Thank you for that clarification, Ryan. I was a bit nervous when you declared JPMS as a requirement, but this explanation helps soften that requirement and makes it a little easier to digest... :-)
Caveat with my modest proposal below, I don’t mean an overall modularization of the platform as I know that some containers use OSGI etc. That would be big and contentious I imagine. What I would propose with the re-packaging:
All implementations provide module-info
Remove any split packages from implementations
This would enable Jakarta EE technologies to be used to build applications that are deployed using jlink. For example, I am thinking of situations where I would like to use CDI in a Function as a Service (FaaS).
Again, I am with Ryan here. We need to define modularity. It is past due.
That said, perhaps new features are more important at the moment since most implementations already have a notion of modularity.
On 5/7/2019 8:38 AM, Ryan Cuprak wrote:
One more thought, tie the namespace change to modularization (Java 9 modularization). A namespace change isn’t a feature that vendors can sell and might be hard to justify the engineering and testing effort. However, if we tie it to modularization, to enable compact cloud apps using jlink, I think it will spur adoption.
I’ve been trying to put together a CDI application (using WELD) on Java 12 that uses modularization with jlink and it hasn’t gone too well. Between split packages, JAR names that weren’t very friendly etc.
Following up with my individual hat on. If we can get this wrapped sooner than a month, great. If there's clear consensus on any particular direction, let's just go for it.
There are items in the "Decisions" sections. The thinking is that we should decide what proposal we want (either of these two or something else) within the timeframe allotted. Decisions associated with the proposal that need to be cleared before it can be executed may naturally take longer to make.
[Contents of this email represent discussions of the Jakarta EE Specification Committee over the last several meetings. The statements here have been reviewed by and represent the voice of the Jakarta EE Specification Committee]
As announced in the Update on Jakarta EE Rights to Java Trademarks[1] post on Friday, future modification of the javax namespace will not be allowed. While this is not what was envisioned when Jakarta EE started, in many ways this in our best interest as the modification of javax would always have involved long-term legal and trademark restrictions.
To evolve Jakarta EE, we must transition to a new namespace. The primary decisions we need to make as a community and industry are how and when. Given all delays and desires on everyone’s part to move forward as fast as possible, we would like to have this discussion openly as a community and conclude in one month. It is the hope that in one month a clear consensus emerges and can be presented to the Specification Committee for final approval.
In an effort to bootstrap the conversation, the Specification Committee has prepared two proposals for how we might move into the new namespace. These should be considered a starting point, more proposals are welcome. No final decisions have been made at this stage.
The guiding principle for Jakarta EE.next will be to maximize compatibility with Jakarta EE 8 for future versions without stifling innovation.
Other proposals should incorporate the following considerations and goals:
The new namespace will be jakarta.*
APIs moved to the jakarta namespace maintain class names and method signatures compatible with equivalent class names and method signatures in the javax.* namespace.
Even a small maintenance change to an API would require a javax to jakarta change of that entire specification. Examples include:
Adding a value to an enum
Overriding/adding a method signature
Adding default methods in interfaces
Compensating for Java language changes
Binary compatibility for existing applications in the javax namespace is an agreed goal by the majority of existing vendors in the Jakarta EE Working Group and would be a priority in their products. However, there is a strong desire not to deter new implementers of the jakarta namespace from entering the ecosystem by requiring they also implement an equivalent javax legacy API.
There is no intention to change Jakarta EE 8 goals or timeline.
Community discussion on how to transition to the jakarta namespace will conclude Sunday, June 9th, 2019.
It is envisioned binary compatibility can be achieved and offered by implementations via tooling that performs bytecode modification at either build-time, deploy-time or runtime. While there are open questions and considerations in this area, the primary goal of the discussion that must conclude is how do we move forward with future modifications to the APIs themselves.
Proposal 1: Big-bang Jakarta EE 9, Jakarta EE 10 New Features
The heart of this proposal is to do a one-time move of API source from the javax namespace to the jakarta namespace with the primary goal of not prolonging industry cost and pain associated with the transition.
Were we to take this path, a compelling approach would be to do the namespace rename and immediately release this as Jakarta EE 9. Additional modifications would be put into a Jakarta EE 10 which can be developed in parallel, without further delays.
Some or all Jakarta EE APIs under javax would move immediately into jakarta as-is.
Any packages not moved from javax to jakarta could be included in Jakarta EE, but would be forever frozen and never move to the jakarta namespace.
Jakarta EE 9 would be refocused as quick, stepping-stone release, identical to Jakarta EE 8 with the exception of the javax to jakarta namespace change and immediately released.
Jakarta EE 10 would become the new release name for what we imagined as Jakarta EE.next with only minor impact on timeline.
Work on Jakarta EE 10 could start immediately after rename is completed in the GitHub source and need not wait for the Jakarta EE 9 release to actually ship.
Pros:
One-time coordination and cost to the industry, including; conversion tools, users, enterprises, cloud vendors, IDE creators, platform vendors, trainers and book authors.
Easily understood rule: everything Jakarta EE 8 and before is javax, Jakarta EE 9 and after is jakarta
Consistent with the javax to jakarta Maven groupId change.
Highest degree of flexibility and freedom of action, post-change.
Industry would have the opportunity to begin digesting the namespace change far in advance of any major new APIs or feature changes.
Cons:
Largest upfront cost for everyone.
Specifications that may never be updated would still likely be moved.
Decision to not move a specification is permanent and therefore requires high confidence.
Decisions:
Which specifications, if any, would we opt not to move?
Would we take the opportunity to prune specifications from Jakarta EE 9?
Do we change the language level in Jakarta EE 9 to Java SE 11 or delay that to Jakarta EE 10?
Proposal 2: Incremental Change in Jakarta EE 9 and beyond
Evolve API source from javax to the jakarta namespace over time on an as-needed basis. The most active specifications would immediately move in Jakarta EE 9. Every Jakarta EE release, starting with version 10 and beyond may involve some javax to jakarta namespace transition.
The most active APIs would immediately move from javax to jakarta
APIs not changed or determined by the community to be unlikely to change would stay in javax
Jakarta EE 9 would be a mix of javax and jakarta packaged APIs
If a change was needed to a javax API post Jakarta EE 9 for any reason, that API would transition from javax to jakarta.
Jakarta EE 10 would be a mix of javax and jakarta packaged APIs, but a different mix than Jakarta EE 9.
At some point down the road, Jakarta EE xx, it may be decided that the migration from javax to jakarta is “done” and the final APIs are moved.
Pros:
Cheaper up front cost and reduced immediate noise.
No need to move specifications unless there is an immediately visible benefit.
Potential for less impact from API change overall.
Cons:
Prolonged coordination, cost and complexity to industry affecting conversion tools, users, enterprises, cloud vendors, IDE creators, platform vendors, trainers and book authors.
Use of restricted javax namespace prolonged.
Frustration of “always changing” packages may deter application developers and become a permanent perception of the brand.
Difficulty in remembering/knowing which Jakarta EE release an API was moved. “Is Connector javax or jakarta in Jakarta EE 11?”
Difficulty in keeping the industry in sync.
New implementations may find themselves having to deal with the javax to jakarta transition, unable to avoid legacy costs and therefore decide not to enter the space.
Transitive dependencies to other specifications may make incremental change difficult or impossible.
Restrictions on what Java SE implementation can be used for certification
Decisions:
Do we start small or start large?
Which APIs would immediately need to be changed?
Out of Scope
The following are very important community discussions, but do not require a decision in the time-frame allotted:
Roadmap or release date for any Jakarta EE.next that would contain new features
List of specifications that may be deprecated, pruned or removed from Jakarta EE.next, if any
Specification text around backwards compatibility requirements, if any
What profiles should be defined
However, depending on the path chosen, some of these topics may require immediate resolution before the chosen path can be executed.
_______________________________________________ jakartaee-platform-dev mailing list jakartaee-platform-dev@xxxxxxxxxxx To change your delivery options, retrieve your password, or unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/jakartaee-platform-dev
