www-repository mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From robert burrell donkin <rdon...@apache.org>
Subject Re: [repo] /www/people.apache.org/repo/m1-ibiblio-rsync-repository/
Date Sat, 30 Dec 2006 12:20:17 GMT
On Fri, 2006-12-22 at 16:12 +0000, Steve Loughran wrote:
> On 22/12/06, Trustin Lee <trustin@gmail.com> wrote:

<snip>

> > > >   -POMs have schema declaration and are valid against the schema
> > > >   -POMs have no unexpanded ${project.version} values
> > > >   -all dependencies resolve. You cannot depend on sun stuff that
> > > > doesnt at least have a stub.
> > > >   -dependency graph is acyclic and no ambiguities (conflicting
> > > > artifacts at the same depth)
> >
> >
> > What would happen we develop a compiler?  There's a chicken and an egg
> > problem with developing a compiler.  Even if a project is not a compiler, a
> > project can have this kind of acyclic dependencies.  A very careful rule on
> > approving releasing an artifact with an acycle dependency though.
> 
> There already is an implicit loop in the build, as ant depends on an
> XML compiler, and xerces depends on Ant.
> 
> So I wouldnt ban stuff, just look closely at the dependency chain.

+1

i can think of other examples:

it's not uncommon for generators (for xmlbeans) to generate themselves
using a previous iteration

users dislike lots of small atomic jars. so, it's not unusual for a
single jar to be composed both core and optional components. it is not
uncommon for cyclic dependencies between jars to actually be standard
dependencies between components composed to form the jars.

this could be captured by sufficiently rich meta-data but IMHO we're not
close yet. so banning may exclude a *lot* of necessary artifacts since
this practise is especially common amongst low level libraries.

- robert

Mime
View raw message