[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: octave/4.4.0-3 appears to break dynare/4.5.4-2 in testing



Hi Sébastien,

On 14-06-18 10:08, Sébastien Villemot wrote:
>> I saw an upload of dynare 4.5.5-2, which apparently tried to fix part of
>> this, but now the test for dynare migration¹ fails with a segfault of
>> octave. Looking at the list of installed packages and their versions, I
>> suspect dynare is missing a *versioned* dependency on the latest octave.
>> What do you think?
> 
> First, note that all this happens in the context of the octave 4.4 transition,
> see #901155.

Fully ack.

> Dynare does de facto have a (kind of) versioned dependency, because it depends
> on liboctave5, which is provided by octave 4.4. And octave 4.4 has a Breaks:
> liboctave4.

No, the binary package liboctave5 is provided by the source package
octave 4.4. The binary package octave doesn't Provide anything as far as
the d/control file is concerned. Dynare does have a Depends on the
binary package octave, technically it looks to me this Depends should be
versioned.

> I guess what's happening here is that the test bed contains dynare 4.5.5-2,
> liboctave5 and octave 4.2.2-3 (and not 4.4), which is a combination which is
> technically possible but indeed broken.

So let's make sure the archive knows that.

> Maybe we could put the "Breaks: liboctave4" in liboctave5, but that would
> probably complicate such transitions.

Please don't do that. You have introduced the binary liboctave5
precisely because it enables us to keep the old library around for a
little while (called smooth transitions). Otherwise you could as well
have a liboctave package and be done with it.

> In any case, even though the dependency relationships are not perfect, nothing
> is actually broken, and the problem will disappear after the transition is complete.
I think the versioned Depends solution is correct. The reason why I
think we (as maintainers) should start caring more is that we need
autopkgtest to be able to figure this out by itself and for that it
needs to be fed with the right relations. This *is* making the
transition more difficult because the RT now has to judge if you made
the right assessment. It's so much easier if the test just passes.

Paul

Attachment: signature.asc
Description: OpenPGP digital signature


Reply to: