Shmuel Fomberg <shmuelfomb...@gmail.com> writes:

> I disagree. A failure to install is a bug.

In principle, yes, but...

> as a module user I don't really care why it failed to install - it is as
> unusable to me as any other problem.
> as a module author, I would like to know if my dependencies are not
> dependable.

This reasoning is correct, the question is how to deal with it. The
single outstanding problem with this aspect of the dependency chain is
the multi dimensional version dependency. A module with 64 direct and
indirect dependencies is not rare. And the dependencies may all have
had, say, 16 different versions on cpan. Gives easily 1024 ways to fail
on a dependency. And among these there is a plentitude of irrelevant
combinations.

Or, as David Golden has put it: noise.

> Of course, visiting the deps site can show this information, but if the
> author isn't aware of the problem, why would he visit it?

I would ask the corresponding question to you: what is it that stops you
from reporting to the author via RT?

> Of course, I can decide that I don't care. similar to one of my modules
> that fail on Perl 5.6, and I don't really care about that version, so I
> ignored them.
>
> It was said in the original discussion that the failure, if marked as one,
> should be flagged differently then the regular pass / fail / na / unknown.
> depfail was suggested.

I have doubts this would be the right approach. I don't know how to do
it right but I would suggest as a start you report some of these
problems with the appropriate amount of detail in RT to the authors and
let us know about the tickets. If we see enough such user stories, maybe
we can invent something appropriate. Doing the right thing automatically
requires some exercise beforehand. Try to do it right manually and we
will probably see.

> And finally, if a module author upload a new version, and this new version
> have test failure, I think that it is possible to notify the authors of
> modules that depend on it without spamming them, but it is really tricky.

Fill the bugtrackers and keep track of the tickets yourself, so we can
learn from the experience.

-- 
andreas

Reply via email to