On 4 Jul 06, at 2:37 PM 4 Jul 06, Steve Loughran wrote:
Carlos Sanchez wrote:
The repository is as good as the users/projects make it. There's no
difference at all with using ant and including the wrong jars, maybe
the problem is that how to fix it in maven is not as easy as in ant.
If project A says it depends on B 1.0 and C says it depends on B 1.1,
there's a conflict in Maven, Ant and anything you want to use, the
difference is that Maven tries to do it for you, but you still can
override that behaviour.
It seems to me that the difference in ant, the duty to set up your
classpath belongs to the project alone, so you, the build.xml
author are the only one who can make a mess of your CP.
This also means that you are the only one who can help create a
correct classpath.
However, on any system with transitive dependencies, you are ceding
control of your classpath to other programs out there.
No, you are ceding control of the dependency chain to the projects
which you are depending on.
Even if you think you know exactly what the dependency graph of
your app is, an update to a new version of any your dependencies
can pull in new metadata, with a new set of dependencies, and
potentially a new classpath.
So how is that different then with Ant when you try a new version of
a dependency? You're going to have to go hunt down the dependencies
or copy in a new set of dependencies provided by that project. It
doesn't matter what you use to build if you change a dependency you
are subject to change.
This is not a maven-specific problem; anything that supports
transitive dependency logic can suffer from it. Gump and Ivy could,
though in both cases the descriptors tend to be hand-written tuned
to not exhibit the problem. (in gump most projects dont export
dependencies, as the default is compile-time-only).
Ok, I agree this is not a Maven specific problem.
You cannot maintain metadata for a growing number of projects and you
can't stay on top of projects changing all the time except in
reaction of problems. The projects themselves must provide useful
metadata and it is of value to them to do so.
> Right now we are in a good position with a huge number of users
trying
> and testing the metadata in the repository, and forcing projects to
> support maven by providing good data.
That still assumes that transitive dependencies are a good thing,
and that perfect metadata is achievable. I'm not sure about either
of those. I also think they're straying dangerously close to one of
the big software engineering tar-pits: versioning.
Transitive dependencies in the state they are in now can be
problematic. What can be done to fix this? The repository manager
running integrity checks, some visualization tools to help people see
the chains easily, and visualization tools to help detect conflicts.
For both of these things IDE integration will be key to making this
easy for users. Big challenges, and not trivial but certainly not
untenable. The metadata will never be perfect but right now I still
think it's far from being ideal because we have no real active
process of improving it on a large scale. Carlos puts in a _lot_ of
time trying to correct things and absorb changes submitted for
improvement but as mentioned in the previous message it's a matter of
education and automated tools running to point people in the right
direction.
-steve
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Jason van Zyl
[EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
