On 09/02/07, Simon Laws <[EMAIL PROTECTED]> wrote:
On 2/9/07, Andrew Borley <[EMAIL PROTECTED]> wrote:
>
> On 1/30/07, Simon Laws <[EMAIL PROTECTED]> wrote:
> > On 1/30/07, Jean-Sebastien Delfino <[EMAIL PROTECTED]> wrote:
> > >
> > > [snip]
> > > Simon Nash wrote:
> > > > A repackaging into a kernel and language extensions as suggested
by
> > > > Pete, Ignacio, and Jeremy seems like a good direction. We'll
still
> > > > have to find a name for the (native, C++, scripting, ???) kernel,
> > > > though. And we'll have to decide what kind of distribution
bundling
> > > > is helpful for our users. Jeremy's suggestion of combining
related
> > > > pieces from the "Java" and "C++" worlds seems like a good
approach.
> > > >
> > > > I am not sure why we would put the cpp language extension into the
> > > > kernel, as suggested by Pete. Is this because it is needed by all
> > > > other extensions that support scripting languages?
> > > >
> > > > Simon
> > > >
> > >
> > > After having thought about various other code names, I'm starting to
> > > like "native".
> > >
> > > I am +1 with separating a kernel from the extensions. The CPP source
> > > tree and build structure already have this separation anyway. The
> kernel
> > > does not depend on the C++ component type support extension so I
don't
> > > think that it should include that extension.
> > >
> > > I also like Jeremy's suggestion of combining the current java/ruby
and
> > > cpp/ruby into a Ruby top level module containing the code to support
> > > both Jruby/Java and the Ruby native interpreter and a shared set of
> > > samples. I think that the same idea will work well for other
Scripting
> > > component types (Python, Javascript etc.) and will help us ensure
> > > consistency across runtimes.
> > >
> > > I propose to go even further and generalize this idea to bindings
(WS
> > > binding, REST binding, an AMQP/Qpid binding may be a good candidate
> too)
> > > and other component types as well.
> > >
> > > I would like to stage some of these changes as they will generate
> > > significant work to adjust all the build files etc. I'd suggest to
> start
> > > changing the terminology (core to kernel, C++ to native) and see how
> we
> > > can share some samples, but attack the refactoring of the source
trees
> > > and the build system later, after the C++/native M3 release.
> > >
> > > Thoughts?
> > >
> > > --
> > > Jean-Sebastien
> > >
> > >
> > >
---------------------------------------------------------------------
> > > To unsubscribe, e-mail: [EMAIL PROTECTED]
> > > For additional commands, e-mail: [EMAIL PROTECTED]
> > >
> > > All of this refactoring sounds interesting and many systems
distribute
> a
> > core/kernel and then optionally allow you to download extensions, PHP
> for
> > example. I should say that in the PHP case poplar extensions find
their
> way
> > into the kernel/core over time once they have proved their worth
because
> > people don't particularly like downloading extensions.
> >
> > As Jean-Sebastien pointed out this separation already exists in the
make
> > system so the user can choose precisely which bits to build once the
> source
> > distro has been downloaded. In the case of the binary distro I think
the
> > case will be that the system only picks up those extensions that you
> > reference in SCDL so that fact that you have installed all of them is
> not an
> > issue given the number of extensions we have now. I would therefore
> suggest
> > that the extra complexity of delivering physically separate tars for
> > extensions and kernel is not required for M3 particularly because
> Jeremy's
> > ideas are interesting and may lead to rehashing the details of how
this
> is
> > done for M4 anyhow.
> >
> > Simon
> >
>
> I think I'm leaning towards this strategy too now - keep a single
> binary downloadable package - as Simon says "the system only picks up
> those extensions that you reference in SCDL so that fact that you have
> installed all of them is not an issue".
>
> I do think we still need a name change, and "native" seems to be the
> favourite. I'm happy with this - what do others think?
>
> The only issue is with our samples - most of them require more than
> one extension (e.g. the PythonCalculator shows the use of both the
> Python and WS binding extensions) - maybe we should split some of them
> up a bit more. With clear docs that explain what each sample requires
> this should be OK.
>
> So my proposal now is that the C++ M3 release provides source and
> binary "Tuscany SCA Native" packages for Windows, Linux and Mac OSX,
> with only the extensions (and their respective samples) that can be
> built on that OS available within the package (e.g. no Axis WS binding
> on Mac).
>
> This has gone round in circles a bit - apologies for that! What do
people
> think?
>
> Cheers
> Andy
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>
> Andy
Thanks for pulling us back to this.
+1 to "Tuscany SCA Native" & to source and binary packages for Windows,
Linux and Mac OSX
Re. samples. Good point! We have mostly implementation specific samples
with
a few higher level samples sprinkled in. (PHPCalculator is perhaps a
little
unusual in that it includes C++ components because, at this point in time,
it has to).
AlertAggregator
CppBigBank
CppCalculator
HttpdBigBank
PHPCalculator
PythonCalculator
PythonWeatherForecast
RestCalculator
RestCustomer
RestYahoo
RubyBigBank
RubyCalculator
There is a general pattern that has a sample direcotry organized as:
xyz
sample.xyz
sample.xyz.composite
sample.xyz.client
sample.xyz.wsclient
sample.xyz.app.composite
The source structure for the sample is reflected when it's deployed. This
seems to make it difficult to mix and match the components of samples, or
indeed provide differing bindings in different situations or on different
platforms.
The first options that come to mind (i'm sure there are lots more):
1/ Allow components to be resused across samples. I don't know how to do
this in as much as the runtime seems to recurse down the application
directoy looking for composites. Could be a deployment step to copy in
composites as required.
2/ Can we use the deployment step to choose the application level
composite
that's appropriate. I.e rather than having LocalCppCalculator and
WSCppCalculator have
CppCalculator
...
samples.calcualtor.local.app.composite
samples.calculator.ws.app.composite
And allow both or either verisions to be deployed if required.
3/ Change the runtime to allow us to point to a specific application
composite files. It's possible that it already does this but I just don't
know how.
...
Simon
I think a reorganisation of the samples is required but I'm not sure whether
this would be too much churn for the M3 release. I think there should be at
lease one sample for each language extension that simply demonstrates the
language binding. The xxxCalculator series seems to be the best (ans
easiest) choice.
For Cpp I have deleted the Axis2C client sample for cpp client and will also
do this for the CppBigBank sample. We are not trying to demonstrate how to
code Axis samples! SO.. now I can build the core/kernel tuscany_sca and the
cpp extension tuscany_sca_cpp and be able to run the CppCalculator sample.
That's all I need.
If we want a sample to demonstate using ws bindings then it should be a
different sample, not just a separate client to the same samples: I don't
want the <binding.ws > element in my basic sample scdl because I would not
be able to run the basic CppCalculator without a ws binding extension. It
would be good to show a ws binding sample re-using the components from the
base sample though.
I'll start a thread for the M3 release content. It would be nice to have a
matrix showing which samples require which extensions and on which platforms
each extension is available (e.g. ws will not be available on Mac and PHP
may not be available on Windows).
Cheers,
--
Pete
