maven-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From John Casey <>
Subject Re: Merging in our Aether and Guice changes to Maven 3.x
Date Wed, 04 Aug 2010 15:54:19 GMT
On 8/4/10 11:03 AM, Jason van Zyl wrote:
> On Aug 4, 2010, at 10:35 AM, John Casey wrote:
>> On 8/3/10 2:21 PM, Jason van Zyl wrote:
>>> Hi,
>>> We have two major pieces that we, Sonatype, would like to merge into Maven 3.x
>>> The first are the Guice changes that we've been talking about for a while, and
the second is the introduction of Aether which is our second attempt at a stand-alone repository
API. The PMC is aware of Aether as Brian reported it in our quarterly report to the Apache
Board, but other developers who are not on the PMC and the community in general might not
know much about it.
>>> I just posted an entry giving a very high level description:
>>> There is a resources section at the bottom of the post for those interested in
the sources, issue tracking, wiki and mailing lists. As part of some of the research we are
about to embark on with Daniel Le Berre, Aether will likely look more like p2 as time passes
and as a final resting place the Eclipse Foundation is more likely then Apache. I know people
will ask so I'm answering that now. Sonatype is just about to fully move Tycho over the Eclipse
Foundation and we want to see how that goes. If that works, then M2Eclipse is next, and then
Aether will follow.
>>> At any rate we would like to merge these changes in and make plans to release
>>> So please let us know if you have any objections.
>> There's too much in this thread that I this is a tad distracting from the important
points, so I'm replying to the top post.
>> I _really_ appreciate all the work done in getting M3 into a usable form, and in
general I like the way Aether looks (I haven't had time to look into the guice shim yet).
I realize there are newer thoughts on repository design since Maven took its swing at things,
and we need to find a way to transition forward..."transition" because we have a large legacy
of artifacts already under the Maven repository format. HOWEVER, there are a couple things
here I'm pretty deeply concerned about.
>> 1. The repository format is a Maven concept. I'd argue that it's one of Maven's two
great contributions to the world of software (the other being a decent build tool). As such,
the Maven community should have some role in guiding the future of that format.
>> If Maven relinquishes all ownership of the API and implementation for the piece that
resolves artifacts, then we have no say in the future design of the repository Maven uses
as its lifeblood. Many people who aren't Sonatype people have spent time working on that de
facto specification, and they've shown the merit to earn a voice in guiding this
least, if it's going to be billed as a Maven-compatible Repository API.
>> 2. Jason, you mentioned sponsoring a Sat4j developer to work with Sonatype in the
future to improve Aether. What effect is this likely to have on the aether-api module? My
concern here is that we're talking about releasing Maven 3.0-beta-2 with a completely rewritten
API / implementation for one of the most pivotal parts of Maven. It's not that I don't trust
Benjamin and Kristian to produce high-quality code.
> Once the API is set for Aether it will be supported forever. Essentially the Eclipse
way of supporting APIs. Just like we're supporting the old Artifact APIs now with Aether being
the backing implementation. We're sensitive to external consumes.
>> What I'm actually worried about having Aether API drift AFTER we adopt it in Maven.
This will hamper anyone wishing to integrate with the Maven 3 core, whether that's Maven plugin
development or Maven embedding.
> It can't drift. Whatever is in place needs to be supported, all the plugins that use
the artifact resolution APIs as they stand here now still work with.
>> What I'd actually prefer to see is the Aether API published in some neutral location
where we have an iron-clad guarantee that we won't be locked out of its design. Then, put
the implementations wherever you think is best. IMO the key moving forward is to establish
a standard API for resolving artifacts. IMO, this is our great failure with Plexus, that we
depended directly on a container implementation, not on a container API.
>> Having a stable set of specifications define their interaction with Maven would make
plugin development and embedding MUCH better. In fact, I think establishing this practice
might be the single best contribution we can make to Maven in the near term.

All due respect, but that dodges the question of separating and 
standardizing the API from the implementation. It also dodges the 
discussion about who sets the design of the repository format and the 
API spec used to access it.

You're asking the Maven community to give up one of its greatest 
creations - the repository format that has become a de facto standard - 
and become completely dependent on a project whose future may be 
uncertain. It's easy to talk about companies as these fixtures in the 
market, but the fact is we're talking about giving complete control over 
the Maven repository API / format to a start-up. Start-ups are not known 
for their stability. Then, the company in control _may_ decide 
(unilaterally) to move the whole shebang to Eclipse. There's absolutely 
no role for Maven developers in this model, unless they go out and 
re-establish their merit on a new project.

I'm not talking about the merit to contribute implementation details - 
though the ASF concept of non-expiring merit argues strongly against 
losing access to that. What I'm talking about is the right to contribute 
to the design of the repository format, API, and SPI (now that I notice 
that's separate from the API). The language we use to share artifacts 
and metadata should not be under the sole control of a private entity.

Sure, there haven't been too many contributors to Maven 3. But how much 
of that has to do with the velocity of work done and paid for by 
Sonatype, the dramatic and repeated shift in direction by those paid 
contributions (mercury for example), the need to chase code from SVN to 
GitHub, to still other GitHub repositories, and the lack of discussion 
of the design of any of it?

It makes me uneasy to see how much this has become a skunkworks type of 
project, where much of the development takes place behind closed doors 
and then gets dumped on the Maven community.

Maven contributors established the foundational concepts (and code, from 
what I can tell) for Aether; Aether is a refactoring of that essential 
design and format. If you expect Maven to use Aether, then the Maven 
community deserves some say in the future of the format and API. That's 
my opinion.

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message