Having consistency is definitely in ours and Maven's best interest. We (us and the Maven guys) need
to examine how people manage their versions. Derby is clearly a case where 4 elements is standard.
I also agree with David that more dots isn't necessarily good but I've also noticed on the user
list that people like jars with and without version numbers. The point is that there are a number
of standards that are not necessarily consistent.
More inline...
David Jencks wrote:
On May 30, 2006, at 10:49 AM, Joe Bohn wrote:
Am I the only one concerned about this?
I think this is an important issue for our users. They won't have the
luxury to wait for a completely new Geronimo image to fix a problem
with an embedded component. They will also face these issues with
their own versioned application modules. I would appreciate your input.
Thanks,
Joe
Overall I think we need to keep consistent with maven on this. Having
slightly different rules for geronimo compared to maven will create more
confusion that any advantage we might gain. There's also some desire to
actually use the maven classes to work with repositories and artifacts,
which would make it even more likely that we follow maven conventions.
Being consistent is good. However, Geronimo is a separate project and I think we need to also be
defensive as well. IMHO that means consistency is good but remembering that users of Geronimo could
probably care less about our build tool. Notes on the user list seem to indicate the users see our
adoption of Maven paradigms confusing and of no additional value to them.
I certainly think that we need very clear documentation on what the
version numbers mean.
Yup
I think we might be able to work with maven to come up with some
additional possibilities. I think that our basic use case that maven
may not support too well is producing a private build of an artifact
that is already released from an outside project. Currently the only
ways to do this are by incrementing a build number (which can conflict
with an official later release) or by incrementing the incremental
version and including a qualifier: so
5.5.15
gets replaced by
5.5.16-MyPrivateBuild
I think the above is confusing. I would expect this to be a component that is modeled on 5.5.16 +
something else. I think something like:
5.5.15-*patch*n...n implies we started with 5.5.15 and this jar is that + something. Using a
specific suffix like PATCH and some number would eliminate and ambiguity. I think if Maven defined
that naming convention that would help us all.
which will in turn be replaced by any official 5.5.16 release from the
project.
I'm not sure what problems this last might cause.
Perhaps we could lobby maven for a special qualifier keyword that is
after all build numbers? e.g. 5.5.15-PRIVATE-23455
Yup...I prefer PATCH; the question is does this imply a whole replacement for the component. I
would think that is easiest but thought we should clarify.
A couple more comments inlne
thanks
david jencks
Joe Bohn wrote:
I'm trying to get my head around the way that we make a version
selection when multiple versions of a package are available. This
will be important as users need to include different versions of
packages beyond what geronimo bundles or if they need to override a
package with a local version.
I was working with the tomcat jars and so I was looking for ways to
drop in a modified version of the jars and have them picked up
without removing the 5.5.15 versions. Here are the items that I
tried and which was chosen when compared to 5.5.15
1) 5.5.15.1
- Apparently any version with more than 2 dots is considered invalid
and so the entire version is considered to be a qualifier (with a
null for the major, minor incrementalVersion, and build - basically
treated as 0.0.0-"5.5.15.1"). Any valid version is considered newer.
- 5.5.15 is chosen over 5.5.15.1
- 5.5.10 is chosen over 5.5.15.1
I'm not sure more dots are a good thing.
2) 5.5.15-1
- The "-" is used to specify a qualifier or buildnumber. Since the
value after the dash was numeric, it was considered to be a
buildnumber. It appears that a build number is always considered
newer than a package without a buildnumber.
- 5.5.15-1 is chosen over 5.5.15
3) 5.5.15-01
- The code (Version.java) treats the leading "0" as a special case.
This makes the last part a qualifier rather than a build number. Any
qualified version is considered to be lower than a non-qualified
version (such as with -SNAPSHOT). Anybody know why this special
check for "0" is in there?
- 5.5.15 is chosen over 5.5.15-01
4) 5.5.15-alpha
- If the portion following the "-" starts with an alphabetic
character then this last portion is considered a qualifier. Once
again, the qualified release is considered older than the same
version non-qualified.
- 5.5.15 is chosen over 5.5.15-alpha
First, we need to document this behavior very clearly for users that
need to replace packages we ship (or their own packages included in
the repo).
Second, I would like to propose some changes:
- IMO a qualified release should generally be considered *newer*
than a non-qualified release. I think SNAPSHOT would be the only
exception. Right now we treat that exception as the rule for all
qualifiers. I think we should add specific code for "SNAPSHOT" and
have all other qualified releases take precedence over a
non-qualified release. I can imagine users wanting to add
myjar-1.1-patch1.jar to replace myjar-1.1.jar.
A lot of people use -DEV which is definitely before a plain build
number. I don't think moving most qualifiers to after build numbers
will fly: I think a special keyword for this might.
- I think we should treat a third "." to be the logical equivalent
of a "-" in the version. Most users would expect 5.5.15.1 to be
major version 5, minor version 5, incremental version 15,
build/rev/patch/whatever 1 and consider this to be newer than 5.5.15.
See #1 above for how we really treat 3 dots. Providing 5.5.15-1
gives substantially different results than providing 5.5.15.1 which
is not intuitive.
I don't think these definitely need to have the same meaning. I think
that allowing 3 dots means we should allow any number of dots.... and
I'm not sure we really need the resultant complexity.
Joe
--Joe Bohn
joe.bohn at earthlink.net
"He is no fool who gives what he cannot keep, to gain what he cannot
lose." -- Jim Elliot
