Package: buildd.debian.org Severity: normal dh-autoreconf depends on automake|automaken, where automake is a real package currently built by src:automake-1.15, and automaken is a virtual package provided by both automake and automake1.11.
When building packages in any suite other than experimental, automake is chosen, and automake1.11 is rejected. This is as expected. When building packages in experimental, it seems that about half the time the aspcud solver used in experimental will select automake1.11 to satisfy the dependency. This can cause a FTBFS, or even a silent mis-build. For instance, in flatpak 0.8.4-1: https://buildd.debian.org/status/logs.php?pkg=flatpak&ver=0.8.4-1&suite=experimental you'll see that 4/10 builds chose automake and succeeded, but 6/10 builds chose automake1.11 and failed. I have been working around this (in at least flatpak, glib2.0 and ostree so far) by adding explicit dependencies on automake, and I've opened a bug against dh-autoreconf asking the dependency to be changed to just automake. However, given that the behaviour of other suites is to ignore everything after the "|", I think it would be good if the aspcud scoring used for experimental tried to avoid selecting non-preferred alternatives. The aspcud scoring appears to be set up via dsa-puppet https://anonscm.debian.org/cgit/mirror/dsa-puppet.git/tree/modules/buildd/files/sbuild.conf and is currently: -removed,-changed,-new,-count(solution,APT-Release:=/experimental/) which is why I'm reporting this bug to buildd.debian.org and not sbuild - I think it's likely to be a problem with the optimization criteria more than with aspcud or sbuild. Sorry, I don't know how to spell "if you see foo|bar, prefer to install foo" in aspcud notation... S
