brooklyn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Aled Sage <aled.s...@gmail.com>
Subject Re: [PROPOSAL] Refactor packages to avoid the split package problem
Date Thu, 06 Aug 2015 15:56:17 GMT
Hi Hadrian,

+1; makes sense to fix the split-package problem for the 0.8.0 release.

These should be different PRs from the org.apache.brooklyn, so that they 
can be reviewed accordingly.

I'll need to look again at the packages to give a definite opinion on 
naming. It's probably not as simple as org.apache.brooklyn.core.* in ./core:

  * There are utility classes in core named brooklyn.util.*, which
    should be compared with the brooklyn-util-common
    (main difference is that many utils in core have dependencies on
    other things in core, rather than being truly independent utils).
  * I'd like us to extract the core tasks framework
    (brooklyn.util.tasks) to its own module.

---
We should also review what is public (i.e. what packages should be 
exported). For example, is core's brooklyn.internal.* definitely not 
used by any other projects?

---
An alternative approach (definitely less appealing for me long term) 
would be to use OSGi fragments. I think something like that was done in 
the early days of OSGi'ifying jclouds, but don't recall the details off 
hand. There's an e-mail thread [1] and a blog post about it [2].

Aled

[1] https://groups.google.com/forum/#!topic/jclouds-dev/FdsbfELc1o0
[2] http://iocanel.blogspot.co.uk/2012/06/osgification-good-bad-purist.html


On 06/08/2015 15:50, Hadrian Zbarcea wrote:
> Hi,
>
> I opened a new jira [1] for this. While doing my refactoring in ./api 
> I discovered quite a bunch of places where we have split packages 
> among api, utils, core, etc.
>
> I would recommend using something like:
>
> org.apache.brooklyn.api.* in ./api
> org.apache.brooklyn.core.* in ./core
>
> and so forth. There are also a bunch of circular dependencies and 
> other smaller issues that we should correct.
>
> WDYT?
> Hadrian


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message