On Thu, Sep 11, 2025 at 10:39:08PM +0200, Nicolas Boulenguez wrote: > I think the right fix is to skip a candidate when > * the pattern mentions no explicit architecture (unchanged) > * Architecture differs from host (unchanged) > * Multi-Arch differs from 'same' (was: Architecture differs from 'all')
I think you mean s/differs from/equals to/ in the third item as that's what the code says. > > I have implemented this change in > https://salsa.debian.org/debian/dh-builtusing/-/commit/2a96171b2b724fb333ea61d291f7676fb2c20134 I'm not sure I fully understand the change here, but I have a few observations. The comment in line 77 still refers to filtering Arch:all. The comment in line 84 states that Arch:all packages cannot be M-A:same. This presently is correct, but there is an effort to make dpkg support that particular combination. This use raises the question of how dpkg exposes such packages in the dpkg-query output and whether it should print the "instantiated" architecture or "all" there. Would you mind keeping the check for Arch:all in order to better handle this future situation? With the change implemented (as is), my understanding is that libc-dev-bin can be used as that's M-A:foreign and then trips the multiarch condition to be kept. Thank you Helmut
