And the JCP model has worked quite well for 20 years.
I really don't want to move closer to a position where the spec is
divorced from any implementation. What if there is never a
final implementation that corresponds to a spec?
Experimentation and verification of a spec should occur before
the spec is finalized, not after. I'd much rather take more time to
finalize a spec and allow multiple implementations to "verify" the
spec. That's much more likely to produce a good result than rushing
to finalize a spec before even one implementation is final.
Kevin Sutter wrote on 6/6/19 11:47 AM:
I don't agree
that
a final implementation has to be available immediately. That
was
the Glassfish model (for the most part). But, it doesn't have
to
continue like that. We've had excellent results with
MicroProfile.
We provide a link to some "workable" implementation for
immediate experimentation and verification. The real, final
implementations
are produced at a later time by the various vendors and
development communities.
That model has worked quite well.
---------------------------------------------------
Kevin Sutter
STSM, MicroProfile and Jakarta EE architect
e-mail: sutter@xxxxxxxxxx Twitter: @kwsutter
phone: tl-553-3620 (office), 507-253-3620 (office)
LinkedIn: https://www.linkedin.com/in/kevinwsutter
From:
Bill
Shannon <bill.shannon@xxxxxxxxxx>
To:
Jakarta
specification committee
<jakarta.ee-spec.committee@xxxxxxxxxxx>,
Kevin Sutter <sutter@xxxxxxxxxx>, Scott Stark
<sstark@xxxxxxxxxx>
Date:
06/06/2019
12:50 PM
Subject:
[EXTERNAL]
Re: [jakarta.ee-spec.committee] Compromise for approval of
Jakarta EE 8
specification projects
At a minimum it needs to be a
distinguished
and stable release of some sort. It definitely can't be "last
Thursday's nightly build".
When the spec is finalized, we want to encourage the community
to download
the compatible implementation(s) and start using them
immediately.
If there is no "final" implementation available, that's going
to discourage people from using it, and decrease the value of a
final release.
I would rather delay the spec approval until a final release is
available,
and continue to make "release candidate" or "proposed final
draft" versions available for testing until then.
Kevin Sutter wrote on 6/6/19 5:47
AM:
Agree, Scott.
It doesn't have to be final. But, it does have to be
accessible
via some open-source repo (milestone driver, rc driver, some
type of defined
driver).
---------------------------------------------------
Kevin Sutter
STSM, MicroProfile and Jakarta EE architect
e-mail: sutter@xxxxxxxxxx Twitter:
@kwsutter
phone: tl-553-3620 (office), 507-253-3620 (office)
LinkedIn: https://www.linkedin.com/in/kevinwsutter
From: Scott
Stark <sstark@xxxxxxxxxx>
To: Jakarta
specification committee <jakarta.ee-spec.committee@xxxxxxxxxxx>
Cc: Kevin
Sutter <sutter@xxxxxxxxxx>
Date: 06/05/2019
10:34 AM
Subject: [EXTERNAL]
Re: [jakarta.ee-spec.committee] Compromise for approval of
Jakarta EE 8
specification projects
I don’t agree that a final release of a compatible
implementation should
be required for a spec to be approved. The point of at least one
compatible
implementation is to validate that the spec can be implemented.
All of
the window dressing of a final release has nothing to do with
implementation
details.
On Jun 4, 2019, at 6:30 PM, mdenicola@xxxxxxxxxxxwrote:
Personally
I think a final release should be required, but it's a good
discussion
topic.
MikeD
From: jakarta.ee-spec.committee-bounces@xxxxxxxxxxx<jakarta.ee-spec.committee-bounces@xxxxxxxxxxx>
On Behalf Of Bill Shannon
Sent: Tuesday, June 04, 2019 3:45 PM
To: Jakarta specification committee <jakarta.ee-spec.committee@xxxxxxxxxxx>;
Kevin Sutter <sutter@xxxxxxxxxx>
Subject: Re: [jakarta.ee-spec.committee] Compromise for
approval of
Jakarta EE 8 specification projects
Kevin
Sutter wrote on 6/4/19 1:05 PM:
Hi Bill,
We're in a classic chicken-and-the-egg situation... It's
difficult
to have any Compatible Implementation that uses the APIs that
are being
submitted for review. We have this situation with every
MicroProfile
release. We require a Compatible Implementation for the Release
Review,
but it doesn't have to be a final GA product. It has to be an
accessible
open-source project that demonstrates the functionality defined
by the
Spec and API and it passes the defined TCK. But, it doesn't
have
to be a final X.Y release.
At
the JCP we required the implementation to be released as a
"final"
product soon after the spec was approved. (It usually took
hours
to weeks after the approval to release the implementation.)
Allowing
some slop between the spec approval and the implementation
release solved
the chicken-and-egg problem.
I guess we need to decide what we require for the JESP. I was
assuming
it would be the same. I don't think it would be appropriate
to approve
the spec with only a "day in the life" of an implementation
that
may never be recreated. Perhaps a final release is not
required,
but it seems like it should at least be some sort of
distinguished milestone
release. Personally I think a final release should be
required, but
it's a good discussion topic.
What is even more unique with this Jakarta EE 8 release is that
we want
to make a big splash with this initial release. When we
announce
Jakarta EE 8, we want to also announce that we have several
Compatible
Implementations -- Eclipse Glassfish, OpenLiberty, and Wildfly
(at least,
maybe there are others). So, how can we accomplish this when
the
Specs, APIs, and even the TCKs are still going through the final
Reviews?
It's
not our plan to hold up this release until other compatible
implementations
are available. If you think we need to do that, you should
bring
that up at the Steering Committee, although that would clearly
blow our
schedule.
If other implementations can finish in the same time as the
existing schedule,
that would be great! Clearly the TCKs have to be "frozen"
at the start of the release review, so that gives other
implementations
some time to ensure that they pass the TCKs. If they can show
that
they pass the frozen-but-not-yet-approved TCKs at the end of
the review
period, I think we should consider them compatible
implementations as part
of the release. And of course if they pass the finalized TCKs
shortly
after the approval, we should add them to the list of
compatible implementations.
If other implementations are intending to be compatible at the
current
release date, that would be helpful to know ASAP. That might
require
additional resources to support usage of the Jakarta EE TCKs
and we should
plan accordingly.
We do have to remember that the TCK is testing functionality,
not whether
the proper license and/or javadocs have been updated.
We
expect the Specification Committee to be checking that the
submitted javadocs
include the correct license, spec name, etc. as part of the
specification
approval process.
So, technically, once these CIs pass the TCK (with or without
the updated
APIs), then they should be ready for the Release Review and the
eventual
Announce.
Yes,
other compatible implementations don't need to depend on these
API jar
files.
But, I do get your point about having to test these updated APIs
before
releasing them. I thought that was part of the TCK testing that
Dmitry
and Ed were going over this morning?
Yes,
it is. Every changed artifact needs to be tested before
release.
Every API jar file will be updated and released. Some
implementations
might also be updated and released.
During our PMC call just prior to the Steering Committee call,
Dmitry explained
that the individual API TCKs would need to be executed to ensure
they are
compliant. These individual tests would not be comprehensive
though
-- we should still do some testing that they work throughout the
whole
TCK bucket.
All
of the existing EE4J implementations need to be tested with
the new Jakarta
EE TCKs since those are the implementations that we will
include in the
approval process as compatible implementations.
Maybe we need all of the Compliant Implementations to re-run the
TCK with
the proper APIs during the Review period? Given that the
individual
API TCKs were run prior to the review, and the CIs would already
passed
the required TCKs (minus the API updates), maybe re-running the
CIs with
the updated APIs would be sufficient at this time? Maybe this
is
the gist of your compromise?
Yes,
we just need to do something to ensure that the updated API
jar files still
match the spec and no unexpected changes or errors have crept
in. Again,
we're not expecting to include these updated API jar files in
most of the
Compatible Implementations at EE4J for Jakarta EE 8, and no
other Compatible
Implementation is required to use them (although they can if
they want
to).
> We need to update
all of
the javadocs for all of the APIs to use the new
specification names and to use the Eclipse Foundation
Specification License.
Since we're actively discussing how far-reaching the acronym
usage limitation
is, I'll leave the extent of the javadoc updates out of this
discussion
and just focus on the requirements of the Release Review and
the Compatible
Implementations.
Even
ignoring the acronyms, most of the javadocs will include the
name of their
own spec (which is changing), and many will include references
to JSRs
(which need to be removed or replaced). The javadoc updates
may not
be extensive, but they include at least the new license, and
in most cases
will include some other required changes.
_______________________________________________
jakarta.ee-spec.committee mailing list
jakarta.ee-spec.committee@xxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe
from this list, visit
https://www.eclipse.org/mailman/listinfo/jakarta.ee-spec.committee
_______________________________________________
jakarta.ee-spec.committee mailing list
jakarta.ee-spec.committee@xxxxxxxxxxx
To change your delivery options, retrieve your password, or
unsubscribe
from this list, visit
https://www.eclipse.org/mailman/listinfo/jakarta.ee-spec.committee
|
