From Fedora Project Wiki
(→‎Scope: still needs some work, but better :))
Line 1: Line 1:
{{admon/important | Incomplete! | This page is incomplete, and liable to change significantly before being proposed!}}
<div class="messagebox" style="float: right; margin-left: 10px;">__TOC__</div>
<!-- All fields on this form are required to be accepted by FESCo.
<!-- All fields on this form are required to be accepted by FESCo.
Line 8: Line 8:
= Feature Name =
= Feature Name =
Fully Testable Perl.
Fully Testable Perl.
<div class="messagebox" style="float: right; margin-left: 10px;">__TOC__</div>
== Summary ==
== Summary ==

Revision as of 06:49, 13 November 2008

Feature Name

Fully Testable Perl.


Provide the ability to re-test installed Perl packages (primarily CPAN-hosted packages) by automatically providing the test suites of each distribution as their own "perl-Foo-tests" subpackage, and providing a framework to test.


Current status

  • Targeted release: Fedora 35
  • Last updated: (DATE)
  • Percentage of completion: XX%

Detailed Description

Perl has a long history and culture of testing, which has resulted in a very high percentage of the packages on the CPAN (Comprehensive Perl Archive Network) containing significant test suites. While these test suites are executed at build time, a large number of modern Perl packages (e.g. Moose, DBIx::Class, Catalyst, etc) depend on a significant number of other packages, which may be owned by a different maintainer and updated independently.

That is, the tests results from %check are only valid so long as the environment in which they are run does not change from the set of packages the package is built and tested in. Additionally, there are many reasons one or more tests may be disabled under the buildsystem:

  • Network or $DISPLAY access is required; or some other operation not permitted / possible in the buildsystem is required.
  • Including packages the test requires would cause a circular build dep loop.
  • A highly specific test environment is needed, e.g. a database setup.

This feature proposes to automatically bundles the package test suite into a -tests subpackage without requiring additional maintainer work (a la debuginfo), which can then be installed to provide the capability to test (or retest) functionality post installation.

Benefit to Fedora

Being able to execute the tests of a Perl dist post-build will leverage existing code (the test suites) to allow the end user to perform sanity checks and rule out problems with the underlying code when tracking down bugs.

The net benefit to Fedora is to create an environment where end users (support, programmers, etc) will be able to execute the test suites of the 900+ CPAN dists Fedora packages natively, to track down bugs, test functionality the buildsystem was unable to provide, or just validate their environment.

Having the test suite used to build the package under koji would allow someone to test their modules without having to move away from the Fedora perl packages. It will:

  • Make finding regressions between multiple packages owned by multiple maintainers easier.
  • Provide the end user with the tools to test, debug and validate their environments with the same test suites originally used to build the modules.
  • Leverage existing RPM functionality to create the -tests subpackages without needing any package maintainer interaction, just as -debuginfo packages are created today.
  • Provide an additional source of documentation (e.g. Moose::Role and MooseX::Types::DateTime explicitly refers one to the test suite as the "best documentation").


For example, Moose is a modern metaobject-based class framework for Perl 5. It's been around for awhile, has a stable user API, and is extremely powerful. It has a slew of dependencies, as well as extensions in the MooseX::* namespace. Sometimes these extensions take advantage of the metaobject system -- those that do can be highly sensitive to backend changes in both Moose and Class::MOP. MooseX::AttributeHelpers is one such extension -- providing alternate attribute metaclass objects for different data structures. Indeed, Moose's build script explicitly checks for certain levels of modules it knows it will break.


There are three main components to this feature.


We should work with upstream for input on how to retain these tests in a coherent, consistent, usable form. (Upstream in this case would be the perl-qa mailing list.)

RPM macros implementation

The model for this effort is the "behind the scenes" packaging of debugging information into -debuginfo subpackages. While it doesn't appear that we have any sort of generic hook to tie into at this point, there are a couple things that could be done.

1. Work with upstream to take the %debug_package macro and reimplement it on top of a "generic" packaging hook. 1. Overload the %debug_package macro to automatically create -tests subpackages for Perl dists.

The first would be cleaner, but would require upstream involvement. Working proof-of-concept rpm macros overloading %debug_package have been created at Features/FullyTestablePerl/RpmMacros.

Test Execution Framework

Create a test-framework: e.g. to test perl-Moose and all its deps, an end user could execute "perl-tests Moose" and step back.

Test Plan

User Experience


Contingency Plan


Release Notes

Comments and Discussion