Am Fri, Apr 24, 2026 at 03:07:10PM +0200 schrieb Ludovic Courtès: > Making it opt-out is the better option, but whoever wants to tackle this > has to understand that testing it entails building the entire collection > and fixing issues one by one. > So perhaps a more reasonable option is to make it opt-in and to > gradually switch packages to the strict behavior.
I would prefer opt-out, since in this case repairing it is as simple as adding a "#:parameter #f" to a package recipe. It is similar to switching openssl to version 3.5, and then replacing any openssl inputs that pose problem by openssl-3.0. So the only drawback is that a world rebuild will be required, but I think the outcome will be worth it. With opt-in, my fear is that we will never finish, because people will forget as there is no immediate problem. And even if we succeed, we will in the end need to make an additional step of switching the default, which will entail dropping tons of "#:parameter #t" that are no more needed (and maybe adding a few "#:parameter #f"). This approach sort of assumes that everything builds before the change, but we are not that far from it anymore! And with a bit of luck, we do not have that many substitute* that do not apply anymore. Andreas
