CI Requirements - Lessons Not Learnt?

[Kevin Kofler]

Nicolás Alvarez wrote:
> It is not true that users will be no worse off. An application could
> increase the dependency of libfoo to 1.3 and add code using a feature that
> was broken in 1.2. If you then revert the version bump, you get code that
> uses the new feature but allows libfoo 1.2, where it's broken. Users are
> now worse off than if you had stuck to the old version.

Sure, that can happen (that the code will build just fine against the old 
library, but not actually run properly), but that is not the common case. 
The common case is that the new library version is used for an API addition, 
and that reverting the dependency bump in the application will necessarily 
also revert the application code using the new library API (because 
otherwise it won't build) and restore the known state from the previous 
release of the application. (This can reintroduce bugs, but only ones which 
were already in the previous release.) As I understand it, this is exactly 
the situation we are in with KWin and xkbcommon now.

        Kevin Kofler