Hi Martin, On 18-11-18 21:21, Martin Pitt wrote: > Paul Gevers [2018-11-18 19:48 +0100]: > It's less about "options" but more which foreign architectures are enabled in > the testbeds.
If I understand correctly, your fix for this bug only enables pulling packages from foreign architectures if those architectures are otherwise enabled in the testbed. > The common case is certainly to pull in i386 dependencies on > amd64, and if there's a good case to be made for it, our amd64 containers > could > certainly grow a "dpkg --add-architecture i386" (unless they already have it). They don't. > Other than that, tests of course always have the option to do that manually, > and install the foreign-arch dependency manually with apt-get instead of > d/t/control Depends:. Well, but if tests go the manual way, they don't need the fix to this bug anyways. Also, I really recommend against manual apt-getting except when apt-getting is part of the functionality that the package provides. I have seen multiple packages that get it wrong because the testbed of us (and Ubuntu) is weird because we mix unstable and testing and have the apt pinning in place. Autopkgtests shouldn't need to know the details of our set-up, especially because we may want or need to change them (like my attempt with disable-apt-fallback). So, I rather keep this bug open until autopkgtest really supports multi-arch test dependencies. (Of course your fix is still valuable in it's own right, I just don't think it fixes this bug). Paul
