Comments inline.
Simon
Simon Laws wrote:
On Jan 2, 2008 1:23 PM, ant elder <[EMAIL PROTECTED]> wrote:
On Jan 2, 2008 12:00 PM, Simon Laws <[EMAIL PROTECTED]> wrote:
On Jan 2, 2008 10:53 AM, Mike Edwards <[EMAIL PROTECTED]>
wrote:
Folks,
Some comments....
Yours, Mike.
ant elder wrote:
On Jan 2, 2008 8:58 AM, Simon Laws <[EMAIL PROTECTED]>
wrote:
For http://issues.apache.org/jira/browse/TUSCANY-1608 I've put in a
change,
based on the ant generator plugin, to bring some automation to the
process
of building the ant files for the samples and demos. For any sample
or
demo
that requires explicit dependencies, e.g. the webapp samples, I've
replaced
the static ant file with and automatically generated one. In the
case
that
some hand crafted ant script is needed, for example, to generate
SDOs,
then
I have the ant generator just build build-dependency.xml which has
the
dependencies listed and which can then be included in the manually
generated
build.xml script.
I haven't applied this change to all of the samples but it could be
done.
If
we did have all of the dependencies explicitly described for all of
the
samples can we get rid of the "all" and "manifest" jars?
Simon
I think its better if applications don't have to know or care about
Tuscany
internals, that includes knowing all the different Tuscany module
names
and
all the dependencies they use.
+1 - applications should ideally have ZERO dependence on Tuscany
internals. They should be deployed to an "SCA capable runtime"
without
having to know anything about that runtime.
We haven't got this right yet so each time we
release our sample Ant builds break as the build.xml files get out
of
date -
this will be happening for any Ant builds our users have as well.
The
"all"
jar is an attempt to fix this, its a better way IMHO than having
applications specify every Tuscany module but theres a bit of work
still
to
do to make it work better for webapps. We've also talked before
about
changing all the samples to be simple sca contributions that don't
need
any
mention of the Tuscany internals, this is something I think we
really
need
to do. Both of those things seem better to me than messing about
trying
to
generate build scripts.
I agree with this sentiment. We should be building:
a) runtimes of various kinds (SCA standalone, embedded within Tomcat,
etc)
b) applications, containing only the code and other artifacts required
for the application itself
and then have some regular means of deploying the applications to
appropriate runtimes - some applications could be deployed to "almost
any" SCA runtime while others need specific runtime capabilities such
as
a Web server and Servlet support.
...ant
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
We are describing configuration scenarios in this thread that are not
currently well covered in the samples that ship with Tuscany. Take the
following steps as an example,
1 - choose a runtime
2 - configure the runtime
3 - start the runtime
4 - assemble application(s)
5 - contribute the application(s)
6 - make top level contributed composites available
90% of our samples currently assume that
1-3 means start a standalone runtime with all features enabled by using
the
"all" jar
4-6 means add a single contribution with a single composite
The ant scripts we have target step 2 and have no impact on 4,5,6. The
ant
scripts do not imply that the "applications have to know or care about
Tuscany internals". I think you are confusing the act of starting the
runtime with that of contributing the application. It just so happens
that
in our samples these steps all happen in close proximity, I.e. our
samples
generally include a runtime and the application itself. In the
particular
case of webapps the ant scripts configure the runtime (the war in this
case)
so that it contains just enough to run and no more in order to reduce
the
size of the war.
Maybe if we had a way to say "Tuscany standalone with all extensions" or
"Tuscany Tomcat with BPEL, WS and JavaScript" then ok but if the
application
build process has to mention tuscany-assembly, tuscany-core,
tuscany-definitions, tuscany-databinding and all the other dozens of
required jars then that does seems like "knowing and caring about Tuscany
internals" to me.
This is the case as at some point someone decided to package our samples
along with the runtime that will run them. So to configure the runtime you
need to configure the sample build. There is no hardwired relationship
though between the runtime build and the application build.
The case where we do need to build the two together is when we package the
runtime within the WAR that holds the application. There have been numerous
posts during previous releases about reducing the size of these wars which I
believe is why the WAR builds were different from the standalone runtime
builds. We could take the view that we always package everything and make
the build simpler.But I prefer what you are suggesting of being able to
describe required features rather than the explicit jars that implement a
feature.
Surely all this should be automatic. The runtime should load in the
pieces that are needed by the applications that are deployed to run
on it. It should not be necessary to predict ahead of time what
runtime pieces will be needed and configure a statically tailored
runtime containing exactly those pieces. The base runtime should
include whatever bootstrap code is needed to load in the extensions
that are needed by applications. I'd be very interested to work on
implementing such an approach.
Having said this I still agree that having the samples be just
applications
that are then contributed to a runnning runtime of your choice is a more
well rounded target to aim for. I note that there a series of modules
under
construction.
tuscany-runtime
tuscany-runtime-tomcat
tuscany-runtime-war
which could be extended to include tuscany-runtime-standalone,
tuscany-runtime-jms or whatever else is required.
We still need a mechanism for configuring these runtimes. In particular
we
have binding, databinding and implementation extensions that can be
included/excluded. Also some of these extensions may provided
alternative
implementations that must be selected explicitly. It seems that steps 1,
2
and 3 resolve to
1 - choose hosting approach
2 - choose extensions
3 - build useable runtime
It may be that the "all" jar can satisfy many of the cases now as we
don't
currently have any cases of alternative extension implementations.
However
I
would be nervous in the long run because of the lack of step 2. If
nothing
else we should remove the hosting options from the all jar.
I think the "all" jar worked ok as a point in time thing to make the
releases more usable after all the problems with M2 but now Tuscany is
getting more complex it might be coming to the end of its useful life.
Agree about removing the hosting options from it, they seem like they
should
be separate things, maybe what we need is to separate out all the runtime
environment stuff from the all the other code so there's a jar for each
environment and a jar with all the common stuff and you'd use them
together,
eg:
tuscany-runtime-standalone.jar and tuscany-sca.jar
or
tuscany-runtime-tomcat.jar and tuscany-sca.jar
or
tuscany-runtime-geronimo.jar and tuscany-sca.jar
And the various runtime jars would include the impls for the specific
environment so tuscany-runtime-geronimo.jar would include the Geronimo
specific impls of things like the Tuscany ServletHost and JMS host etc.
Yes. this is the kind of thing I was thinking. Maybe we ship
tuscany-sca.jarwith everything in it and provide a build script that
creates it based on a
selection of feature names if you want/need to cut is down for whatever
reason.
You mean everything except the environmental modules, I presume?
Can we enumerate the concrete scenarios we want to satisfy to help us.
I've
seen the following so far. SCA applications running with
A - a standalone runtime
B - a war runtime
C - Tomcat with no Tuscany code in the war but with Tuscany embedded in
Tomcat
D - Geronimo with no Tuscany code in the war but with Tuscany embedded
in
Geronimo
E - a standalone runtime and a JMS provider
Also the distributed store tutorial brings up the case where the SCA
application uses SCA components to describe a service running outside
the
SCA runtime, i.e. the host has no Tuscany runtime but you can wire to
it's
services.
What else do we need to cover?
Simon
Those A to D sound like a good start to me, not sure about E, why is JMS
different than say RMI - does A include the RMI host?
Maybe this is just my misunderstanding of what the JMS host does. I thought
it was ultimately going to be an optional extra in the case that you needed
to support the JMS binding with an auto started JMS broker (as the samples
do currently) as opposed to some existing broker. From you question it
sounds like you are expecting it to be around all the time in which as it
would be the same as A
I believe the auto-starting JMS broker is the same as case A. If I run
a standalone application that exposes an SCA service with a Web Service
binding, the Tuscany standalone runtime will automatically expose a
SOAP/HTTP endpoint using Tomcat or Jetty. This is case A, and I'd expect
RMI or JMS services to work the same way.
Simon
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
