To copy comments from our discussion in IRC... In my case, blocking a migration until the entire Package-List is filled out or waiting for arch: all binaries if the source package lists all as an architecture does not solve the problem. I regularly put Britney2 in situations where a package doesn't have arch-indep binaries but that's acceptable, then the binaries show up later. The software presents different behavior compared to when arch-dep binaries are presented. With arch-dep binaries, the source package is re-evaluated. With arch-indep binaries, the source is ignored. That difference in behavior is what I really wanted to raise, but preventing Britney from being in this situation at all in Debian might be an acceptable path for Debian.
I think that with further discussion, we're not whether it's advisable to delete the section of `'_should_upgrade_srcarch' that ignores arch-indep binaries.` Doing so at least exposes #1064427 in current tests, so there are obviously other differences we can't well predict. I think that you should still go forward with a change which prevents this situation from occurring in the first place. I might still need to make a different change which causes *something* to come out in excuses, since I will still but britney in this situation on the regular and need it to at least alert me. Initial testing reveals that causes more tests to fail, but that's alright... I think that covers your questions. Thanks again for taking a look at this. I'm not sure at what point this bug should be closed now, though. When the situation is effectively prevented in Debian?
