Re: Build-Depend on virtual packages
On Tue, Jun 14, 2005 at 02:01:42PM +0200, Stefano Zacchiroli wrote:
> On Tue, Jun 14, 2005 at 11:09:14AM +0200, Goswin von Brederlow wrote:
> > No you don't. You have one per entry in the sources.list. As I said
> > even the same version can make apt-get build-dep fail. Two versions,
> > like when you just uploaded a new one that a local mirror doesn't have
> > yet or users with stable/testing/unstable entries, makes apt always fail.
> I made some test with current unstable apt and "apt-get build-dep". I
> set up a local repository of packages. First I tried to store there a
> version of ocaml (providing the virtual package on which several other
> packages build depend on) greater than that in unstable. In that case
> apt-get build-dep indeed fails.
> Then I tried to store in the repository the same version of ocaml than
> that in unstable. In this case apt-get build-dep work properly.
> I see no problem with autobuilders and virtual build-dependencies since
> they're supposed to be configured with just unstable repositories.
> However, to be kind to users it would be better to do as you suggest (a
> real metapackage) so that they can happily live with an additional local
> apt repository.
No, i refuse to do hacky workaround and drop our current elegant solution just
because others are too lazy or uninterested or whatever ot fix this. I think
we should never again let others dictate how we should handle this or not, and
if they get annoyed by imperfections or bugs in apt, dpkg or whatever, they
can just fix it.
/me still a bit angry about elmo's email about that a couple of years ago.
> Do you know if there are plans to fix this behaviour in future versions
> of apt? The expected behaviour for it should be to choose the greater
> version if several version of a single package provide a virtual
No, but now is the time to fix this, at the start of the release cycle, so
maybe we should fill an important bug report about this ?