| Hi Ian,
there is a discussion with Wayne on the technology PMC mailing list on that. The problem is: To support JEE, we need example applications that show how people use these libraries. For JEE libs these example applications are hard to get since most applications are developed in companies. Code Recommenders has a 'crowd-sourcing' plug-in which enables users to share their (anonymized) code with the server so that we are able to learn valuable patterns from them. With that data (== usage data == example applications) we can generate models for almost all libraries in use by Eclipse developers.
This 'usage data', however, has some implications that need to be considered. The anonymization has to be done right, the data needs to be stored somewhere, legal aspects need to be addressed etc. Also, Wayne pointed out that the Foundation has no bandwidth for this kind of service. Maybe it makes sense you talk to Wayne and join the discussion here http://dev.eclipse.org/mhonarc/lists/technology-pmc/msg03863.html ?
Marcel On 11.01.2012, at 14:19, Ian Skerrett wrote:
Can you tell me more about what you mean by ' a usage data collector
for community-driven model generation would be supported. ' I
unfortunately do not know how code recommenders work. :-(
Ian
On 1/11/2012 2:59 AM, Markus Knauer wrote:
Sounds great. Let us know when your bundles are
available from the Simultaneous Release repository (i.e. the
staging repository) and what we need to include.
Thanks,
Markus
On Tue, Jan 10, 2012 at 20:02, Marcel
Bruch <bruch@xxxxxxxxxxxxxxxxxx>
wrote:
Hi Markus, Steffen, Ian,
We are currently rewriting these parts and
will probably be ready mid/end of January.
most of the critical rewrites are done or will be
done until the end of the week. We've hudson builds
for 3.x and 4.2 and will likely join the aggregation
build by the end of the week. I'll get in contact with
Java Package and RCP/RAP via Bugzilla to discuss next
steps and requirements.
Ian, for JEE I need to discuss with the Foundation
whether a usage data collector for community-driven
model generation would be supported. It's challenging
to find a sufficient amount of (open source) data to
create a reasonable coverage for existing JEE
libraries.
Best,
Marcel
Hi Markus, Steffen
and Ian,
Java and RCP/RAP would be my favorite
packages. RCP is a perfect fit since we already
have lots of the data for it. Java is excellent
since we can generate models for Java SDK quite
easily and many tools work out-of-the-box.
However, I'm a bit cautious at the moment. In
the past months we received a lot feedback and
learned that some of our concepts fall too short
and need improvement. We are currently rewriting
these parts and will probably be ready mid/end
of January.
Would it be ok to test the package in January
after the rewrite of the code analyzer and
extdoc parts?
I was a little bit
selfish when I asked you to propose Code
Recommenders for package inclusion. How
about opening another bug against the
RCP/RAP package? If it is not too large I
would be willing to give it a try in the
RCP/RAP package as well, but I will need
some testing support.
Thanks,
Markus
On Mon, Dec 5,
2011 at 14:11, Ian Skerrett <ian.skerrett@xxxxxxxxxxx>
wrote:
I also think the Java EE package
would be a logical place for Code
Recommenders. You might open a bug
for the package too.
Ian
On 12/5/2011 8:09 AM, Steffen
Pingel wrote:
Hi
Marcel,
great to hear that Code
Recommenders will be part of
the release train!
Steffen
On
Mon, Dec 5, 2011 at
12:10 PM, Marcel Bruch <bruch@xxxxxxxxxxxxxxxxxx>
wrote:
Hi
EPP-Dev,
after short
conversation with
Technology PMC we
decided to join
the annual release
train for
Juno. Markus
Knauer told me
that with joining
the annual release
train we may also
be join some EPP
packages - if
there is a package
owner willing to
add Code
Recommenders to
their package.
Can any package
owner imagine to
integrate Code
Recommenders into
his package? If
so, what
requirements
should we met? If
not, it would be
great to learn
why.
What does Code
Recommenders offer
for annual
release?
We have several
components that
can be packaged
separately (more
details on
request):
* Intelligent
Method Call
Completion
& Method
Overrides
Completion
* Call Chain
Completion
* Snipmatch
* Extdoc
Platform
* Subwords
Completion
* Usage Sharing
Client to build
continuously
growing knowledge
bases for
intelligent
completion +
extdoc
* (maybe) new
code search client
+ engine for
Eclipse
A (not
self-learning)
version of Call
Chain and Subwords
Completion may
already be handed
over to JDT 3.8
until Juno. All
tools work for
Java only.
A video showing
the above tool
suite in action is
available here:
PMC
conversation about
joining the
release train:
Code
Recommenders
source code
repository:
Thanks,
Marcel
_______________________________________________
epp-dev mailing list
epp-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/epp-dev
epp-dev mailing list
epp-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/epp-dev
|
