On 11 December 2012 23:26, Robert Matthews <rmatth...@nakedobjects.org>wrote:
Dan
1. A quick look at Eclipse and I'm struggling to see any seemingly random
placing. All my components are together and so on. What I do have though is
very long names as the group ids are also showing. This is set up in the
Eclipse plug-in settings. That said I'd rather have shorter project names,
and after making the change to shorter name everything is all over the
place. I'm happy with the proposed naming convention as I prefer the
benefit of the ordering. That said, the core modules will still appear
randomly.
Yeah, if you import with the advanced > template of [groupId]:[artifactId],
then everything is indeed ordered logically.
However, when Jeroen and I tried this on Monday, we decided not to use this
template, and just went with the default template, which is just the
artifactId.
Try it... you'll see it's quite a mess.
I was just playing around with Idea 12, and (I think) it also does the
equivalent, of naming its projects based on the artifact.
2 & 3. My previous expectations were that the core would be released at a
particular point and the components would then be released independently to
work against the core, and would probably be re-released again against the
same core as they get improved. So the components would evolve more quickly
than the core and the core would be improved in the background. (Although
isolated fixes to the core would be made without causing the components any
concern.)
Yes, no change... that's my expectation as well.
WRT the parenting, are we expecting the core modules to stay in step - all
released each time - or will specific modules be release as they needed?
All the core modules would be released together, yes.
An alternative strategy to the two options you suggested would be for the
parent pom to be released as part of the core, that way it would always be
ready for the components. Generally, though, I'm still trying to get my
head around how this could/would/should work. Maybe a skype call, were
share our thoughts, might be worth while at this stage.
I think this is basically the same as having everything parented by core's
pom. Indeed, the more I think about it, the more I am happy to go this
way. I'm trying it out in a branch locally to see how it flies (ie testing
to see if I release core and then if I can release one of the components).
3. Such numbering makes a lot of sense. And as you say a large number is
just a large number.
Ok, that's three of us in agreement. It might be worthwhile doing a formal
vote on this one; perhaps in a day or two.
Regards
Rob
On 12/10/12 20:17, Dan Haywood wrote:
All,
Following on from the previous discussion thread relating to renaming of
artifacts etc, I've been doing some further work on the release process,
up
to and including have now pushed our isis-core up to the Nexus staging
repo. Won't be calling a vote, just wanted to see if it all worked using
git (which it does, after a few modifications).
Anyway, we have a few alternatives in some of the detail; if you have an
opinion on any of this, please respond...
1. artifactId names: eg isis-jdo-objectstore vs isis-objectstore-jdo
This point came up in the previous thread, and at the time I was
ambivalent
as to which way we went on it. Now, though, I think we should use the
"reverse polish notation", ie "isis-objectstore-jdo".
The reason is mostly this: if a developer imports the entire project from
the root aggregator pom (oai:isis-all) into Eclipse, then the project name
will (by default at least) be based on the artifactId. This means that
artifacts appear pretty randomly in the Eclipse workspace.
Of course, the same argument applies if listing a directory, eg ls
WEB-INF/lib.
Opinions?
2. Should our releasable modules (core and each of the 20-ish components)
have a separate org.apache.isis:isis-parent as their parent (which defines
the maven-release-plugin configuration etc), or should they all use
org.apache.isis.core:isis as their parent.
The benefit of the former (and the code is mostly organized this way) is
that, well, it seems to follow what many other maven projects do. But the
downside is that we would need to release this isis-parent artifact first,
and we might tend to forget about it and not update it.
The benefit of the latter is we will always re-release it whenever there
is
a new release of core, and - as a bonus - all the child modules will
inherit version definitions by way of <dependencyManagement>.
Put another way: do we think that the lifecycle of our release process
itself be decoupled from the lifecycle of releases of Isis core? If yes,
we should have a separate org.apache.isis:isis-parent. If no, we should
just use org.apache.isis.core:isis as the parent of all releasable
modules.
Opinions?
3. What should our version numbers be?
This is quite a big one, possibly worth its own thread, but I was
wondering
if we should adopt semantic versioning [1]. Thus, this first release will
be v1.0.0 of the core; any bug fixes would be 1.0.1, 1.0.2, any backwardly
compatible enhancements (to the APIs) would be 1.1.0, 1.2.0, any breaking
changes to the APIs would be 2.0.0, 3.0.0, etc.
There's an argument that things aren't quite stable enough in core to do
this; the counter argument is that they are just numbers; if we end up
with
isis-core v 12.0.0 in two years time because of a succession of breaking
API changes, so what?
Opinions?
Thanks
Dan
PS: fyi, I'm hoping to be in a position to put a release out of isis core
for voting by the end of this week or beginning of next, latest.
[1] http://semver.org.
