Re: Are pure virtual Depends/Recommends entries bugs?
On Mon, Oct 31, 2005 at 10:55:18PM -0500, Daniel Burrows wrote:
> As I'm sure everyone knows, pure virtual entries in a Depends line are
> strongly deprecated, due to the fact that frontends have a tendency to pick
> a random provider of the package. What I'm not sure is if this is just "ugly"
> or actually considered a bug. In particular, I can't remember and would like
> to know:
> (a) Is a pure virtual entry with no prior alternative in a Depends line an
> actual bug? i.e., do we have a consensus on this?
> (b) If the answer to (a) is "yes", is a pure virtual *Recommendation* a bug?
> Rationale: Recommendations are intended to be installed by default, so
> their fields should be just as friendly to automatic tools as Depends is.
I don't think a pure virtual dependency (or recommendation) should be a
bug. If there is really no reason to choose one provider or another, why
choose one artificially?
I expect a tool like aptitude might pick the candidate with highest
priority, or else one at random (or the first one in some sort order).
A package maintainer could pick one this way too, but then it's one more
thing to maintain.
Steinar mentioned problems with pure virtual build-deps. Those should
work too. In the past bugs have been filed about those, but I think
those are a problem because of bugs in the auto-build tools, rather than
being an intrinsic problems with the deps.
Hamish Moffatt VK3SB <email@example.com> <firstname.lastname@example.org>