their expected values set and e.g. %configure will fail when trying to run rpmbuild. To allow creating the macros file for such a custom platform, the shell variables listed below must be set. If RPM_CUSTOM_ARCH is not set, the rest is ignored. export RPM_CUSTOM_ARCH=genericx86_64 export RPM_CUSTOM_ISANAME=x86 export RPM_CUSTOM_ISABITS=64 export RPM_CUSTOM_CANONARCH=x86_64 export RPM_CUSTOM_CANONCOLOR=0 # to use /usr/lib for %_libdir export RPM_CUSTOM_CANONCOLOR=3 # to use /usr/lib64 for %_libdir make install This also creates and installs the new platform file e.g. /usr/lib/platform/genericx86_64-linux/macros Rpm comes with an automated self-test suite. The test-suite requires podman (https://github.com/containers/podman/) or docker (https://github.com/docker/). It is enabled by default but can be disabled with -DENABLE_TESTSUITE=OFF to avoid the dependencies. The test-suite can be executed with: make check Additionally, rpm supports executing the test-suite in a way that's similar to how the project CI is set up. You can do that with: make ci Finally, if you wish to prepare an rpm source tar ball, you should do make dist To package RPM: -------------- After RPM has been installed you can run rpm to build an rpm package. Edit the rpm.spec file to mirror any special steps you needed to follow to make rpm compile and change the specfile to match your taste. You will need to put the rpm source tar file into the SOURCES directory and we suggest putting the specfile in the SPECS directory, then run rpmbuild -ba rpm.spec. You will end up with two rpms which can be found in RPMS and SRPMS. If you are going to install rpm on machines with OS package managers other then rpm, you may choose to install the base rpm package via a cpio instead of a tar file. Instead of running "make tar" during the build process, as described above, use the base rpm packages to create a cpio. After the rpms have been created run rpm2cpio on the base rpm package, this will give you a cpio package which can then use to install rpm on a new system. rpm2cpio rpm-4.0-1.solaris2.6-sparc.rpm > rpm-4.0-1.solaris2.6-sparc.cpio Non Linux Configuration Issues: ------------------------------ OS dependencies: ---------------- Under RPM based Linux distributions all libraries (in fact all files distributed with the OS) are under RPM control and this section is not an issue. RPM will need to be informed of all the dependencies which were satisfied before RPM was installed. Typically this only refers to libraries that are installed by the OS, but may include other libraries and packages which are available at the time RPM is installed and will not under RPM control. Another common example of libraries which may need dependency provisions are precompiled libraries which are installed by the OS package manager during system build time. The list of dependencies you will wish to load into RPM will depend on exactly how you bootstrap RPM onto your system and what parts of the system you put into packages as well as on the specific OS you are using. The script vpkg-provides.sh can be used to generate a package which will satisfy the dependencies on your system. To run it you will need to create a specfile header for this empty package and run the progam with: --spec_header '/path/to/os-base-header.spec and if you wish to ensure that some directories are not traversed you can use the option: --ignore_dirs 'grep-E|pattern|of|paths|to|ignore By default the generated rpm will include a %verifyscript to verify checksum of all files traversed has not changed. This additional check can be suppressed with: --no_verify The result of running the script will be a specfile which will create a package continging all the dependencies found on the system. There will be one provides line for each depednecy. The package will contain none of the actual OS library files as it is assumed they are already on your system and managed by other means. Here is a example (truncated) of the provides lines used by one user of Digital Unix. (I have put several provides on the same line for brevity) provides: /bin/sh /usr/bin/ksh /usr/bin/csh provides: libc.so.osf.1 libm.so.osf.1 libcurses.so.xpg4 libdb.so.osf.1 provides: libX11.so libXaw.so.6.0 libXext.so libXm.so.motif1.2 libXmu.so provides: libdnet_stub.so.osf.1 libsecurity.so.osf.1 libpthread.so.osf.1 provides: libexc.so.osf.1 libmach.so.osf.1 libdps.so libdpstk.so The script vpkg-provides2.sh is underdevelopment as a more advanced version of vpkg-provides.sh which is aware of many different unix vendor packaging schemes. It will create one "dependency package" for each unix package your OS vendor installed. rpmfilename: ----------- If you plan on packaging for more then one OS you may want to edit /etc/macros or /usr/lib/rpm/macros and change the line which has rpmfilename to something which include both the %{_target_os} and %{_target_cpu}. This will cause the name of the generated rpm files to the operating system name as well as the architecture which the rpm runs under. The line to change looks like: %_rpmfilename %%{ARCH}/%%{NAME}-%%{VERSION}-%%{RELEASE}.%%{ARCH}.rpm you may wish to include both the %{_target_os} and %{_target_cpu} in the final base name, so that it's easier to distinguish between what package is appropriate for a particular arch-os-version combo. We suggest: %_rpmfilename %%{_target_platform/%%{NAME}-%%{VERSION}-%%{RELEASE}.%%{_target_platform}.rpm There is no %{_target_os_version} tag, so if you need to also distinguish between RPMs for certain versions of the OS, you can hard-code the version in the rpmrc on the build machine, so that .rpm files are generated with the version as part of the filename. For example when one user builds RPMs for Digital Unix 4.0b and 4.0d, optimization is important and he will build one set of RPMs for the EV4 processor and another set for the EV56 processor. He specifies both the OS version (if it's important, as it is for a few packages) and the processor version by default by setting a special rpmfilename: on the particular build machine. The "rpmfilename: "tag on one machine (Digital Unix 4.0d, EV56 PWS 433) looks like: rpmfilename: %{_target_os}/4.0d/%{_target_cpu}/%{name}-%{version}-%{release}.%{_target_os}-%{_target_cpu}ev56.rpm For package `foo-1.1', at build time that would translate into: osf1/4.0d/alpha/foo-1.1-1.osf1-alphaev56.rpm The hyphen between the %{_target_cpu} and ev56 is left out for compatibility with GNU Config.guess and because `alphaev56' looks more "normal" to people with an alpha than alpha-ev56 for someone on an Intel Pentium Pro would want `i586pro' over `i586-pro', but it does make parsing this filename by other programs a bit more difficult. GPG --- To use the signing features of rpm, you will need to configure certain rpm macros in ~/.rpmmacros: %_gpg_name %_gpg_path %(echo $HOME)/.gnupg