Yeah, it sounds right. The ids should definitely be normalized when
reading them so that matching classifier/extension combos map back to
the same types.
- Brett
On 02/03/2009, at 11:43 AM, Brian E. Fox wrote:
Seems logical to me. I don't know why there are these special mappings
in the first place. Sources is the other one that comes to mind... if
true you should handle that the same.
-----Original Message-----
From: Benjamin Bentmann [mailto:[email protected]]
Sent: Sunday, March 01, 2009 9:11 AM
To: Maven Developers List
Subject: MNG-4056, please comment
Hi,
I recently created a patch for MNG-4056 and would appreciate some
comments whether that's the proper way to address the issue.
In short, this issue is about the subtle difference between
<dependency>
<groupId>gid</groupId>
<artifactId>aid</artifactId>
<version>0.1</version>
<classifier>tests</classifier>
</dependency>
and
<dependency>
<groupId>gid</groupId>
<artifactId>aid</artifactId>
<version>0.1</version>
<type>test-jar</type>
</dependency>
i.e. <type>test-jar</type> vs. <classifier>tests</classifier>. While
both declarations work during builds of the consumer project that
run up
to the "install" phase, only the latter declaration will allow proper
dependency resolution from the reactor during an earlier lifecycle
phase
like "package" (see also comments in MNG-2045 [0]).
The cause for this difference is that resolution from the reactor
matches artifacts by their dependency conflict id which has the form
gid:aid:type:classifier. For the first dependency declaration above,
the
type is "jar" which doesn't match the type "test-jar" as used for the
attached test JAR.
In case matching by dependency conflict id fails, the proposed patch
falls back to another id I called repository conflict id (well, it
needed to have a name...). The important difference is that the
repository conflict id has the form gid:aid:extension:classifier, i.e.
uses the file extension instead of the type.
The rationale for this approach is the observation that the repository
layout does not use the artifact type but the file extension to
distinguish files. In other words, the ids
gid:aid:test-jar:tests:0.1
and
gid:aid:jar:tests:0.1
get mapped to the same physical file in the repository, namely
gid/aid/0.1/aid-0.1/aid-0.1-tests.jar
i.e. there's an aliasing effect in the repository which the patch
mimics
for dependency resolution from the reactor.
It solves the problem from a user's perspective but I'm not sure
whether
this kind of artifact identity is clean from a design perspective.
WDYT?
Benjamin
[0]
http://jira.codehaus.org/browse/MNG-2045?focusedCommentId=152064&page=co
m.atlassian.jira.plugin.system.issuetabpanels%3Acomment-
tabpanel#action_
152064
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
--
Brett Porter
[email protected]
http://blogs.exist.com/bporter/
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
