[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
RE: [dsdp-tm-dev] RE: SubSystemConfiguration vs. SubSystemFactory ??
|
Hi Dave,
I absolutely agree: There needs to be one place that holds
all these factories
together. Note that currently, these are _not_ all in one
place since the
RemoteElementAdapters are typically registered by the
activator, and not
by the configuration.
The place that's holding all things together could
be
1.) The ISubSystemFactory class.
That's how it has been in RSE 7,
the class has been renamed to ISubSystemConfiguration.
I don't like the plain renaming
because it's misleading.
2.) An ISubSystemConfiguration class.
But then, the configuration should
not take on duties of the factory (by deriving from
the factory), but it should
delegate to the various factories where needed. That's
in-line with the common best
practice that "composition" of classes is usually better
than "extending" classes in order
to add functionality.
3.) The subSystemConfiguration extension
point.
This would allow for plain
"reconfiguration" of existing services, by naming existing
factories where needed. Compared
to (2), it's basically the same pattern but moving
from a programmatic approach to a
data-driven approach. This might eventually
be helpful if we want to support
headless (UI-less) operation by instanciating only
service classes instead of the
full-blown UI-dependent classes from a headless
application.
I'm most inclined towards (3), and I see the path towards
it gradual: Leave everything
in the factory
for now (because this _is_ how things still work), and split out the
various tasks into separate factories or a configuration
class gradually.
Thanks for your thoughts and
discussion!
I consider this really exciting and
helpful.
Thanks,
--
Martin
Oberhuber
Target Management Project Lead, DSDP PMC Member
http://www.eclipse.org/dsdp/tm
Hi Martin,
So do you think we'd need an IServiceFactory for the
configuration? If we start down that path, then we also need
IConnectorServiceFactory, and then depending on the underlying model, we'd
need something to create service model to subsystem model adapters, such as
IHostFileToRemoteFileAdapter, which
converts IHostFile to IRemoteFile. The other thing is that some
subsystems have additional services, such as the ISearchService for files -
would that just be created from the IServiceFactory? For each of these
factories, we'd still need one object to hold them altogether so that there's
a clean switch when you change from one configuration to another for a given
subsystem. The concept of service didn't exist when the documentation
was written, so I'm not sure it buys us that much there if we role up the
configuration into the factory.
____________________________________
David McKnight
Phone: 905-413-3902 , T/L: 969-3902
Internet:
dmcknigh@xxxxxxxxxx
Mail:
D1/140/8200/TOR
____________________________________
| "Oberhuber, Martin"
<Martin.Oberhuber@xxxxxxxxxxxxx>
11/08/2006 08:52 AM
|
|
To
| David
McKnight/Toronto/IBM@IBMCA
|
|
cc
| "Target Management developer
discussions" <dsdp-tm-dev@xxxxxxxxxxx>
|
|
Subject
| RE: [dsdp-tm-dev] RE:
SubSystemConfiguration vs. SubSystemFactory
?? |
|
Hi Dave,
I thought about your suggestion again.
We'll
probably need a bit more time to sort out the actual details of
separating
ISubSystemConfiguration from ISubSystemFactory. What's important for
me,
though, is that whenever a
class is responsible for creating something, I'd like
to name it "...Factory".
Bringing back the
name ISubSystemFactory instead of ISubSystemConfiguration,
for what essentially _is_ a factory, has the very
big advantage that all documentation
referring to ...factories would be correct again. And that's a
lot!!
For me it looks like even if a user re-uses an existing
FileServiceSubSystemFactory,
he'd supply his own IFileService. In other words, the configuration
would need to name
a factory for
creating IFileService objects, wouldn't it?
The extension
point, finally, names a "type" or "configuration" of subsystem.
Elements
of the extension point
(which is a configuration) can be the ISubSystemFactory class,
the IConnectorService class, and the
IServiceFactory class. Such an extension point
would (I think) make the duplicate code for the current
factories eventually unnecessary,
and all the "plumbing" of the configuration would occur via the
extension point.
The extension point would be the "configuration" but it
would name the factory
classes which are responsible for creating objects of proper
type.
This would also be a little bit in line with what the
Platform does for
extension
points
org.eclipse.update.core.featureTypes
--> element <feature-factory>
org.eclipse.update.core.siteTypes --> element
<site-factory>
I suggest we go ahead with renaming classes
accordingly for now. I'll send out
a separate E-mail with requested refactorings. We can think about the
split-up
later on if we want --
it would affect the code much less than doing all at once,
since it would just be one additional item in the
extension point.
How does that sound?
Cheers,
--
Martin Oberhuber
Target Management
Project Lead, DSDP PMC Member
http://www.eclipse.org/dsdp/tm
From: dsdp-tm-dev-bounces@xxxxxxxxxxx
[mailto:dsdp-tm-dev-bounces@xxxxxxxxxxx] On Behalf Of Martin
Oberhuber
Sent: Thursday, August 10, 2006 10:19 PM
To:
David McKnight
Cc: Target Management developer
discussions
Subject: Re: [dsdp-tm-dev] RE: SubSystemConfiguration
vs. SubSystemFactory ??
Hi
Dave,
ahh, now I see! Your suggestion sounds excellent.
I guess
there's still a few things to sort out, like where does the ConnectorService
come from (would there
be
ISubSystemConfiguration.getConnectorService()?
Then, what about
methods like supportsFilters() which are more a static configuration property
than a dynamic one and thus be more associated with the factory, than the
actual config -- after all they define capabilities of the subsystem
implementation, and not its actual configuration.
Finally, the
extension point... should the extension point name both the config and the
factory classes?
Or should the config have a method like
getSubSystemFactory()?
For me it sounds like the config is "above" the
factory, it's like the master putting all items
together.
Cheers,
Martin
David McKnight schrieb:
I'm seeing the value of the
configuration not so much for things like "isCaseSensitive" but for providing
the actual service implementations. We define the FileServiceSubSystem
independently of any service implementation. Currently the means of
providing each service implementation is via each the subsystem configuration
however each is also the thign that creates the subsystem. Each
subsystem configuration does some redundant thing - they each create
FileServiceSubSystem. RSE does allow you to switch configurations and
thus thus services such that the subsystem configuration that was intially
used to create the subsystem would no longer be used after a subsystem
configuration gets switched, which is kind of weird. That problem would
be solved with an independent factory.
If no subsystem configurations are contributed then
there would never been a subsystem to create, so I don't see the value of
having a default configuration. I guess I'm sort of thinking along these
lines:
class FileServiceSubSystemFactory implements ISubSystemFactory
{
public ISubSystem createSubSystemInternal(ISubSystemConfiguration
initialConfiguration) {
return new FileServiceSubSystem(
initialConfiguration, ... );
}
}
There would never be an SshFileServieSubSystem, nor a
DStoreFileServiceSubSystem - there's only FileServiceSubSystem with a
configuration that provides the service implementation.
class
SshSubSystemConfiguration implements ISubSystemConfiguration {
public boolean isCaseSensitive() { return true; }
public
IFileService getFileService(IHost host);
....
}
Does that make any
sense?
____________________________________
David McKnight
Phone: 905-413-3902 , T/L: 969-3902
Internet:
dmcknigh@xxxxxxxxxx
Mail:
D1/140/8200/TOR
____________________________________
Hi
Dave,
I'm afraid I cannot follow you thoroughly.
I didn't think about
contributing the configuration and the factory separately, but
only provide an
extension point for the factory. The factory would be responsible
for creating the
subsystem, and its initial configuration. I wouldn't see what the
advantage of separate contributions for configuration and factory would
be.
We probably shouldn't deviate from what we currently have too much
right now.
Currently, we have a static configuration that is tied 1:1 to the
factory. With my
proposed change, the factory could provide configurations that are
not so much
tied to it any more, and thus more flexible.
I didn't think about
persisting modified configurations though, so allowing
configurations to
change at runtime is probably something to consider for
2.0 (and keeping them static for
now).
Perhaps an example could help:
class
SshSubSystemFactory implements ISubSystemFactory {
public ISubSystem
createSubSystemInternal() {
return new SshSubSystem(
getDefaultConfiguration(), ... );
}
public ISubSystemConfiguration
getDefaultConfiguration {
//the configuration can be an
anonymous inner class,
//or a real class defined
outside
return new DefaultSubSystemConfiguration
{
// define overriders here
public boolean isCaseSensitive() { return true; }
}
}
}
Or, if we want to keep code closer to what it is right
now:
class SshSubSystemFactory implements ISubSystemFactory,
ISubSystemConfiguration {
public ISubSystem createSubSystemInternal()
{
return new SshSubSystem( this, ... );
}
public boolean
isCaseSensitive() { return true; }
}
In both cases, the Subsystem can replace its current
configuration with
something different later on.
Another option, for
DStore for instance, would be to have
class DStoreWindowsSubSystemConfiguration
extends DefaultSubSystemConfiguration {
public boolean isCaseSensitive() {
return true; }
}
class DStoreUnixSubSystemConfiguration extends
DefaultSubSystemConfiguration {
public boolean isCaseSensitive() { return false;
}
}
Comments?
Cheers,
--
Martin Oberhuber
Target Management Project
Lead, DSDP PMC Member
http://www.eclipse.org/dsdp/tm
From: dsdp-tm-dev-bounces@xxxxxxxxxxx [mailto:dsdp-tm-dev-bounces@xxxxxxxxxxx] On Behalf Of David McKnight
Sent:
Thursday, August 10, 2006 5:01 PM
To: David Dykstal
Cc:
Oberhuber, Martin; Target Management developer discussions
Subject:
[dsdp-tm-dev] RE: SubSystemConfiguration vs. SubSystemFactory ??
I like the idea but I'm
thinking that it would be good to still keep the service creation with the
configuration rather than the factory. There could be a single factory
for each different type of service subsystem:
Example:
FileServiceSubSystemFactory
--> produces --> FileServiceSubSystem
ShellServiceSubSystemFactory --> produces -->
ShellServiceSubSystem
ProcessServiceSubSystemFactory --> produces
--> ProcessServiceSubSystem
...
The factory would be responsible
for the lifecycle of the subsystem but would use the configuration to define,
not only the attributes in terms of "isCaseSensitive()" and such but also the
services themselves. The factory could use the the current to setup the
service configuration for a subsystem. For each, service there could be
a different configuration:
Example:
DStoreFileServiceConfiguration
SSHFileServiceConfguration
FTPFileServiceConfiguration
A given factory may use
one of the available configurations for creating the subsystem as well as
changing it's configuration - for example, when switching between FTP and
DStore.
If we
were to take this approach, we could keep the configuration extension point
pretty much the same - since it's really there to contribute the services, but
we'd need to introduce a new extension point for the subsystem factory.
So there would be a FileServiceSubSystemFactory contribution before any
service configurations are defined.
What do you think of this?
____________________________________
David McKnight
Phone: 905-413-3902 , T/L: 969-3902
Internet:
dmcknigh@xxxxxxxxxx
Mail:
D1/140/8200/TOR
____________________________________
| David
Dykstal/Rochester/IBM@IBMUS
10/08/2006 10:13 AM
|
|
Interesting
idea.
In most
cases where we have to grab the SubSystemConfiguration from the subsystem we
would continue to do so. So its possible this won't be as bad as I
initially suspected. This is a pretty pervasive hit though and it affects the
extension points. Would you expect to define both subystem factory and
subsystem configuration extension points independently or would a subsystem
factory provide a subsystem configuration to the subsystems it
creates?
_______________________
David Dykstal
david_dykstal@xxxxxxxxxx
