[bug #3797] EOEditingContext, objectWillChange

[nobody]

=================== BUG #3797: FULL BUG SNAPSHOT ===================
http://savannah.gnu.org/bugs/?func=detailbug&bug_id=3797&group_id=99

Submitted by: dlatt                   Project: GNUstep                      
Submitted on: Thu 05/29/2003 at 09:58
Category:  gdl2                       Severity:  5 - Major                  
Bug Group:  None                      Resolution:  None                     
Assigned to:  ayers                   Status:  Open                         

Summary:  EOEditingContext, objectWillChange

Original Submission:  EOEditingContext should (according to the docs) add an 
invocation of -processRecentChanges to the end of the current run loop.

This may probably be done like in EODelayedObserverQueue.

Follow-up Comments
*******************

-------------------------------------------------------
Date: Wed 07/09/2003 at 14:57       By: ayers
I've attached two patches to the bug report, one for -base and one for -gdl2 
which currently "fix" the problem.  The patch for -base defines and uses 
NSUndoCloseGroupingRunLoopOrdering.  The patch for -gdl2 also uses the 
NSUndoCloseGroupingRunLoopOrdering in the EOUndoManager's implementation of 
registerUndoWithTarget:selector:object: after calling super, to register a call 
to _loop: with the run loop.  I would need someone with OS X to verify whether 
this actually belongs on NSUndoManager as WO4.5 suggests (and I still need to 
figure out whether this truely called unconditionally, when I find the time.)
The -gdl2 patch also 
- fixes the -hasChanges implementation of EOEditingContext to include checks 
for _unprocessedChanges/Deletes/Inserts,
- initializes the EOUndoManager instance for an EOEditingContext,
- sets the event grouping instead of querying it and ignoring the return value 
in _enqueueEndOfEventNotification,
- adds an empty implementation of _clearChangedThisTransaction: which is setup 
during processing and
- invokes _enqueueEndOfEventNotification in objectWillChange: the first time 
the object is changed and not just on subsequent changes.

PS: Sorry for the bandwidth.  I was hoping merely adding files to the report 
wouldn't trigger messages to bug-gnustep. I guess I was wrong :-( .


-------------------------------------------------------
Date: Wed 07/09/2003 at 08:17       By: ayers
Yesterday, Dirk and I did find out that EOEditingContext  -hasChanges should 
also check the _unprocessedChanges/Deletes/Inserts (patch coming up).

But the real issue, the update handling, is done through the NSUndoManager 
which currently isn't initialized for EOEditingContext.  Initializing it, 
doesn't quite do the trick yet though.  What seems to be missing in 
-[NSUndoManager registerUndoWithTarget:selector:object:] is a [[NSRunLoop 
currentRunLoop] performSelector:_loop target:self argument:nil 
order:NSUndoCloseGroupingRunLoopOrdering modes:NSDefaultRunLoopMode].  Yet I'm 
still unsure whether this is done uncondictionally.
Still investigating....

-------------------------------------------------------
Date: Tue 07/08/2003 at 11:07       By: dlatt
I am referring to the implementation of EOEditingContext's objectWillChange: 
method (didn't write it into the text because I thought it was sufficient to 
put it into the bug subject, sorry).
It is not _really_ explicitly written in the reference, but the mechanism is 
mentioned in several places, e.g. EOF Developer's Guide pp 209-211. Perhaps the 
most obvious description is in the EOF Control Reference, page 5, last 
paragraph of "Tracking Enterprise Objects Changes".
I interpret this in the way that in objectWillChange:, after doing its 
bookkeeping, the editing context should put a processRecentChanges invocation 
to itself at the end of the current run loop, so that the window can redraw all 
changes that occurred (this assumes using EOInterface).

I can't compare with Apple's EOF, so this is only an interpretation of the 
docs, but as it's implemented now in gdl2, changes are not displayed if I don't 
add an explicit call to processRecentChanges in the application. I don't 
believe this is intended.

The call to processRecentChanges in saveChanges is mentioned in the docs (see 
description of processRecentChanges in EO Control reference), but modifications 
in the object graph should be displayed in the user interface even if they are 
not yet saved (i.e., committed to the DB).

Wait... maybe the issue that the display doesn't get refreshed has something to 
do with hasChanges, which displays the correct value only after 
processRecentChanges (this seems wrong, too?!). I can't examine 
MulleEOInterface now, will investigate this evening.

-------------------------------------------------------
Date: Mon 07/07/2003 at 21:57       By: ayers
Sorry for the long delay in looking into this.  I tried to verify it a while 
back couldn't.  Now I had some time to dig deeper.

To which passages of the docs are you refering to?  I don't see any queing of a 
-processRecentChanges in any run loop (at least not through the standard 
methods: performSelector:target:argument:order:modes:, 
performSelector:withObject:afterDelay: or 
performSelector:withObject:afterDelay:inModes:)
during objectWillChange. (Testing with WO 4.5)

It is directly invoked from EOEditingContext-_processEndOfEventNotification: 
somewhere during EOEditingContext-saveChanges.  So, to me, our implementation 
seems correct.



CC list is empty


File Attachments
****************

-------------------------------------------------------
Date: Wed 07/09/2003 at 14:32  Name: base.patch  Size: 1KB   By: ayers
Temporary Patch for -base
http://savannah.gnu.org/bugs/download.php?group_id=99&bug_id=3797&bug_file_id=518


For detailed info, follow this link:
http://savannah.gnu.org/bugs/?func=detailbug&bug_id=3797&group_id=99

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/