Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 363856

Summary:	[DI] Erroneous field injection?
Product:	[Eclipse Project] Platform	Reporter:	Brian de Alwis <bsd>
Component:	UI	Assignee:	Oleg Besedin <ob1.eclipse>
Status:	VERIFIED FIXED	QA Contact:
Severity:	normal
Priority:	P3	CC:	pwebster, remy.suen
Version:	4.2
Target Milestone:	4.2 M4
Hardware:	PC
OS:	Mac OS X - Carbon (unsup.)
Whiteboard:

Description Brian de Alwis

2011-11-15 13:16:51 EST

BuildId: I20111028-1100
OS: MacOS X (only because it demonstrates the problem)

The CocoaUIProcessor requests the MApplication through injection.  But an interesting thing occurs: when the MApplication instance is set (in E4Application#createE4Workbench()), the injector actually tries to inject with a created instance of MApplication, which fails.  This can be seen if you place a breakpoint on InjectionException:

Daemon Thread [Thread-1] (Suspended (exception InjectionException))	
	InjectorImpl.internalMake(Class<?>, PrimaryObjectSupplier, PrimaryObjectSupplier) line: 326	
	InjectorImpl.resolveArgs(Requestor, PrimaryObjectSupplier, PrimaryObjectSupplier, boolean, boolean, boolean) line: 450	
	InjectorImpl.resolveArguments(IRequestor, boolean) line: 334	
	FieldRequestor(Requestor).resolveArguments(boolean) line: 115	
	ContextObjectSupplier$ContextInjectionListener.update(IEclipseContext, int, Object[]) line: 76	
	TrackableComputationExt.update(ContextChangeEvent) line: 106	
	EclipseContext.processScheduled(Set<Scheduled>) line: 307	
	EclipseContext.set(String, Object) line: 321	
	E4Application.createE4Workbench(IApplicationContext, Display) line: 231	

(I often trap InjectionException as it seems the best way to discover why a handler isn't being invoked.)

Comment 1 Oleg Besedin

2011-11-15 14:53:55 EST

If I understand the description right, this seem like a result of the change in the bug 302533.

Is that correct? How can I duplicate the problem (and where is the CocoaUIProcessor coming from)?

Comment 2 Brian de Alwis

2011-11-17 08:03:19 EST

(In reply to comment #1)
> If I understand the description right, this seem like a result of the change in
> the bug 302533.

Maybe, although in this case there is a value available in the context.

> Is that correct? How can I duplicate the problem (and where is the
> CocoaUIProcessor coming from)?

The CocoaUIProcessor is part of the org.eclipse.e4.ui.workbench.renderers.swt.cocoa fragment, and installed by default.  (I vaguely recall that you had a mac, so this would be an easy demo of the issue.)

Comment 3 Oleg Besedin

2011-11-17 14:10:53 EST

Thank you, I can duplicate the problem. (My Git repo was messed up on Mac.)

There seems to be two issues there:
- Originally ModelAssembler#runProcessor() injects CocoaUIProcessor with a context that gets disposed (localContext.dispose());
- The CocoaUIProcessor gets update notification even through its context was disposed.

I'll look into those.

Comment 4 Oleg Besedin

2011-11-21 14:11:43 EST

(In reply to comment #3)
> There seems to be two issues there:
> - Originally ModelAssembler#runProcessor() injects CocoaUIProcessor with a
> context that gets disposed (localContext.dispose());

Fixed:
http://git.eclipse.org/c/platform/eclipse.platform.ui.git/commit/?id=0e5242977d5dd079aa45de99c7d85d1e41f5b60a

> - The CocoaUIProcessor gets update notification even through its context was
> disposed.

Fixed:
http://git.eclipse.org/c/platform/eclipse.platform.runtime.git/commit/?id=c8d5768a057413a47ccfd713aa89b2b3b01f25e4

Comment 5 Brian de Alwis

2012-01-24 13:31:15 EST

Verified in I20120123-2200 — it doesn't happen in any of my sample apps.