Hi! On Mon, 2024-07-01 at 04:43:40 +0200, Guillem Jover wrote: > On Sat, 2024-06-29 at 17:29:49 +0200, Guillem Jover wrote: > > There's another regression that I found on various BSDs, where they > > already have GNU make 4.4.1, where the buildtools test fails for the > > AR and AR_FOR_BUILD checks, and I've not managed to track down yet > > what the problem is (and whether this is an intentional behavior > > change from its NEWS entry), but this needs to be fixed before a release > > because Debian will eventually hit that, and other systems with a new > > GNU make will also fail. If you could have a look that would be > > appreciated, otherwise I'll try during the weekend. > > I've tracked this down. The problem was that on the BSDs the dpkg > build system sets CC as the value detected at configure time, which is > cc, and for which there was no mocked binary, so the test code ended > up running as if it was cross-building, which cause the value to not > inherit from AR. I'll be adding a mock for cc, but that will not catch > CC being set to clang or another program, so this needs to be dynamic, > and so for now I'll explicitly override AR_FOR_BUILD to avoid the > cross-compilation check.
Actually simply overriding the CC to gcc in the mk.t does it and makes it into a controlled environment. > (I'm also going to add some CI runs for the BSDs to avoid regressions, > but probably on a branch for github.) Thanks, Guillem
