Re: Bug#908757: r-cran-processx: autopkgtest regression

[Ian Jackson <ijackson@chiark.greenend.org.uk>]

Paul Gevers writes ("Re: Bug#908757: r-cran-processx: autopkgtest regression"):
> On 17-09-18 15:37, Ian Jackson wrote:
> > Paul Gevers writes ("Bug#908757: r-cran-processx: autopkgtest regression"):
> >> With a recent upload of r-cran-processx the autopkgtest of
> >> r-cran-processx fails in testing when that autopkgtest is run with the
> >> binary packages of r-cran-processx from unstable. It passes when run
> >> with only packages from testing.
> > 
> > Was there a slip in that description ?
> 
> No, but I can see it confuses a bit. But improvements to the text (which
> I use in a template) are welcome. Very much so. What I meant to say is,
> take src:r-cran-processx and all it's binaries from unstable, add it to
> testing and run the testsuite (in this case, from unstable).

I see you have tried to clarify this in recent emails.  I think the
result is simply far too convoluted a sentence.  It is really hard to
understand.

How about a table:

The recent upload of r-cran-processx seems to have introduced a
regression:

                     passing                   failing

  r-cran-processx    <version>                 <version>
   autopkgtest       (currenlty in testing)    (currently in unstable)

  r-cran-processx    <version>                 <version>
   binary packages   (currenlty in testing)    (currently in unstable)

  some-dependency    <version>                 <version>
   binary package    (currenlty in testing)    (currently in unstable)

  other packages     those from testing        those from testing

or something ?

> > I think that it is not unusual for the autopkgtest of X (1.0) to
> > fail when run against the binaries for X (1.1).
> 
> That is not what I meant here. I agree with you that that would not be a
> big issue. Nowadays, with my britney improvements, this should be only
> the case with uncatched versioned dependencies or breaks/conflicts.
> 
> > Doing that test is not really useful for testing migration; the
> > testing migration should run the tests from the new X (1.1).
> 
> Ack, I think that happened also.
> 
> > It is probably not worthwhile anyone declaring an explicit version in
> > the test dependency.
> 
> I don't see what you mean here even.

It would be possible to to avoid this non-big issue by, whenever a
test is changed in a way that makes it fail with old binaries, adding
a versioned test dependency on the new binaries to the affected test.

I am saying that I think this is not worthwhile (and possibly even
harmful, although I haven't thought about it that clearly).

Ian.

-- 
Ian Jackson <ijackson@chiark.greenend.org.uk>   These opinions are my own.

If I emailed you from an address @fyvzl.net or @evade.org.uk, that is
a private address which bypasses my fierce spamfilter.