Eric and I discussed this in person, so to relay the discussion:

- The CSS support needs to be initialized very early, prior to any widget 
creation.

- The services so far are for use by the parts.  We have a styling service 
at present that sets the CSS id/cass. This may never get called for a 
given part and thus is not a good candidate for CSS engine init.  Eric 
observed correctly that the real relationship is between CSS and the SWT 
renderers.  Thus I think the services aren't the right mechanism for this 
initialization.

- My argument, which I think Eric agreed with, is that if we're talking 
about workbench model + SWT then we might as well also be talking about 
CSS for any practical use of the workbench + SWT.

- It's possible that the same model though could be used with a different 
renderer and therefore maybe you don't want to use CSS.  Thus again it 
seems reasonable to pair renderer + styling engine (or different renderer 
+ different styling engine).

- Currently the optional dependencies and reflective code are in 
org.eclipse.e4.ui.workbench.swt.  Thus I think it's ok to make the use of 
CSS explicit there.

- Eric observed that most if not all the contents of that project could be 
moved to workbench.renderers.swt.  We also noted that many of the classes 
in the project seem like left over experiments to be deleted.

Eric chime in if I missed anything.

Regards,
Kevin





Eric Moffatt/Ottawa/i...@ibmca 
Sent by: [email protected]
04/07/2009 09:22 AM
Please respond to
E4 Project developer mailing list <[email protected]>


To
E4 Project developer mailing list <[email protected]>
cc

Subject
Re: [e4-dev] Re: CSS relationship to e4 workbench







The way we handle this for the legacy code (also an optional dependency) 
is that the legacy bundle registers a service using the 
org.eclipse.e4.services extension point. Then code that needs to gets the 
service from the global context. A returned 'null' means that the bundle 
isn't there and the code can take appropriate actions (if any). This 
mechanism uses the service layer/context rather than reflection to 
determine the availability of the necessary classes. It also avoids the 
eager loading cases since services themselves are lazy loaded. 

Not sure if this will help in the CSS case but it might provide some 
avenues to look at... 

Eric 



Ed Merks <[email protected]> 
Sent by: [email protected] 
04/07/2009 06:11 AM 

Please respond to
E4 Project developer mailing list <[email protected]>


To
E4 Project developer mailing list <[email protected]> 
cc

Subject
Re: [e4-dev] Re: CSS relationship to e4 workbench








Gunnar,

Well, you still have to be careful about Java wanting to load classes that 
might not be available.  Certainly EMF makes quite heavy use of the type 
of pattern you describe, i.e.,  non-reflective code guarded by tests for 
availability of the classes used in that code.  That's how we support 
stand alone execution with no Eclipse dependencies.  Often it's necessary 
to put the logic in a separate class to prevent Java from eagerly loading 
it.  It's obviously very important to test for proper behavior when the 
classes aren't available.

Cheers,
Ed


Gunnar Wagenknecht wrote: 
Kevin McGuire schrieb:
  
1) The use of optional dependencies is cool, but makes writing more code
here tricky since it must be reflective which is brutal to write (I
believe Boris has a "reflectify" refactor but that only helps the first
time and change/debug is still a challenge).  It also is a potential
point of introducing errors if we forget to add the CSS plugins to the
product since the tooling won't warn you of their absence (being 
optional).
    

The trick is to only use reflection to verify that an optional
dependency is satisfied. Thereafter it's totally fine to call out to a
different class which uses the API provided by the optional dependencies
directly.

An example of this can be found in the Equinox Preferences bundle.
org.eclipse.core.internal.preferences.Activator#start
org.eclipse.core.internal.preferences.Activator#addingService

OSGi services actually do support this very well because you only need
to reference the service class by name.

-Gunnar

 _______________________________________________
e4-dev mailing list
[email protected]
https://dev.eclipse.org/mailman/listinfo/e4-dev
_______________________________________________
e4-dev mailing list
[email protected]
https://dev.eclipse.org/mailman/listinfo/e4-dev

_______________________________________________
e4-dev mailing list
[email protected]
https://dev.eclipse.org/mailman/listinfo/e4-dev

Reply via email to