Hi, On Wed, Dec 11, 2019 at 08:18:39PM +0100, John Paul Adrian Glaubitz wrote: > numpy is currently BD-Uninstallable on a number of architectures due to > a circular dependency with scipy. > > Normally, this is fixed by adding a stage1 sbuild profile for boostrapping > the package. When building with the stage1 profile, the build dependency > on python3-scipy should be disabled (i.e. python3-scipy [!stage1]) such that > the numpy package can be built without scipy.
I disagree. stage1 usually is the last resort for such situations. It should not be added lightly. In fact, stage1 has been deprecated as a profile for this reason. It shouldn't be introduced at all anymore. > Can you add such a stage1 profile so I can bootstrap numpy for all the > architectures where it's currently BD-Uninstallable? Please don't. stage1 is undescriptive and in three years nobody will know what it was being added for. That said, the BD-Uninstallable situation and the circular dependency are real and they deserve a solution. Let's try the less heavy approaches first: * Maybe scipy is only used for testing numpy or vice versa? Then one could be annotated <!nocheck> and we'd be done. * Similarly, maybe one is only needed for building Arch:all packages? Then maybe it could be demoted to Build-Depends-Indep. With luck that could also mean we're done. * If that really doesn't help, we need to figure out how they integrate with one another. We need to describe the feature that they enable and add a descriptive profile for disabling that feature. So please try <!nocheck> and Build-Depends-Indep before asking for a profile. These standard measures are regularly tested and have meaning beyond bootstrapping. Only if that fails, document and understanding for what features actually cause the cycle and add a profile disabling such a feature in particular. Thank you. Also thanks for Ccing me. Helmut
