Hi Matthias, On Tue, Mar 26, 2019 at 04:08:48PM +0100, Matthias Klose wrote: > According to http://crossqa.subdivi.de/src/openjdk-11 there is one missing > issue > to cross build OpenJDK itself (#925467, please make ant and ant-optional M-A: > foreign). Note that the Debian tracker gives this M-A as well.
This is not correct. Technically, dose-builddebcheck only yields one issue at a time, which is why crossqa.subdivi.de (and buildd.debian.org) only give one reason even though there may be more reasons. There is a link to bootstrap.debian.net giving more details using josch's botch-buildcheck-more-problems. This gives a more accurate picture of dependency issues: https://bootstrap.debian.net/cross_all/openjdk-11.html > Looking further, we should make other build systems like gradle and groovy > M-A: > foreign as well, together with *all* their binary-indep dependencies. And > probably mark every binary-indep package as M-A: foreign ... Blindly marking all build systems Multi-Arch: foreign is bad as that tends to be false in surprising ways. A build system can have architecture-dependent output and that often means that it shouldn't be M-A:foreign. When it doesn't have architecture-dependent output, it should never occur in Build-Depends and moved to Build-Depends-Indep instead. In any case, M-A:foreign is not the silver bullet. However, I'm happy to help with figuring out whether M-A:foreign is appropriate. That tends to require a lot of multiarch-specific as well as domain-specific knowledge. The number of developers that have both tends to be small (or 0). Therefore, such tagging is usually resolved by discussion. > Time to add that to the java policy? Maybe not now, but maybe for bullseye. > But > a fix for ant would be appreciated now. Afaiui, lib*-java is in a similar role as python-* and lib*-perl. This is the infamous multiarch interpreter problem and it's not magically solved for java. That said, I'm happy to help with resolving these issues. Helmut
