On Sat, 12 Jan 2019 01:24:54 +0200, Yavor Doganov wrote: > Emilio Pozuelo Monfort wrote: > > And yes, a combined list would be appreciated if the rebuilds need > > to be done in order. > > In previous transitions, the order was guaranteed because rdeps higher > up the stack were in state BD-Uninstallable until the library packages > they depend upon were rebuilt and installed. But in this cycle we > have relaxed the dependencies between the -base/-gui binary packages > to allow co-installation of different library versions, thus > supporting partial upgrades.
I would appreciate the release team's opinion regarding this experiment; I think it failed not only because of the regressions [1] when both library versions are installed but because testing migration is now blocked by regression in gnustep-sqlclient's autopkgtests. [1] https://alioth-lists.debian.net/pipermail/pkg-gnustep-maintainers/2019-January/004782.html Should we switch back to tight dependencies, thus allowing only one gnustep-{base,gui} version to be installed? I believe that's also the reason for the gnustep-sqlclient's autopkgtest failure in testing as libperformance0.5 (from src:gnustep-performance) linked against gnustep-base/1.25 is installed during the test. If the dependencies were tight the test would be skipped in testing. Are you going to force gnustep-base to testing or you want me to do something else? > What's important is that any package in this category which compiles > and runs tests at build time will FTBFS because the tests will abort. > This is precisely what happened to me with sogo when I tested it for > this transition; see #918795 for explanation (I closed the bug as it > turned out that sogo builds fine). TTBOMK, gnustep-sqlclient and sogo > are the only rdeps that have tests, Same problem here; gnustep-sqlclient and sogo failed to build on kfreebsd-amd64 as the builds were attempted with non-binNMUed gnustep-performance and sope/sbjson. Other problems so far: * lusernet.app was built against libpantomime1.2 on all release architectures + hurd-i386. Builds for debian-ports are fine AFAICS. Would you schedule another round with the appropriate extra-depends or you want me to make a sourceful upload? * All binNMUs on kfreebsd-i386 were in vain because gnustep-base is not installed there yet and gnustep-gui is not even built. * There are some installability issues on kfreebsd-* due to libheif not being rebuilt for the last x265 transition. Should I ask on -bsd/-wb-team for help here? * 3 packages FTBFS on GNU/Hurd due to a known issue with the buildds; I asked Samuel to give them back.
