Control: close -1
Control: tag -1 + wontfix

On Fri, Aug 03, 2018 at 10:54:44AM +0800, Ralf Treinen wrote:
> Package: autopkgtest
> Version: 5.4.1
> Severity: normal
> 
> Hi,
> 
> the flaky restriction does not seem to work in case of an unsatisfiable
> test dependency. This can be observed on ci.debian.net here:
> 
> https://ci.debian.net/data/autopkgtest/unstable/amd64/w/why3/757615/log.gz
> 
> In *this* particular case, the test dependency is not installable due to
> a stupid mistake by myself, but that is besides the point. The offending
> test why3+cvc4 is declared in debian/test/control with restriction flaky.
> Strangely enough, my local installation of autopkgtest 5.4.1 on my laptop
> works fine and treats its correctly:
> 
> autopkgtest [10:32:17]: @@@@@@@@@@@@@@@@@@@@ summary
> why3+alt-ergo        PASS
> why3+cvc3            PASS
> why3+cvc4            FLAKY non-zero exit status 1
> why3+spass           PASS
> why3+z3              PASS
> why3+coq             PASS
> 
> Maybe it has something to do with how ci.d.n sets up the test bed.

The real bug is the fact that autopkgtest is OK with uninstallable
dependencies. `flaky` is to be used when *the tests* are unreliable.
Having uninstallable dependencies should still be treated as a failure,
i.e. in your example above, The `why3+cvc4` test should actually be a
failure.

Now, there is the *valid* issue you reported in #905311 about how to
handle the case when you have multiple alternatives where each of them
might not be available on all architectures, but let's discuss that case
separately.

Attachment: signature.asc
Description: PGP signature

Reply via email to