Hi, On 5/28/26 6:50 AM, Colin Watson wrote:
I'm not sure that this is something I would personally want to opt into as a maintainer, though. The trade-off is between buildd time and the likelihood of delivering more broken things to unstable.
Also, a lot of architectures don't have autopkgtests, and those are precisely the ones where assumptions the authors made about computing in general, like "all computers are little-endian and pointers fit into a 'long'", fall flat.
It seems to me that improving the bootstrapping situation for cases where loops can be broken using nocheck would be a safer way to go.
These packages should probably define a "stage1" profile as well that just happens to be equivalent to "nocheck".
Simon
