From Fedora Project Wiki
(Missing backslash)
(Section on patching gem versions in gemspec)
Line 188: Line 188:
Here we actually install into the <code>%{buildroot}</code>.  We create the directories that we need and then copy what was installed into the temporary directories into the <code>%{buildroot}</code> hierarchy.  Finally, if this ruby gem creates shared objects the shared objects are moved into the arch specific <code>%{gem_extdir}</code> path.
Here we actually install into the <code>%{buildroot}</code>.  We create the directories that we need and then copy what was installed into the temporary directories into the <code>%{buildroot}</code> hierarchy.  Finally, if this ruby gem creates shared objects the shared objects are moved into the arch specific <code>%{gem_extdir}</code> path.


==== Packaging for Gem and non-Gem use ====
==== Patching required gem versions ====
One common patching need is to change overly strict version requirements in the upstream gemspec.  This may be because upstream's gemspec only mentions versions that they've explicitly tested against but we know that a different version will also work or because we know that the packages we ship have applied fixes for problematic behaviour without bumping the version number (for instance, backported fixes).  To fix these issues, find the <code>add_runtime_dependency</code> call in the gemspec and patch it with the corrected version similar to this:
 
<pre>
Gem::Specification.new do |s|
  # [...]
-  s.add_runtime_dependency(%q<mail>, ["~> 2.2.19"])
+  s.add_runtime_dependency(%q<mail>, [">= 2.3.0"])
  # [...]
end
</pre>
 
=== Packaging for Gem and non-Gem use ===


{{admon/caution|Packaging for non-Gem use is no longer needed| Originally, rubygem modules were not placed in ruby's library path, so we packaged rubygems for use with both gems and non-gems.  This allowed code that used <code>require('ARubyModulePackagedAsAGem');</code> to function.  The current <code>rubygem</code> module adds all gems to the ruby library path when it is <code>require</code>'d. So, packagers '''must not''' create non-Gem subpackages of rubygems for new packages. Since the majority of Ruby packages in Fedora are now packaged as installed gems, you may need to patch the code to use <code>require('rubygem')</code> as early in the program as possible to ensure that these ruby components are properly found.  Packages for ruby gems that currently create a non-gem subpackage should be adapted to stop shipping the non-gem subpackage (with a [[Packaging:Guidelines#Renaming.2FReplacing_Existing_Packages | proper Obsoletes and Provides]] in the main rubygem package).}}
{{admon/caution|Packaging for non-Gem use is no longer needed| Originally, rubygem modules were not placed in ruby's library path, so we packaged rubygems for use with both gems and non-gems.  This allowed code that used <code>require('ARubyModulePackagedAsAGem');</code> to function.  The current <code>rubygem</code> module adds all gems to the ruby library path when it is <code>require</code>'d. So, packagers '''must not''' create non-Gem subpackages of rubygems for new packages. Since the majority of Ruby packages in Fedora are now packaged as installed gems, you may need to patch the code to use <code>require('rubygem')</code> as early in the program as possible to ensure that these ruby components are properly found.  Packages for ruby gems that currently create a non-gem subpackage should be adapted to stop shipping the non-gem subpackage (with a [[Packaging:Guidelines#Renaming.2FReplacing_Existing_Packages | proper Obsoletes and Provides]] in the main rubygem package).}}

Revision as of 14:37, 21 March 2012

Warning.png
This page is a draft only
It is still under construction and content may change. Do not rely on the information on this page.
Warning.png
Different guidelines for older Fedora and RHEL
In RHEL 4, 5, and 6 and Fedora before Fedora 16, a ruby version that does not have all the features needed to support these guidelines exists. If you're packaging for those distributions you should follow the old ruby guidelines instead.

There are three basic categories of ruby packages: ruby gems, non-gem ruby packages, and applications written in ruby. These guidelines contain sections common to all of these as well as sections which apply to each one individually. Be sure to read all the guidelines relevant to the type of ruby package you are building.

Ruby ABI

Each Ruby package must indicate the Ruby ABI version it depends on with a line like

Requires: ruby(abi) = 1.9.1
Questionmark.png
Fact check this
Is the intention that jruby can provide ruby(abi)? Or must the standard ruby interpreter always be installed even if the system uses jruby exclusively?

ruby(abi) is independent of ruby interpreter. jruby might satisfy the ruby(abi) check just as easily as the main ruby interpreter package.

Naming Guidelines

  • Packages that contain Ruby Gems must be called rubygem-%{gem_name}.
  • The name of a ruby extension/library package must start with the interpreter it is built for (ruby, jruby, etc) and then the UPSTREAM name. For example: ruby-UPSTREAM. If the upstream name UPSTREAM contains ruby, that should be dropped from the name. For example, the SQLite database driver for ruby is called sqlite3-ruby. The corresponding Fedora package should be called ruby-sqlite3, and not ruby-sqlite3-ruby.
  • Application packages that mainly provide user-level tools that happen to be written in Ruby must follow the general NamingGuidelines instead.

Macros

Non-gem ruby packages and ruby gem packages install to certain standard locations. The ruby-devel and rubygems-devel packages contain macros useful for the respective package types.

Macro Expanded path Usage
From ruby-devel; intended for non-gem packages. Alternate ruby interpreters will have equivalent locations (To be added to this table later)
%{ruby_vendorarchdir} /usr/lib{64}/ruby/vendor_ruby Place for architecture specific (e.g. *.so) files.
%{ruby_vendorlibdir} /usr/share/ruby/vendor_ruby Place for architecture independent (e.g. *.rb) files.
%{ruby_sitearchdir} /usr/local/lib{64}/ruby/site_ruby Place for local architecture specific (e.g. *.so) files.
%{ruby_sitelibdir} /usr/local/share/ruby/site_ruby Place for local architecture independent (e.g. *.rb) files.
From rubygems-devel; intended for gem packages
%{gem_dir} /usr/share/rubygems Top directory for the Gem structure.
%{gem_instdir} %{gem_dir}/gems/%{gem_name}-%{version} Directory with the actual content of the Gem.
%{gem_libdir} %{gem_instdir}/lib The lib folder of the Gem.
%{gem_cache} %{gem_dir}/cache/%{gem_name}-%{version}.gem The cached Gem.
%{gem_spec} %{gem_dir}/specifications/%{gem_name}-%{version}.gemspec The Gem specification file.
%{gem_docdir} %{gem_dir}/doc/%{gem_name}-%{version} The rdoc documentation of the Gem.
%{gem_extdir} %{_libdir}/gems/exts/%{gem_name}-%{version} The directory for binary Gem extensions.

Interpreter independence and directory macros

You may have noticed that the table above has different directories for non-gem libraries on different ruby interpreters but only a single set of directories for rubygem libraries. This is because code written for one ruby interpreter will often run on all ruby interpreters that Fedora ships (ruby, jruby, etc). However, some code uses methods that are not available on all interpreters (for instance, fork). Rubygems have a facility to ship different versions of the code in the same gem so that the gem can run on all versions of the interpreter so we only need to have one common directory for rubygems that all the interpreters can use.

The standard ruby %{vendorlib} directories lack this facility. For this reason, non-gem libraries need to be placed in per-interpreter directories and must have a separate subpackage (or package depending on upstream) for each interpreter that they support.

Dependencies

Due to having multiple types of libraries (gems and non-gems) there are two ways to specify dependencies.

Requires

If your package uses gem 'rubylibrary' or you're packaging a ruby gem that depends on the library in its .gemspec (See the section on building gems for more information) you Must use Requires: rubygem(rubylibrary) so the rpm will pull in the correct ruby library.

If your package just uses require 'rubylibrary' it's always fine to use Requires: ruby(rubylibrary). If you know that the ruby library is provided as a gem despite using require 'rubylibrary' you may use Requires: rubygem(rubylibrary) instead but this is not required as it forces packagers to know unnecessary details of how another ruby package has been created.

When packaging rubygems the gem itself carries some dependency information in the "gem specification". That specification may contain precise versions of dependent gems. If Fedora has packaged a different version of the gem you may have to adjust the .gemspec to work with the Fedora version of the package (see the section on building gems for more information).

Please ensure that the package works properly with the dependencies specified in the rpm spec file.

Provides

Non-gem ruby libraries Must indicate what they provide with Provides: ruby(RUBYLIBRARY) = VERSION. The string RUBYLIBRARY Must be the same as what is used in the require statement in a Ruby script that uses the library. The VERSION should be the upstream version of the library, as long as upstream follows a sane versioning scheme. For example, a Ruby script using the SQLite database driver will include it with require 'sqlite3'. The specfile for the corresponding Fedora package must contain a line Provides: ruby(sqlite3) = 1.1.0, assuming the package contains version 1.1.0 of the library.

Gem packages must have two provides:

Provides: ruby(RUBYLIBRARY) = VERSION
Provides: rubygem(%{gem_name}) = %{version}

This allows people who are packaging something that requires the gem to use either form of the dependency to get the proper package. Since their package may only have a require 'rubylibrary' in the code, there's no reason for them to have to figure out whether we've packaged it as a gem or a non-gem library. The rubygem(%{gem_name}) form of the dependency allows packages which depend on having the gem metadata to specify that the Fedora package must be packaging a gem form of the library, not a non-gem form of the library.

Libraries

These guidelines only apply to Ruby packages whose main purpose is providing a Ruby library; packages that mainly provide user-level tools that happen to be written in Ruby must follow the ruby applications guidelines instead.

RubyGems

RubyGems are Ruby's own packaging format. Gems contain a lot of the same metadata that RPM's need, making fairly smooth interoperation between RPM and Gems possible. This guideline ensures that Gems are packaged as RPM's in a way that such RPM's fit cleanly with the rest of the distribution and to make it possible for the end user to satisfy dependencies of a Gem by installing the appropriate RPM-packaged Gem.

Both RPM's and Gems use similar terminology --- there are specfiles, package names, dependencies etc. for both. To keep confusion to a minimum, terms relating to Gem concepts will be explicitly refereed to with the word 'Gem' prefixed, eg 'Gem specification' (.gemspec). An unqualified 'package' in the following always means an RPM.

  • Spec files must contain a definition of %{gem_name} which is the name from the Gem's specification.
  • The Source of the package must be the full URL to the released Gem archive; the version of the package must be the Gem's version.
  • The package must BuildRequires: rubygems-devel to pull in the macros needed to build.

Building gems

Since gems aren't a standard archive format that rpm knows about and they encapsulate both an archive format and information to build the ruby library building an rpm from a gem looks a little different from other rpms.

A sample spec for building gems would look like this:

%prep
gem unpack %{SOURCE0}
%setup -q -D -T -n  %{gem_name}-%{version}

gem spec %{SOURCE0} -l --ruby > %{gem_name}.gemspec

# Modify the gemspec if necessary with a patch or sed
# Also apply patches to code if necessary
%patch0 -p1

%build
mkdir -p ./%{gem_dir}

# Create the gem as gem install only works on a gem file
gem build %{gem_name}.gemspec

export CONFIGURE_ARGS="--with-cflags='%{optflags}'"
# gem install compiles any C extensions and installs into a directory
# We set that to be a local directory so that we can move it into the
# buildroot in %install
gem install -V \
        --local \
        --install-dir ./%{gem_dir} \
        --bindir .%{_bindir} \
        --force \
        --rdoc \
        %{gem_name}-%{version}.gem

%install
mkdir -p %{buildroot}%{gem_dir}
cp -a .%{gem_dir}/* %{buildroot}%{gem_dir}/

# If there were programs installed:
mkdir -p %{buildroot}%{_bindir}
cp -a .%{_bindir}/* %{buildroot}%{_bindir}

# If there are C extensions, mv them to the extdir.
# $REQUIRE_PATHS is taken from the first value of the require_paths field in
# the gemspec file.  It will typically be either "lib" or "ext".  For instance:
#  s.require_paths = ["lib"] 
mkdir -p %{buildroot}%{gem_extdir}/$REQUIRE_PATHS
mv %{buildroot}%{gem_instdir}/$REQUIRE_PATHS/shared_object.so %{buildroot}%{gem_extdir}/$REQUIRE_PATHS/
%prep

Since gems aren't an archive format that rpm recognizes, the first thing we have to do is explicitly use gem unpack to extract the source from the gem. Then we call %setup -n %{gem_name}-%{version} to tell rpm what the directory the gem has unpacked into. The -T and -D flags tell %setup that we've already unpacked the code

We then run gem spec to output the metadata from the gem into a file. This .gemspec file will be used to rebuild the gem later. If we need to modify the .gemspec (for instance, if the version of dependencies is wrong for Fedora or the .gemspec is using old, no longer supported fields) we would do it here. Patches to the code itself can also be done here.

%build

Next we build the gem. The first step is to create directories in which to temporarily install the built sources. We do this because the gem install command both builds and installs the code in one step so we need to have a temporary directory to place the built sources before installing them in %install. Because gem install only operates on gem archives, we next recreate the gem with gem build. The gem file that is created is then used by gem install to build and install the code into the temporary directory we created earlier.

%install

Here we actually install into the %{buildroot}. We create the directories that we need and then copy what was installed into the temporary directories into the %{buildroot} hierarchy. Finally, if this ruby gem creates shared objects the shared objects are moved into the arch specific %{gem_extdir} path.

Patching required gem versions

One common patching need is to change overly strict version requirements in the upstream gemspec. This may be because upstream's gemspec only mentions versions that they've explicitly tested against but we know that a different version will also work or because we know that the packages we ship have applied fixes for problematic behaviour without bumping the version number (for instance, backported fixes). To fix these issues, find the add_runtime_dependency call in the gemspec and patch it with the corrected version similar to this:

Gem::Specification.new do |s|
  # [...]
-  s.add_runtime_dependency(%q<mail>, ["~> 2.2.19"])
+  s.add_runtime_dependency(%q<mail>, [">= 2.3.0"])
  # [...]
end

Packaging for Gem and non-Gem use

Stop (medium size).png
Packaging for non-Gem use is no longer needed
Originally, rubygem modules were not placed in ruby's library path, so we packaged rubygems for use with both gems and non-gems. This allowed code that used require('ARubyModulePackagedAsAGem'); to function. The current rubygem module adds all gems to the ruby library path when it is require'd. So, packagers must not create non-Gem subpackages of rubygems for new packages. Since the majority of Ruby packages in Fedora are now packaged as installed gems, you may need to patch the code to use require('rubygem') as early in the program as possible to ensure that these ruby components are properly found. Packages for ruby gems that currently create a non-gem subpackage should be adapted to stop shipping the non-gem subpackage (with a proper Obsoletes and Provides in the main rubygem package).

Non-Gem Packages

Non-Gem Ruby packages must require ruby-devel package at build time with a BuildRequires: ruby-devel, and may indicate the minimal ruby version they need for building.


Build Architecture and File Placement

The following only affects the files that the package installs into %{ruby_vendorarchdir} and %{ruby_vendorlibdir} (the actual Ruby library files). All other files in a Ruby package must adhere to the general Fedora packaging conventions.


Warning.png
Site versus Vendor
Previously, %{ruby_sitelibdir} and %{ruby_sitearchdir} were used. However, as they are meant only for local installations, please use %{ruby_vendorlibdir} and %{ruby_vendorarchdir} instead.

Pure Ruby packages

Pure Ruby packages must be built as noarch packages.

The Ruby library files in a pure Ruby package must be placed into %{ruby_vendorlibdir} (or its proper subdirectory). The specfile must use this macro.

Ruby packages with binary content/shared libraries

For packages with binary content, e.g., database drivers or any other Ruby bindings to C libraries, the package must be architecture specific.

The binary files in a Ruby package with binary content must be placed into %{ruby_vendorarchdir} (or its proper subdirectory). The Ruby files in such a package should be placed into %{ruby_vendorlibdir}. The specfile must use these macros.

For packages which create C shared libraries using extconf.rb

export CONFIGURE_ARGS="--with-cflags='%{optflags}'"

should be used to pass CFLAGS to Makefile correctly. Also, to place the files into the correct folders during build, pass --vendor to extconf.rb like this:

extconf.rb --vendor

Applications

Applications are

  • programs that provide user-level tools or
  • web applications, typically built using Rails, Sinatra or similar frameworks.

The RPM packages must obey FHS rules. They should be installed into %{_datadir}. The following macro can help you:

%global app_root %{_datadir}/%{name}

These packages typically have no "Provides" section, since no other libraries or applications depend on them.

Here's an abbreviated example:

%global app_root %{_datadir}/%{name}

Summary: Deltacloud REST API
Name: deltacloud-core
Version: 0.3.0
Release: 3%{?dist}
Group: Development/Languages
License: ASL 2.0 and MIT
URL: http://incubator.apache.org/deltacloud
Source0: http://gems.rubyforge.org/gems/%{name}-%{version}.gem
Requires: rubygem(haml)
#...
Requires(post):   chkconfig
#...
BuildRequires: rubygem(haml)
#...
BuildArch: noarch

%description
The Deltacloud API is built as a service-based REST API.
You do not directly link a Deltacloud library into your program to use it.
Instead, a client speaks the Deltacloud API over HTTP to a server
which implements the REST interface.

%package doc
Summary: Documentation for %{name}
Group: Documentation
Requires:%{name} = %{version}-%{release}

%description doc
Documentation for %{name}

%prep
gem unpack -V %{SOURCE0}
%setup -q -D -T -n %{name}-%{version}

%build

%install
mkdir -p %{buildroot}%{app_root}
mkdir -p %{buildroot}%{_initddir}
mkdir -p %{buildroot}%{_bindir}
cp -r * %{buildroot}%{app_root}
mv %{buildroot}%{app_root}/support/fedora/%{name} %{buildroot}%{_initddir}
find %{buildroot}%{app_root}/lib -type f | xargs chmod -x
chmod 0755 %{buildroot}%{_initddir}/%{name}
chmod 0755 %{buildroot}%{app_root}/bin/deltacloudd
rm -rf %{buildroot}%{app_root}/support
rdoc --op %{buildroot}%{_defaultdocdir}/%{name}

%post
# This adds the proper /etc/rc*.d links for the script
/sbin/chkconfig --add %{name}

%files
%{_initddir}/%{name}
%{_bindir}/deltacloudd
%dir %{app_root}/
%{app_root}/bin
#...

%files doc
%{_defaultdocdir}/%{name}
%{app_root}/tests
%{app_root}/%{name}.gemspec
%{app_root}/Rakefile

%changelog
#...

Note, that although the source is a RubyGem, we have to install the files manually under %{_datadir}/%{name}, %{_bindir}, etc. to follow FHS and general packaging guidelines. If additional Fedora specific files (systemd .service files, configurations) are required, they should be

  • added via another %SOURCE tags
Source1: deltacloudd-fedora
  • placed into appropriate locations during %install stage
install -m 0755 %{SOURCE1} %{buildroot}%{_bindir}/deltacloudd

Running test suites

If there is test suite available for the package (even separately, for example not included in the gem but available in the upstream repository), it should be run in %check. The test suite is the only automated tool which can assure basic functionality of the package. Running it is especially helpful when mass rebuilds are required. You may skip test suite execution when not all build dependencies are met but this must be documented in the specfile. The missing build dependencies to enable the test suite should be packaged for Fedora as soon as possible and the test suite re-enabled.

The tests should not be run using Rake - Rake almost always draws in some unnecessary dependencies like hoe or gemcutter.

Note.png
Do not ship tests
Normally tests are only run at package buildtime. They should not be included in the binary rpms that users install on their systems. You may make an exception for this if the package makes public use of the test suite at runtime (for instance, an application package that has a --selftest command line switch that runs its testsuite.)

Testing frameworks usage

The Ruby community supports many testing frameworks. The following sections demonstrate how several to execute several of the more common test suites.

MiniTest

MiniTest is the default testing framework shipped with Ruby. It is unbundled in Fedora (you must use BuildRequires: rubygem(minitest)). To run tests using MiniTest you can use something like:

%check
testrb -Ilib test

Test::UNIT

To run tests using Test::Unit you must BuildRequires: rubygem(test-unit) and use something like this:

%check
testrb2 -Ilib test
Note.png
Test suites which run using Test::UNIT can typically also be executed by MiniTest. In that case, please prefer MiniTest.

RSpec

To run tests using RSpec >= 2 you must BuildRequires: rubygem(rspec-core) and use something like:

%check
rspec -Ilib spec

Test suites not included in the package

Sometimes you have to get the tests separately from upstream's gem package. As an example lets suppose you're packaging rubygem-delorean, version 1.2.0, which is hosted on Github. Tests are not included in the Gem itself, so you need to get them and adjust the specfile accordingly:

# git clone https://github.com/bebanjo/delorean.git && cd delorean
# git checkout v1.2.0
# tar -czf rubygem-delorean-1.2.0-specs.tgz spec/
Source1: %{name}-%{version}-specs.tgz

# ...
%prep
%setup -q -D -T -n %{gem_name}-%{version}

tar -xzf %{SOURCE1}

# ...

%check
cp -pr spec/ .%{gem_instdir}
pushd .%{gem_instdir}
# Run tests
rm -rf spec
popd

# ...

  • Make sure to include the version of the tests in the source name, so that when updating to new version, rpmbuild will fail because it won't find the proper %SOURCE1 (and this will remind you to update the tests, too).
  • Add the commands you used to get the tests into the specfile as comments. This will make it a lot easier the next time you will need to get them.
  • Run the tests as you normally would.

Some notes

  • Need to move the rubygems library into the per-interpreter directories as it is a non-gem library.
    • Need to rebuild ruby and rubygems package to use the new location
    • Need to rebuild rubygem packages to use the new interpreter-neutral rubygem library location.
  • Should there be more information about jruby, etc in the introductory portion (naming and
  • gem2rpm and rpmdev-newrpmspec can be updated with the new template