Hi Thomas,
On 2/19/2014 1:53 PM, Thomas Watson wrote:
We probably should get a sense
of how final the APIs are looking for RSA 1.1 before making
them available as an official API for ECF. We always run the
risk of putting something out there as a release before it has
become officially final by OSGi. I have been bitten by this
in Equinox when the OSGi core spec slips past a feasible date
to be included in the release train. One example is OSGi
composite bundles. We had several equinox releases where the
composite API (and an implementation) was included in Equinox
releases but it never become a real OSGi specification. For
luna I have finally been able to remove this. But it required
a 2 year notice to the community first because we considered
it real Eclipse API. That was a completely separate 1.0
package so it was easier deal with it being removed. But if
the package is updated to 1.x then it is much more
catastrophic if you put something out there as 1.1 that is not
really the final 1.1. Then when 1.1 is really final you are
left with having to update your version of the API without
incrementing your package version.
Yeah, I completely agree. Remember RFC-119? (precursor to remote
services).
But...happily...I don't believe this is likely to be a huge problem
for the RFC 203[1]/RSA 1.1. Main reason is that the new API is
*very* limited. Basically the new API boils down to adding support
for remote service update (i.e. the local equivalent to service
modified)...and creating a new/additional event listener and event
type to support the update notification. AND adding new
capabilities to represent the RSA subsystems (topology manger,
discovery, distribution). In prose above this sounds like a lot,
but it's not really...and pretty uncontroversial...is my sense from
archives and discussions.
Doing the work in a branch is
good and then not merging into master until you are very
comfortable that the API will be final final by the time you
release. Even doing an RI in a branch is acceptable.
Yeah, that's what we are doing...with the
hope/expectation would be before 'too late' for addition to Luna.
As far as I know OSGi does not
require RIs to be included in any official release from an
open source project. I don't think it even has to be provided
from a download site. All that is required is that the source
code is freely available in a repository and it can be built
in order to get the RI binary.
Yeah, that's understood. A pain for the community if it can't get
into Luna though.
But I do realize it is in your
best interest to have the ECF luna release support RSA 1.1.
But if there is any question on the finalization of the RSA
1.1 API in time for the Luna release then you should consider
not releasing it until Luna SR1.
Yeah...if things go wrong WRT the final approval of RFC 203 for R6
then this is what we will do.
(and sorry for the long winded
answer, did not intend to take that long! ;-)
No problem, it's appreciated. My only request (to you/Tom, rt pmc,
and EF folks that touch OSGi) use any means and/or influence
available to keep the OSGi RFC approval (or disapproval for R6) on
schedule.
Scott
|