Features/RemovePyXML

Remove PyXML from Fedora

Summary

The goal of this Feature is to remove the PyXML package from Fedora.

Owner

• Name: toshio
• Email: toshio@fedoraproject.org

Current status

• Targeted release: Fedora 19
• Last updated: Jan 11, 2013
• Percentage of completion: 45% (9 out of 20 packages resolved)

Detailed Description

PyXML has been dead upstream for many years. The main authors of it have stated this explicitly on the python-dev mailing list. It's successor, the python stdlib's xml module, has been getting bugfixes that PyXML has not. The current Fedora package maintainer (rrakus) asked about removing it in February, 2012.

The Python stdlib in python2.x also has the dubious behaviour of importing PyXML if it is installed and replacing its own code with PyXML's. In some cases, this leads to bugs (For instance: Eric bug, Docutils bug)as the old PyXML code does not cope with some usages that the version in the stdlib does.

We want to remove this package from Fedora. To do that we need to decide what happens to the packages that depend on it. After analyzing the packages that use it, most of them will be ported to another xml library as part of this Feature. However, a few packages will be dropped from Fedora instead.

Benefit to Fedora

Fedora will no longer have to carry old code with known bugs and no upstream maintainance. Packages that depended on that old code will be ported to a more responsive upstream codebase. Use of the python xml module will be more consistent as it will always be implmented with code from the python stdlib rather than code from the stdlib unless PyXML is installed on the user's system.

Scope

An F17 repoquery run revealed 20 packages that had deps on PyXML: https://fedoraproject.org/wiki/User:Toshio/Remove_PyXML#Deps I analyzed them and opened up bugs to deal with them in various ways. The open issues are tracked here: https://bugzilla.redhat.com/show_bug.cgi?id=843176

At the moment, Rawhide is down to 10 packages with a dep on PyXML.

$ repoquery -q --tree-whatrequires PyXML
PyXML-0:0.8.4-28.fc18.x86_64 [cmd line]
 \_  bkchem-0:0.14.0-5.pre2.fc18.noarch [1: PyXML = 0.8.4-28.fc18]
 \_  comoonics-base-py-0:0.1-7.fc18.noarch [1: PyXML = 0.8.4-28.fc18]
 |   \_  comoonics-cdsl-py-0:0.2-20.noarch [1: comoonics-base-py = 0.1-7.fc18]
 |   \_  comoonics-cluster-py-0:0.1-26.noarch [1: comoonics-base-py = 0.1-7.fc18]
 \_  comoonics-cdsl-py-0:0.2-20.noarch [1: PyXML = 0.8.4-28.fc18]
 \_  comoonics-cluster-py-0:0.1-26.noarch [1: PyXML = 0.8.4-28.fc18]
 |   \_  comoonics-cdsl-py-0:0.2-20.noarch [1: comoonics-cluster-py = 0.1-26]
 |   \_  osr-dracut-module-chroot-0:0.8-2.noarch [1: comoonics-cluster-py = 0.1-26]
 |   \_  osr-dracut-module-cluster-0:0.8-2.noarch [1: comoonics-cluster-py = 0.1-26]
 \_  grc-0:0.70-9.fc18.noarch [1: PyXML = 0.8.4-28.fc18]
 \_  katello-cli-common-0:1.1.10-1.fc19.noarch [1: PyXML = 0.8.4-28.fc18]
 |   \_  katello-cli-0:1.1.10-1.fc19.noarch [1: katello-cli-common = 1.1.10-1.fc19]
 |   |   \_  katello-cli-unit-tests-0:1.1.10-1.fc19.noarch [1: katello-cli = 1.1.10-1.fc19]
 \_  katello-cli-unit-tests-0:1.1.10-1.fc19.noarch [1: PyXML = 0.8.4-28.fc18]
 \_  libopensync-plugin-google-calendar-1:0.22-7.fc18.x86_64 [1: PyXML = 0.8.4-28.fc18]
 \_  openerp-0:6.1-2.fc19.noarch [1: PyXML = 0.8.4-28.fc18]
 |   \_  openerp-mpl-0:6.1-2.fc19.noarch [1: openerp = 6.1-2.fc19]
 |   |   \_  openerp-0:6.1-2.fc19.noarch [1: openerp-mpl = 6.1-2.fc19]
 \_  spacewalk-backend-tools-0:1.7.38-2.fc18.noarch [1: PyXML = 0.8.4-28.fc18]

The plan for each of those is:

• bkchem: remove the dep in our package as the code that uses PyXML appears to be unused.
• comoonics*: The maintainer hasn't replied to the bug report. The packages haven't been rebuilt by the maintainer since 2009 but the packager is status active. I plan to ping the maintainer about retiring the packages (and the dependent, osr-dracut-module package that he owns) in bugzilla and retiring them if I still don't get a response.
• grc: This package is obsoleted by gnuradio. I've pinged the maintainer in the bug saying I'll retire the grc package if there's no response.
• katello-cli-common: Port to python-dateutil has been applied upstream. New release, 1.3 should contain the fix which is due out in February.
• libopensync-plugin-google-calendar: No response on the bug. I plan to ping the maintainer about retiring the package if the maintainer is no longer interested or assisting in porting the package if the maintainer is interested.
• openerp: No response from maintainer on the bug. I plan to drop the dep from the Fedora package as analysis of the code looks like that will be sufficient.
• spacewalk-backend-tools: Fixed upstream. Maintainer is planning to package the fix with the 1.9 release, scheduled for the beginning of February.

Once those packages are dealt with, I'll retire and block the PyXML package in F19/rawhide.

How To Test

1. Check that there is no PyXML package in the repositories: On F19, repoquery -q --whatprovides PyXML => should return no output.
2. Check that nothing requires PyXML in the repositories: On F19, repoquery -q --whatrequires PyXML => should return no output.

User Experience

This Feature should barely be noticable to Fedora end users. People who look at the package set itself instead of simply using the applications inside of it will notice that the packages for PyXML, comoonics, and grc have been removed. If anyone uses comoonics, they will notice that it is no longer in Fedora. Users of grc should already have experienced the switch to gnuradio as the gnuradio package obsoletes grc.

Dependencies

Covered in the scope section.

Contingency Plan

We'll ship Fedora 19 with some packages still depending on PyXML and a PyXML package still in the repository. This Feature will be retargetted for F20.

Documentation

• https://fedoraproject.org/wiki/User:Toshio/Remove_PyXML

Release Notes

• Python programmers who use the xml module may find that a few pieces of it work differently than in the past. This is due to Fedora no longer shipping PyXML. This change allows the python stdlib's xml module to be visible to programmers (PyXML replaced the stdlib's code with its own). This was done because PyXML code is older and buggier than the stdlib code and unmaintained upstream.

Comments and Discussion

• See Talk:Features/RemovePyXML