Hi Samuel, Thanks for the response.
On 28-11-2019 13:35, Samuel Thibault wrote: > The package works only by luck, the warning does indicate that something > wrong is happening. I do think that in the case of hwloc it could very > well crash, because structures have changed between hwloc1 and hwloc2. Ack. > Personally I almost always leave the stderr check, because most often > they are a sign of something wrong happening, and we want to fix it > somewhere, even if it is harmless for the test at case. Ack, me too, except when it shows it's too sensitive. E.g. a lot of python tests fail otherwise on deprecation warnings which is *not* what we want in autopkgtests. > Perhaps what we are lacking here is a button for the maintainer of the > migrating package and the maintainer of the failing package to be able > to say "never mind, this is fine", and/or another to say "please retry > this with all to-be-migrated packages". That could cover most of the > cases, without the debian-release team having to delve into them. In most cases "please retry this with all to-be-migrated packages" can be achieved by adding appropriate versioned breaks (for the breaking package) or versioned (test) depends to the broken package. If people find this unpretty, I don't mind scheduling the check for now. What Ivo and I already have been considering, we should maybe accept the failure *if* the current broken package passes in unstable. But I have to think about what that means during transitions *before* the "broken package is rebuild. I'll make sure the package is tested with both packages from unstable, to confirm that everything is OK. Paul
signature.asc
Description: OpenPGP digital signature
