Re: buildd and virtual (build) dependencies :(((
On Fri, Jan 03, 2003 at 10:27:36AM +0100, Stefano Zacchiroli wrote:
> On Mon, Dec 30, 2002 at 07:20:47PM +0100, Sven Luther wrote:
> > Also this may cause problems with Stefano's plan for library depends.
>
> Well, the library dependencies aren't really a problem, there we can
> safely fall back on 'standard' versiones build dependencies (e.g.
> Build-Depends: libxxx-ocaml-dev (>= 1.2.3), libxxx-ocaml-dev (<< 1.2.4))
> that is ugly but works.
It is ugly, and are you sure you never uploaded a package or maybe just
built it forgetting to bump one of the two virtual dependencies.
> I proposed the virtual package solution for 'consistency' with the ocaml
> package solution.
>
> Returning to the 'ocaml' package issue, what does we 'risk' falling back
> to versioned build dependencies? Can something be broken by this falling
> back?
Well, the real problem, is i don't have a cristal ball which would
enable me to know when we will break compatibility (like the libdir
move), nor are we sure there will not be some library or other package
which will do the dependency wrong, and thus break all manner of things.
Now, the complaint is about build depends only, and we really don't need
those. Well at least not as urgently as the true dependencies.
The main problem is to stop a library, bytecode package from building
with the wrong version of ocaml, and thus be incompatible with all the
other ocaml libraries.
I have a feeling that this is not really a problem that is best solved
as we do it, but more with a so-name like dependency, which would need
to be implemented apart from the virtual provides, but which once solved
may help on more than just the ocaml packages, and be friendlier to the
autobuilders.
Now, the real problem is that i did a mistake when i did ocaml 3.06-13 &
-14, about the /usr/include directory, which caused some problems when
uninstalling the packages, which caused ocaml to be only partially
uninstalled, and thus the autobuilder believed ocaml-3.06-1 was already
installed while it was not.
I got a bug report from James Lamont (from the hppa autobuilder) and
after discussion he told me he did fix the problem, well, sort of since
i have not seen any hppa build since then.
Now, James Troup, which maintain the sparc and arm autobuilders did
contact me and told me i should no more use virtual build-depends, and
as i wanted some more info, and was a bit too insistent, he decided not
to speak to me anymore :(((.
So, basically, we are screwed in what concerns the sparc and arm
autobuilders, and should best find another solution, or maybe someone
else could approach James about the subject, maybe someone who had
problems with his packages also, and try to get to the heart of what the
problem really is. James was kind of saying that i am uninformed and that
the info is somewhere about what really breaks and what not, so i will
maybe go hunting for this, and maybe the autobuilder source code to see
what exactly went wrong.
Anyway, glibc did not yet hit testing, so there is no hurry, altough the
glibc bugs are better, the only remaining issue is a licence one with
the sun rpc included in libc.
Friendly,
Sven Luther
> Cheers.
>
> --
> Stefano Zacchiroli - Undergraduate Student of CS @ Uni. Bologna, Italy
> zack@{cs.unibo.it,debian.org,bononia.it} - http://www.bononia.it/zack/
> "I know you believe you understood what you think I said, but I am not
> sure you realize that what you heard is not what I meant!" -- G.Romney
Reply to: