There is an assumption that MP Config wouldn’t accept release alignment as that is an explicit statement of policy i.e.
“MicroProfile creates and evolves specifications without regard to downstream consumer requirements (e.g. Jakarta). For example, specification consumers
will have to manage items like lifecycle, compatibility requirements, namespace, whether org.eclipse.microprofile is a suitable root package, etc.”
https://docs.google.com/document/d/1VQ5cvzhVhKYr27FKC1tVmf081eGLSNiuX-dhQ2BxItc/edit?pli=1&tab=t.0#heading=h.7e20q7f70ond
From: config-dev <config-dev-bounces@xxxxxxxxxxx>
On Behalf Of Roberto Cortez via config-dev
Sent: 21 February 2025 14:21
To: microprofile@xxxxxxxxxxxxxxxx
Cc: Roberto Cortez <radcortez@xxxxxxxxx>; Jakarta Config project developer discussions <config-dev@xxxxxxxxxxx>; Jakarta EE specifications <jakarta.ee-spec@xxxxxxxxxxx>
Subject: Re: [config-dev] [EXTERNAL] Re: [microprofile] Config for Jakarta EE 12 and MP.next
Hi,
Can you please clarify which standards MicroProfile Config cannot meet?
Also, why is there an assumption that MP Config wouldn’t accept some sort of release alignment? MP projects are free to release anytime they want.
Roberto
Yes, MP decided on the pull model. That means that the decision is solely on Jakarta EE, with the pull model, MicroProfile has decided they don't care about EE.
Copying/forking MP Config is not the only option, even with the pull model. Even with this model, Jakarta EE can depend on MP Config, or a core part of it (without CDI). But I doubt that this is the way to go
in reality. MP Config most probably cann't meet the standards of Jakarta EE. Another problem is that MP and EE have different release cycles, and, even if EE depended only on a core MP Config part, releases of MP, MP Config Core, and EE would need to be synchronized.
I doubt that the MP would accept that.
For the arguments above, the only reasonable options I see, are:
1.
MP Config participates in this process and splits to core part and CDI part, so that the core part can be moved to Jakarta EE Core Profile without package changes
2.
Jakarta EE copies the core part of MP config under the jakarta package prefix and adds it to the Core Profile. Jakarta EE can introduce a new CDI Config integration spec, or the existing CDI spec can specify integration in the non-lite
part, which is not part of EE Core Profile. This could still cause conflicts for implementations that provide both MP and EE, but we can hardly do anything about it
3.
Jakarta EE will not care about MP Config and will introduce its own API. This option makes sense only if all other options fail. We tried that already and haven't delivered in more than 2 years.
This basically boils down to extracting core parts of MP Config, and the question is whether to keep the package names or introduce jakarta package names, and whether MP helps and splits the MP Config spec or
not. The third option is probably the least desired right now.
Are any other options on the table?
To be honest the whole thing is a mess that should have been sorted out years ago. As one of the 4 vendors that actually has a compatible implementation of Microprofile 6 and above the two WG should have merged
years ago with innovation occurring in Jakarta ee as standalone specs then maturity being signalled by adoption into a platform spec.
"Downstream consumers will probably require a fork (with changing package names) to meet the downstream projects requirements."
This should not be controversial as it is rehashed continually.
While we have the same debates the market does not care and moves on.
I strongly disagree with forking MP Config to Jakarta with namespace changes. This introduces unnecessary headaches and namespace clashes in the IDEs. Why cannot Jakarta EE adopt MicroProfile Config as it is and
job done? No more hassle.
The namespace update is purely artificial and causes too many headaches and version matching. In order to continue the forking and namespace changes, can anyone show me the rule of Jakarta not allowing the use of MicroProfile specs first?
Thanks for forwarding this, although the quote of quotes of quotes makes it very hard to read.
I fully agree with Otavio's message that both are developed under the Eclipse Foundation, and users of a product don't really care, if it was developed in repository A or B.
However, what Roberto outlined earlier about continuous copying between MP Config and Jakarta Config makes absolutely no sense.
RC> After the initial copy/paste, how would things evolve?
My intention is that technical evolution would take place in the
MicroProfile Config project. In the event of Jakarta specific
1. Cross that bridge when we come to it.
2. Try to come up with solutions that are palatable to both communities, Jakarta
3. If absolutely necessary, we would define content in MP Config that
would have the proviso such as "this only takes effect in Jakarta
EE environments". There is ample precedent for such approaches. See
what we did with Faces when Validation was present (in EE) vs. not
present (such as in Tomcat).
RC> Would Jakarta keep the APIs in-sync?
Yes. Every time Jakarta needed a new version, they would pick up a
chosen release of MP config to give the copy/paste treatment to.
MicroProfile consumes Jakarta EE, so there is no MP application or platform without a Jakarta EE platform, at the very least the Core Profile. So Jakarta Config is expected to be available in every profile. If
the MP Config API was to co-exist with Jakarta Config forever, then applications would have to exclude one of them from their build system, otherwise they risk confusion or even mixing them in the same project with unforseen and unpredictable consequences.
Especially if a Jakarta EE application using Jakarta Config API also wanted to use certain MP features like OpenTelemetry, Health, etc. internally configured via MP Config, but potentially even a different version of the API, if e.g. MP 8.1 used a new version
of MP Config while Jakarta EE was still on 12 or 13 based on an older MP Config API.
If the API is a drop-in-replacement then nothing keeps the MP projects from using the Jakarta one after the next release. And it does not really matter, if it was maintained in
https://github.com/jakartaee/config or
Otherwise everyone, most importantly developers and users of both MP and Jakarta EE would face a "config hell".
I think that, in the case of config, other specifications can just specify the accepted config properties, regardless of how these properties are provided. The TCK could use system properties as a common config
source.
Then Microprofile umbrella can state that these properties are supplied by MP Config, Jakarta EE Platform would state they are supplied by Jakarta Config. Jakarta Core profile wouldn't need to include Jakarta
Config. Or it could, but then MicroProfile would state that Jakarta Config is not required. I'm sure there's a way to define all this in a simple way so that everybody is happy.
I think that makes a lot of sense. Each spec just says what it needs from a "Configuration Provider" and that provider would be either MP Config or Jakarta Config.
With all that, I share the same sentiment with Reza. I always hoped that MP would tend to donate APIs to Jakarta after they become stable, and then completely rely on the Jakarta version of the API.
I agree with this sentiment as well...
We need to also keep in mind the perspective of the end users... they want a "normal" configuration setup that "just works" with all of the different parts of the app. And it would be confusing to have two different
namespaces with the same classes. It's also confusing for the end user to have two different Config specifications, even if they're roughly the same.
I think that, in the case of config, other specifications can just specify the accepted config properties, regardless of how these properties are provided. The TCK could use system properties as a common config
source.
Then Microprofile umbrella can state that these properties are supplied by MP Config, Jakarta EE Platform would state they are supplied by Jakarta Config. Jakarta Core profile wouldn't need to include Jakarta
Config. Or it could, but then MicroProfile would state that Jakarta Config is not required. I'm sure there's a way to define all this in a simple way so that everybody is happy.
With all that, I share the same sentiment with Reza. I always hoped that MP would tend to donate APIs to Jakarta after they become stable, and then completely rely on the Jakarta version of the API. But that's
not necessary, and I think anything is better than the current state, when many implementations rely on MP config even for their Jakarta EE functionality, or even worse, MP parts support MP Config while Jakarta EE parts don't.
I am sure Ed and/or Jared will respond with their own thoughts - in the meantime let me share my two cents, including on some of the broader technical intricacies.
First a purely personal opinion independent of Microsoft. These are some of the intricacies of managing two platforms run by two separate working groups that in practice need to co-exist closely. It's the reason
some of us espoused the hope that common dependencies and possible sources of colliding resources to manage would only be in one direction with the Jakarta EE Core Profile being keenly mindful of the needs of both MicroProfile as well as the other Jakarta
EE Profiles (and hopefully in some distant future other non-Eclipse Foundation platforms that also depend on a stable/high-quality/minimal Core Profile). The hope would have been that platforms such as MicroProfile would deprecate APIs that are effectively
standardized onto the shared space of the Core Profile.
Setting aside the above purely personal opinion, if MicroProfile is very averse to supporting both Jakarta Config and MicroProfile Config, I don't think it's too hard to just keep Config out of the Core Profile
including the small handful of Jakarta EE APIs there (via spec profiles if needed). The major customer pain point is needing to configure the data/external infrastructure related Jakarta EE technologies using old style embedded XML or Java, which are mostly
not in the Core Profile anyway.
On 2/17/2025 11:17 AM, Roberto Cortez via config-dev wrote:
If we have separate namespaces and APIs are synced manually / selectively (and evolve independently), I’m guessing that any Jakarta specification would use its own Jakarta Config, and MicroProfile specifications
would use MP Config?
In practice, even if Jakarta Config is not part of the Core, if any Core specification adopts it, it would force Jakarta Config with its own API to the MP platform. Is this correct?
I plan to bring this up in the Jakarta EE Steering Committee. The technical debate aside, I think there are also process and branding/IP considerations here. For one, it's important to track down what the existing
consensus had been in Jakarta EE WG/Jakarta Config with regards to namespace. A sanity check from the perspective of branding/IP is also in order as these are in reality two different working groups.
I agree that the most prudent approach is avoiding needlessly introducing mutual inter-dependencies. Both MicroProfile and Jakarta EE should be able to independently evolve their configuration approaches when
needed. Separate namespaces with APIs synced manually/selectively when needed does that well.
On 2/11/2025 11:00 AM, Ed Burns via config-dev wrote:
Even with approaches that allow for mitigating the circular dependencies, I am strongly predisposed to prefer the repackaging approach that allows the content of MicroProfile config to be accessed by Jakarta specifications
using an entirely Jakarta namespace.
My responses are inline. We will discuss this issue in more detail at this Tuesday's MP Technical call. Please join if you are interested. The joining details can be found
here.
Thank you for the response.
1. A technical problem regarding introducing circular dependencies.
Yes, the issue is that MP Config is dependent on CDI. This has been discussed many times, and I believe MP is open to make the necessary adjustments and removing that restriction. In the previous Jakarta Config
initiative, that was already a goal. One of the things I’ve been advocating is for MP Config (or any other Config specification), to work standalone without any other dependency. This would allow any Java project to consume it without requiring the use of
the platform. As for the CDI, that can be an addendum to the specification or even be integrated into CDI itself.
+1. We can rework MP Config to use the CDI approach by dividing MP Config to MP Config Core and MP Config full, where MP Config Core will just have the spi part while the other part will include CDI. In this way,
Jakarta can simply include MP Config Core in the Jakarta Core profile. With this, I think Jakarta can simply include MP Config Core with the need of having Jakarta Config. With this approach, renaming microprofile-config.properties to application.properties
is not mandatory as the package namespace would contain microprofile.
2. A non-technical problem where Jakarta specs may not make
dependencies on MicroProfile artifacts.
Can we clarify what the problem is exactly? Is this something that can be worked out?
What I’m trying to understand is if we could work on the technical and non-technical issues that prevent Jakarta from adopting MP as is (without copying and renaming packages), and assuming we can have those fixed,
would Jakarta be able to use it as a regular dependency?
To my best knowledge, there was no restriction for Jakarta to use MP as long as the MP spec do not depend on Jakarta. If MP Config Core is consumed by Jakarta, there would be no circular dependency.
RR> Just using application.properties is a good idea indeed.
RR> I am sure Ed and Jared will respond, but I believe the idea here
RR> is to allow both Jakarta EE and MicroProfile to evolve
RR> independently in accordance with their own needs and also
RR> collaborate when best seen fit.
RC> A few comments inline. Thank you!
JA> This email is a follow up to the discussion at the 2025-02-04
JA> Jakarta EE platform call.
JA> In that call, we discussed an approach where Jakarta EE 12 could
JA> effectively use MicroProfile Config "as is" with some important
JA> non-technical accommodations.
JA> 1. The APIs for Jakarta Config would be the MicroProfile Config APIs,
JA> but with jakarta namespace. Yes, a copy/paste.
RC> After the initial copy/paste, how would things evolve?
My intention is that technical evolution would take place in the
MicroProfile Config project. In the event of Jakarta specific
1. Cross that bridge when we come to it.
2. Try to come up with solutions that are palatable to both communities, Jakarta
3. If absolutely necessary, we would define content in MP Config that
would have the proviso such as "this only takes effect in Jakarta
EE environments". There is ample precedent for such approaches. See
what we did with Faces when Validation was present (in EE) vs. not
present (such as in Tomcat).
RC> Would Jakarta keep the APIs in-sync?
Yes. Every time Jakarta needed a new version, they would pick up a
chosen release of MP config to give the copy/paste treatment to.
RC> What restricts Jakarta from using the API as-is?
1. A technical problem regarding introducing circular dependencies.
2. A non-technical problem where Jakarta specs may not make
dependencies on MicroProfile artifacts.
JA> 2. The implementation may delegate to the MicroProfile Config implementation.
JA> 3. The Spec document would be one-line: see the corresponding
JA> MicroProfile config spec document. May need additional text to
JA> talk about the difference in namespace and adding in
JA> jakarta-config.properties until a new MP Config version added that
JA> to its specification. See #5 below.
JA> 4. The TCK would be a copy/paste of the MicroProfile Config TCK
JA> and updating the name space and adding jakarta-config.properties
JA> 5. Need to introduce a new line in the ConfigSource (MicroProfile
JA> Config API) “Some configuration sources are known as default
JA> configuration sources. These configuration sources are normally
JA> available in all automatically-created configurations, and can be
JA> manually added to manually-created configurations as well. The
JA> default configuration sources are:
JA> 1. System properties, with an ordinal value of 400
JA> 2. Environment properties, with an ordinal value of 300
JA> 3. The /META-INF/jakarta-config.properties resource, with an ordinal value of 200
JA> 4. The /META-INF/microprofile-config.properties resource, with an
RC> How about dropping microprofile-config.properties (keep it for
RC> compatibility) and jakarta-config.properties, and use
RC> application.properties? This one is already used by many popular
RC> runtimes like Spring, Quarkus, and Micronaut, to name a few.
Roberto, that's a rad idea. I like it.
_______________________________________________
config-dev mailing list
config-dev@xxxxxxxxxxx
To unsubscribe from this list, visit
https://accounts.eclipse.org
--
_______________________________________________
config-dev mailing list
config-dev@xxxxxxxxxxx
To unsubscribe from this list, visit
https://accounts.eclipse.org
--
You received this message because you are subscribed to the Google Groups "MicroProfile" group.
To unsubscribe from this group and stop receiving emails from it, send an email to
microprofile+unsubscribe@xxxxxxxxxxxxxxxx.
To view this discussion visit
https://groups.google.com/d/msgid/microprofile/CABd%3DrHcUknfeWNWBR0CBgP495qVTUY4QZeX7ZsQqgbxAv3MHjA%40mail.gmail.com.
|
