Frank,
The approach I am thinking is to make DefaultHelperContext a public interface
that its implementation can be provided by others. This is similar to the way
o.a.t.s.h.HelperProviderImpl was discovered in the
META-INF/services/commonj.sdo.impl.HelperProvider file and bootstrapped. Of
course, the default impl will be still using TCCL.
Fuhwei
Frank Budinsky <[EMAIL PROTECTED]> wrote: Hi Guys,
I finally managed to catch up on this. I think you're heading in the wrong
direction by using HelperProvider.getDefaultContext() because:
1) it's a singleton, and we're trying to get away from singletons:
static HelperContext defaultContext = new DefaultHelperContext();
The fact that the default context in Tuscany has some special TCCL-based
managment (under the covers - thanks to EMF), is a Tuscany-specific
behavior - but the default context is itself a single object (singleton).
2) HelperProvider is an impl class (in package commonj.sdo.impl) ... It's
bad to make that the application API
The reason we added HelperContext to the spec to get away from the
singleton problems. The default context was needed as an interim step,
until the spec comes up with other ways to access scoped HelperContexts.
Many of the problems we had with EMF were exactly because we were trying
to replace the global EMF INSTANCEs - e.g., EcoreFactory.eINSTANCE - but
then we ran into problems because there were multiple apps trying to
control the value. We don't want a repeat of that with the default
context.
That said, I think that a better way to look at this is that we want a
pluggable mechanism for managing HelperContexts in Tuscany. Since the
ability to manage scope-specific context's is a general issue, not just
for SCA, it does seem to make sense to put this into Tuscany/SDO. Maybe we
can feed back our (hopefully successful) result to the spec, so in the
future there will be proper SDO APIs for this.
For now, however, we're talking about some new SDOUtil methods for this.
For example, what if we provide a HelperContextManager, where you can
register keyed HelperContexts. The key could be a ClassLoader, Thread, or
anything else. Given such a manager, we could provide some SDOUtil methods
to register and access the HelperContext. Maybe something like this:
HelperContext SDOUtil.getHelperContext(Object key);
and maybe even some convenience methods like this:
HelperContext SDOUtil.getHelperContext(Thread thread);
HelperContext SDOUtil.getLoaderHelperContext(Class Loader classLoader);
These helper methods would be used by clients to access the HelperContext.
The implementation would just use the HelperContextManager to get the
keyed HelperContext. Maybe we could also provide ways to set up chained
(according to parent CL's, for example) HelperContext's, etc.
Thoughts?
Frank
Fuhwei Lwo wrote on 04/20/2007 01:36:37 PM:
> Raymond,
>
> I agree with your suggestion below. In addition, I think SCA still
> needs to provide an option (injection or API) for the applications
> to explicitly retrieve the data model scope from the Contribution.
> Other databinding technology APIs beside SDO may not have default
> context helper concept.
>
> 1) SDO defines the pluggability to supply the default HelperContext.
> 2) SCA plugs its own scoping scheme to the SDO default HelperContext.
> The
> HelperContext will be populated based on the Contribution.
> 3) Application code will use HelperProvider.getDefaultContext() to
> retrieve
> the default HelperContext.
>
>
> Raymond Feng wrote: Hi,
>
> Please see my comments inline.
>
> Thanks,
> Raymond
>
> ----- Original Message -----
> From: "Jean-Sebastien Delfino"
> To:
> Sent: Friday, April 20, 2007 9:23 AM
> Subject: Re: Scoping SDO metadata, was: How to access a composite's data
> model scope in an application?
>
>
> > Fuhwei Lwo wrote:
> >> Hi Sebastien,
> >>
> >> Here is my understanding of requirements about getting rid of
import.sdo
> >> and switching to contribution -
> >>
> >> 1) A contribution will be created by contribution processor for each
> >> application. - Contribution processor has been done for Jar and file
> >> system.
> >>
> >>
> >
> > Yes
> >
> >> 2) The contribution processor will create a SDO scope (HelperContext
> >> instance) to associate with the contribution. Currently calling
> >> SDOUtil.createHelperContext() is enough.
> >>
> >
> > That's what I was poking at in my previous email. Creating our own
> > context, different from the default SDO context forces SCA to
introduce a
> > new API to get to that context, and forces all SDO users to use that
new
> > API. So I'm wondering if it wouldn't be better to play more nicely
with
> > SDO, and have the SCA runtime just populate the default SDO context in
use
> > in a particular application in the server environment.
> >
>
> I have a slightly different view here. IMHO, the SDO should provide the
> scoping mechanism and the pluggability of scoping schemes. I assume the
> HelperContext is provided by SDO for scoping metadata. What's missing
from
> SDO is the pluggability of the scoping schemes. Currently, the default
> HelperContext is based on TCCL and it's not replaceable. I agree SDO
cannot
> define scoping schemes for all environment so the pluggability is
desirable.
>
> >> 3) Tuscany SCA needs to provide a way for the application to get hold
of
> >> the HelperContext in association with the contribution in step 2
above.
> >> Currently the application is forced to use SDO API -
> >> HelperProvider.getDefaultContext() which is using TCCL.
> >>
> >
> > I'm not getting this one :) Is it bad for an SDO user to be "forced
to"
> > use an SDO API to get an SDO context? It seems better to me than
forcing
> > an SDO user to use an SCA API, simply because his code may be used at
some
> > point in an SCA environment... and then his code wouldn't work in a
JSP, a
> > servlet, or any other non-SCA environment...
> >
> > If the fact that HelperProvider.getDefaultContext() is using the TCCL
to
> > find the correct SDO context is a problem, then we just need to fix
that.
> > We went through the same discussion with SCA CompositeContext about a
year
> > ago. Associating context with the TCCL is not always convenient in a
> > server environment, and it may be better to associate context with the
> > current Thread (using a threadlocal or an inheritable thread local for
> > example). This is what we did for SCA CompositeContext. Maybe SDO
could
> > provide a way to associate an SDO context with the current thread
instead
> > or in addition to associating the SDO context with the TCCL?
>
> I agree that we should try to use the SDO API to retrieve the current
> context. But I think in the SCA application, the default context should
be
> associated with the Contribution. Then it would be a win-win situation
if we
> can do the following:
>
> 1) SDO defines the pluggability to supply the default HelperContext.
> 2) SCA plugs its own scoping scheme to the SDO default HelperContext.
The
> HelperContext will be populated based on the Contribution.
> 3) Application code will use HelperProvider.getDefaultContext() to
retrieve
> the default HelperContext.
>
> >
> > This would seem a good thing to have anyway since these contexts are
not
> > thread safe as far as I know :)
> >
> > Thoughts?
> >> I am not sure my understanding above is correct so please bear with
me.
> >> Based on my understanding above, currently there is no additional
> >> requirement from SDO.
> >
> > I wouldn't reach that conclusion so fast :) I think that there is a
> > requirement to provide a way to get to an SDO context independent of
TCCL
> > if people don't like that association with TCCL.
> >
> >> In the future, if we decided to support contribution import/export
that
> >> may require SDO scoping hierarchy support. But I think we should
start
> >> using contribution and getting rid of import.sdo as the first step.
> >>
> >>
> >
> > Yes I'd like to get rid of import.sdo, as I indicated earlier in this
> > discussion thread.
> >
> > I would like to support contribution import/export at some point. I'm
not
> > sure that we'll be able to use SDO scope hierarchy support as an SCA
> > contribution import does not necessarily import the whole scope of
another
> > SCA contribution, but I guess we'll know more when we start to look at
the
> > details.
>
> I'm thinking of the following approach to discover SDO metadata from a
SCA
> contribution.
>
> When the Contribution is processed, the generated SDO factories (the
class
> name and the namespace) are recognized. Other models such as WSDL/XSD
are
> handled as well. We don't have to convert all of them into SDO model
upfront
> as the conversion can be performed on-demand upon the query of a
particular
> namespace.
>
> >
> >> What do you think? Thanks for your reply.
> >>
> >> Fuhwei Lwo
> >>
> >> Jean-Sebastien Delfino wrote: Fuhwei Lwo wrote:
> >>
> >>> Hi,
> >>>
> >>> In my composite, I defined in the default.scdl file that would
prompt
> >>> the SCA container to register my data types using SDO databinding.
The
> >>> question I have is what API I should use in my service
implementation
> >>> code to obtain the registered data types. If I have two composites
that
> >>> are using two different data type definition but with the same
namespace
> >>> URI, I definitely don't want to obtain the wrong data type
definition.
> >>> Thanks for your help.
> >>>
> >>> Below is the previous message from Raymond Feng about associating
> >>> databinding type system context/scope with a composite. I think this
is
> >>> related to my question but from Tuscany SCA development perspective.
> >>>
> >>> How to associate some context with a composite?
> >>> http://mail-archives.apache.org/mod_mbox/ws-tuscany-dev/200702.
> mbox/[EMAIL PROTECTED]
> >>>
> >>
> >> Hi,
> >>
> >> The short (and not perfect) answer to your question is. With the
current
> >> code in trunk, use:
> >> commonj.sdo.impl.HelperProvider.getDefaultContext()
> >>
> >> But I thought about this a bit and your question triggered some
comments,
> >> and more questions :)
> >>
> >> Import.sdo extension:
> >> I think we should be able to remove that Tuscany extension to SCA
> >> assembly XML, now that we have the SCA contribution service in place.
We
> >> know which WSDLs and XSDs are available in a given SCA contribution
and,
> >> with sca-contribution.xml import elements, we also know which XML
> >> namespaces are imported from other SCA contributions or other
locations
> >> outside of an SCA domain. So we probably don't need another element
> >> duplicating part of this information in .composite files.
> >>
> >> Scope of XML metadata:
> >> My understanding of the SCA assembly spec is that the scope of XML
> >> metadata is an SCA contribution (plus what it imports from outside)
and
> >> not an individual Composite.
> >>
> >> Scope of metadata contributed by Java classes:
> >> Our runtime currently supports SCA contributions packaged as JARs or
file
> >> system folders. With these packaging schemes an SCA contribution is
self
> >> contained and cannot reference application classes in other SCA
> >> contributions. At some point we'll probably want to support packaging
of
> >> SCA contributions as OSGI bundles and then leverage OSGI to allow an
OSGI
> >> bundle to see classes in another bundle, but we don't support that
OSGI
> >> packaging scheme yet. As a side comment I'd like to see if we could
> >> reactivate some work on the OSGI extensions that we have under
> >> java/sca/contrib/ and are not integrated in our build at the moment.
So,
> >> the scope of Java metadata is an SCA contribution as well, with no
> >> external import mechanism.
> >>
> >> So the bottom line is:
> >> References to types in SCA artifacts are resolved at the SCA
contribution
> >> level. There is no relationship between an SCA composite and a
metadata
> >> scope.
> >>
> >> More comments, on databinding specific handling of metadata:
> >> We need to support multiple databindings. Each databinding comes with
its
> >> own form of metadata and different APIs to get to that metadata and
> >> define metadata scopes. I guess it's important for a databinding
> >> technology to define a way to scope metadata if it wants to be
> >> successfully used in a server environment, and isolate the metadata
for
> >> the different applications running on the server.
> >>
> >> In such an environment, our SCA runtime should play nicely with the
other
> >> pieces of runtime and application code (not necessarily running as
SCA
> >> components), and use the metadata scoping mechanism defined by each
> >> databinding in such a way that non-SCA code and SCA component code
> >> running together in the server environment are able to see the same
> >> metadata for a given application.
> >>
> >> I'd like to start a discussion to cover this aspect for our various
> >> databindings and make sure that the metadata story for each
databinding
> >> holds together.
> >>
> >> To help feed this discussion with concrete data, could the SDO folks
jump
> >> in here, and describe the various ways of maintaining SDO metadata
scopes
> >> in a server environment, running with multiple classloaders and
threads?
> >>
> >> Thanks,
> >>
> >>
> >
> >
> > --
> > Jean-Sebastien
> >
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: [EMAIL PROTECTED]
> > For additional commands, e-mail: [EMAIL PROTECTED]
> >
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
