Requirements¶
Given our principles, when updates are proposed, they must be accompanied by a strong rationale and present a low risk of regressions. These requirements therefore follow.
Real world impact. We won’t make a change unless there’s real world impact for users and the proposed change will address it.
Minimal changes only. We think this correlates well with minimising regression risk.
Public documentation. We must explain, in a way that is understandable to technical users not familiar with Ubuntu development, how we ensured the above.
Real world impact¶
Every change carries regression risk, and pushing unnecessary additional downloads to users harms their experience. If it’s a valid bug but nobody appears affected in practice and nobody is likely to be affected in practice, then a change to existing users is not acceptable.
Minimal changes only¶
[This section needs cleaning up]
Some take the view that sticking to upstream point releases is safer. We do not. We have experiences of regression across the archive, from code shipped by tens of thousands of upstream projects. We find that they vary considerably both in quality and in what upstreams consider acceptable to deliberately change.
Users expect us to maintain our standards across our packages.
It is increasingly common for upstreams to tell us that they only support their exact released sources, and minimal patching by distributions to fix bugs is not acceptable to them over taking a more recent upstream release wholesale.
While we value upstream expertise and opinion, this does not extend as far as to overrule our own release policies.
Even the simplest of changes can cause unexpected regressions due to lurking problems:
In bug 81125, the upgrade regression had nothing to do with the content of the change that triggered it: any user who had installed the libpthread20 package would encounter a problem the next time libc6 was upgraded.
In bug 309674, the failure was a misbuild due to timestamp skew in the build process. The underlying problem existed in the source package in the original release, but would only manifest in a small percentage of builds.
In bug 559822, a C++ library (wxwidgets2.8) was uploaded with no code changes. Due to an underlying toolchain change/bug, this caused an ABI change, causing a lot of unrelated packages to break (see bug 610975)
In bug 2055718, updating the package is the trigger for the bug, because the package update reconfigures tzdata.
We never assume that any change, no matter how obvious, is completely free of regression risk.
In line with this, the requirements for stable updates are not necessarily the same as those in the development release. When preparing future releases, one of our goals is to construct the most elegant and maintainable system possible, and this often involves fundamental improvements to the system’s architecture, rearranging packages to avoid bundled copies of other software so that we only have to maintain it in one place, and so on. However, once we have completed a release, the priority is normally to minimise risk caused by changes not explicitly required to fix qualifying bugs, and this tends to be well-correlated with minimising the size of those changes. As such, the same bug may need to be fixed in different ways in stable and development releases.
Public documentation¶
Consider what happens when something goes wrong. Suddenly we’re on the front pages of the industry media. How will we be judged? We think it’ll be on the basis of whether the choices we made appear reasonable, or irresponsible, with respect to users’ production systems. Critics as well as affected and therefore angry users tend to jump to the worst conclusions; that’s human nature. If on the other hand we already have a clear, documented explanation of the trade-offs we made, then suddenly we appear far more reasonable. Otherwise, those worst conclusions appear justified and public confidence in our product is damaged. Timeliness is important here; the media moves faster than we do, so it’s essential to have the documentation in place before a regression is published.
We must therefore document clearly the choices we have made and our justifications for them, such that a technical non-Ubuntu-familiar reader can understand it. This includes publication of this policy itself. For individual SRUs, we must clearly document how the individual SRU meets our policy. This should include:
The real world impact to users that explains why we are making the change in the first place.
What we are doing to minimise risk to existing users, including our analysis of the risks, and a QA plan that mitigates that risk as far as is reasonable.
For details, see Explanation → Reason for requirements → Documentation.