This page documents cases where we needed to revert uploads in the development release due to regressions, or where we considered doing so. Its intent is to help us gather better data for future policy changes.

Add incidents at the top of this list, so that the list is in reverse chronological order. Use this template:


Description: Describe the regression here. Include links to bugs, test failures, or similar where appropriate.

Analysis: Who analysed the regression and what caused it? Include links to Launchpad source package releases and IRC logs where appropriate.

Outcome: What did we do? Include links to Launchpad source package releases where appropriate.


Description: New dependency of python-configglue on python-configparser causes reverse-dependencies to fail to execute (1198480).

Analysis: python-configglue was updated on July 1. A regression in software-center as a result was identified the same week, and fixed in response (the bug is technically in the caller, not in python-configglue). On July 8, a similar problem was identified in python-virtualenv. Barry Warsaw deemed the virtualenv issue not severe enough to warrant a revert. On July 9, a further problem was identified in ubuntuone-client. Sebastien Bacher escalated this last problem to Barry. The problem was investigated throughout July 9, but as of end of day on July 10 the fix had not been landed.

Outcome: seeing that the issue had not been resolved by EOB on July 10, Steve Langasek uploaded a manual revert of python-configglue to the previous upstream version.


Description: Ubuntu images take considerably longer to boot under some circumstances. Installer images fail to boot. (1124330)

Analysis: Moving whoopsie from talking to NetworkManager via GDBus calls to libnm-glib (which also uses DBus) triggered DBus service activation for NetworkManager and a hang inside ConsoleKit.

Outcome: While the cause for the hang is still not known, ev has reverted whoopsie back to using its old GDBus calls.


Description: Ubuntu images not buildable because language-selector-gnome depends on a package in universe (im-config) that has not gone through the MIR process

Analysis: Analyzed by infinity and slangasek on #ubuntu-release

Outcome: infinity reverted the change to language-selector-gnome and the Ubuntu seed.


Description: 1080713: CMakeFile.txt: Unknown arguments specified: "STREQUAL" "TRUE"

Analysis: new upstream release of CMake, changed GETTEXT_FOUND type from string to boolean, causing unity to FTBFS. xnox, Laney, infinity, didrocks & seb128 quickly analysed this and came to conclusion that unity should be adapted to do an if-check, instead of an explicit string comparison. At this time, no other affected packages were found. This actions were found to be insufficient, and CMake was subsequently patched using OpenSUSE patch (see bug details). Patch accepted upstream.

Outcome: No revert of cmake. Patch unity as per this one-liner which works with both old and new cmake versions. Also patch cmake to correctly set that variable. CMake patch accepted upstream, see the bugtasks on the 1080713.


Description: 1076906: Login failure due to regression in libgcrypt11.

Analysis: Iain Lane reported this to Adam Stokes on #ubuntu-devel, noting that the master bug understates the impact and that the duplicate is more serious. Colin Watson suggested reverting since the offending patch was simple, and Adam agreed.

Outcome: Patch reverted. Build and Tested by Iain Lane. Uploaded to raring.

UbuntuDevelopment/RevertLog (last edited 2013-07-11 06:47:57 by vorlon)