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

Bug 346805

Summary:

Object is not detached from it's resource on unload

Product:

[Modeling] EMF

Reporter:

Rimvydas <rimvydas>

Component:

Core

Assignee:

Ed Merks <Ed.Merks>

Status:

CLOSED FIXED

QA Contact:

Severity:

normal

Priority:

CC:

Andreas.Muelder, pierre-charles.david

Version:

2.6.0

Target Milestone:

---

Hardware:

OS:

Windows 7

Whiteboard:

Bug Depends on:

Bug Blocks:

390022

Attachments:

Description	Flags
unit test	none

Description Rimvydas

2011-05-22 04:10:51 EDT

Build Identifier: 20110218-0911

Object folderA is added into contents of resourceA and object folderB is added into contents of resourceB. folderA contains folderB. Unload of resourceB causes that folderB will be attached to resourceA but it was not detached from resourceB.

Expected: folderB should be:
1. detached from resourceB
2. attached to resourceA

Reproducible: Always

Steps to Reproduce:
1. Run attached unit test

Comment 1 Rimvydas

2011-05-22 04:12:02 EDT

Created attachment 196289 [details]
unit test

Comment 2 Ed Merks

2011-11-28 06:35:06 EST

The fix is committed to CVS for EMF 2.8.

Comment 3 Ed Merks

2012-01-09 04:00:16 EST

The fix is available in a build.

Comment 4 Andreas Muelder

2012-09-20 10:40:35 EDT

It seems that this bug fix causes regression to GMF
see Bug #390022

Comment 5 Ed Merks

2012-09-21 00:37:26 EDT

Yes, GMF is calling internal APIs inappropriately.  That method says

  /**
   * Sets this object to be directly contained by the resource 
   * and returns accumulated notifications.
   * This is only done as the inverse of {@link Resource#getContents()}<code>.add(this)</code>.
   * @return accumulated notifications.
   */
  NotificationChain eSetResource(Resource.Internal resource, NotificationChain notifications);

I.e., it's only called as part of bidirectional handshaking.  Calling it directly will leave things in an inconsistent state.  I.e., the other half of the bidirectional isn't updated.


In a newsgroup posting in July I suggested:

The GMF code should be using resource.getContents().remove(...).   I.e., perhaps it should be referencedObject.eResource().getContents().remove(referencedObject) though I'm not sure if the referenced object is necessarily contained directly in a resource so maybe it should check that eResource() isn't null. 

You should open a bugzilla for the GMF runtime to fix this.

Comment 6 Andreas Muelder

2012-09-21 04:39:50 EDT

Thanks for clarification, your suggestion works very well.
I provided a patch for the GMF bug.